generic-poky/bitbake/lib/bb/runqueue.py

2612 lines
105 KiB
Python
Raw Permalink Normal View History

#!/usr/bin/env python
# ex:ts=4:sw=4:sts=4:et
# -*- tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*-
"""
BitBake 'RunQueue' implementation
Handles preparation and execution of a queue of tasks
"""
# Copyright (C) 2006-2007 Richard Purdie
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License version 2 as
# published by the Free Software Foundation.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
import copy
import os
import sys
import signal
import stat
import fcntl
import errno
import logging
import re
import bb
from bb import msg, data, event
V5 Disk space monitoring Monitor disk availability and take action when the free disk space or amount of free inode is running low, it is enabled when BB_DISKMON_DIRS is set. * Variable meanings(from meta-yocto/conf/local.conf.sample): # Set the directories to monitor for disk usage, if more than one # directories are mounted in the same device, then only one directory # would be monitored since the monitor is based on the device. # The format is: # "action,directory,minimum_space,minimum_free_inode" # # The "action" must be set and should be one of: # ABORT: Immediately abort # STOPTASKS: The new tasks can't be executed any more, will stop the build # when the running tasks have been done. # WARN: show warnings (see BB_DISKMON_WARNINTERVAL for more information) # # The "directory" must be set, any directory is OK. # # Either "minimum_space" or "minimum_free_inode" (or both of them) # should be set, otherwise the monitor would not be enabled, # the unit can be G, M, K or none, but do NOT use GB, MB or KB # (B is not needed). #BB_DISKMON_DIRS = "STOPTASKS,${TMPDIR},1G,100K WARN,${SSTATE_DIR},1G,100K" # # Set disk space and inode interval (only works when the action is "WARN", # the unit can be G, M, or K, but do NOT use the GB, MB or KB # (B is not needed), the format is: # "disk_space_interval, disk_inode_interval", the default value is # "50M,5K" which means that it would warn when the free space is # lower than the minimum space(or inode), and would repeat the action # when the disk space reduces 50M (or the amount of inode reduces 5k) # again. #BB_DISKMON_WARNINTERVAL = "50M,5K" [YOCTO #1589] (Bitbake rev: 4d173d441d2beb8e6492b6b1842682f8cf32e6cc) Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2012-02-26 08:48:15 +00:00
from bb import monitordisk
import subprocess
import pickle
from multiprocessing import Process
bblogger = logging.getLogger("BitBake")
logger = logging.getLogger("BitBake.RunQueue")
__find_md5__ = re.compile( r'(?i)(?<![a-z0-9])[a-f0-9]{32}(?![a-z0-9])' )
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def fn_from_tid(tid):
return tid.rsplit(":", 1)[0]
def taskname_from_tid(tid):
return tid.rsplit(":", 1)[1]
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def split_tid(tid):
(mc, fn, taskname, _) = split_tid_mcfn(tid)
return (mc, fn, taskname)
def split_tid_mcfn(tid):
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if tid.startswith('multiconfig:'):
elems = tid.split(':')
mc = elems[1]
fn = ":".join(elems[2:-1])
taskname = elems[-1]
mcfn = "multiconfig:" + mc + ":" + fn
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
else:
tid = tid.rsplit(":", 1)
mc = ""
fn = tid[0]
taskname = tid[1]
mcfn = fn
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
return (mc, fn, taskname, mcfn)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def build_tid(mc, fn, taskname):
if mc:
return "multiconfig:" + mc + ":" + fn + ":" + taskname
return fn + ":" + taskname
class RunQueueStats:
"""
Holds statistics on the tasks handled by the associated runQueue
"""
def __init__(self, total):
self.completed = 0
self.skipped = 0
self.failed = 0
self.active = 0
self.total = total
def copy(self):
obj = self.__class__(self.total)
obj.__dict__.update(self.__dict__)
return obj
def taskFailed(self):
self.active = self.active - 1
self.failed = self.failed + 1
def taskCompleted(self, number = 1):
self.active = self.active - number
self.completed = self.completed + number
def taskSkipped(self, number = 1):
self.active = self.active + number
self.skipped = self.skipped + number
def taskActive(self):
self.active = self.active + 1
# These values indicate the next step due to be run in the
# runQueue state machine
runQueuePrepare = 2
runQueueSceneInit = 3
runQueueSceneRun = 4
runQueueRunInit = 5
runQueueRunning = 6
runQueueFailed = 7
runQueueCleanUp = 8
runQueueComplete = 9
class RunQueueScheduler(object):
"""
Control the order tasks are scheduled in.
"""
name = "basic"
def __init__(self, runqueue, rqdata):
"""
The default scheduler just returns the first buildable task (the
priority map is sorted by task number)
"""
self.rq = runqueue
self.rqdata = rqdata
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.numTasks = len(self.rqdata.runtaskentries)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.prio_map = [self.rqdata.runtaskentries.keys()]
self.buildable = []
self.stamps = {}
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
self.stamps[tid] = bb.build.stampfile(taskname, self.rqdata.dataCaches[mc], taskfn, noextra=True)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in self.rq.runq_buildable:
self.buildable.append(tid)
self.rev_prio_map = None
def next_buildable_task(self):
"""
Return the id of the first task we find that is buildable
"""
self.buildable = [x for x in self.buildable if x not in self.rq.runq_running]
if not self.buildable:
return None
if len(self.buildable) == 1:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
tid = self.buildable[0]
stamp = self.stamps[tid]
if stamp not in self.rq.build_stamps.values():
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
return tid
if not self.rev_prio_map:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.rev_prio_map = {}
for tid in self.rqdata.runtaskentries:
self.rev_prio_map[tid] = self.prio_map.index(tid)
best = None
bestprio = None
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.buildable:
prio = self.rev_prio_map[tid]
if bestprio is None or bestprio > prio:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
stamp = self.stamps[tid]
if stamp in self.rq.build_stamps.values():
continue
bestprio = prio
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
best = tid
return best
def next(self):
"""
Return the id of the task we should build next
"""
if self.rq.stats.active < self.rq.number_tasks:
return self.next_buildable_task()
def newbuilable(self, task):
self.buildable.append(task)
bitbake: runqueue.py: revised completion scheduler The idea is that tasks which complete building a recipe (like do_package_qa) are more important than tasks which start building new recipes (do_fetch) or those which increase disk usage (do_compile). Therefore tasks get ordered like this (most important first, do_rm_work before do_build because the enhanced rm_work.bbclass was used): 1. ID /work/poky/meta/recipes-support/popt/popt_1.16.bb:do_build 2. ID /work/poky/meta/recipes-core/readline/readline_6.3.bb:do_build 3. ID /work/poky/meta/recipes-connectivity/libnss-mdns/libnss-mdns_0.10.bb:do_build ... 464. ID /work/poky/meta/recipes-sato/images/core-image-sato.bb:do_build 465. ID /work/poky/meta/recipes-graphics/xorg-proto/inputproto_2.3.2.bb:do_rm_work 466. ID /work/poky/meta/recipes-devtools/python/python3_3.5.2.bb:do_rm_work 467. ID /work/poky/meta/recipes-core/packagegroups/packagegroup-base.bb:do_rm_work ... 3620. ID virtual:native:/work/poky/meta/recipes-extended/pbzip2/pbzip2_1.1.13.bb:do_install 3621. ID /work/poky/meta/recipes-devtools/qemu/qemu-helper-native_1.0.bb:do_install 3622. ID /work/poky/meta/recipes-core/zlib/zlib_1.2.8.bb:do_compile_ptest_base 3623. ID /work/poky/meta/recipes-extended/bzip2/bzip2_1.0.6.bb:do_compile_ptest_base ... 3645. ID /work/poky/meta/recipes-support/libevent/libevent_2.0.22.bb:do_compile_ptest_base 3646. ID /work/poky/meta/recipes-core/busybox/busybox_1.24.1.bb:do_compile_ptest_base 3647. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_uboot_mkimage 3648. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_sizecheck 3649. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_strip 3650. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_compile_kernelmodules 3651. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_shared_workdir 3652. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_kernel_link_images 3653. ID /work/poky/meta/recipes-devtools/quilt/quilt-native_0.64.bb:do_compile 3654. ID /work/poky/meta/recipes-extended/texinfo-dummy-native/texinfo-dummy-native.bb:do_compile ... The order of the same task between different recipes is the same as with the speed scheduler, i.e. more important recipes come first. (Bitbake rev: 70e297e5c285ce0a02e9efd3117ff62cdc77ec12) Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2017-01-19 19:35:16 +00:00
def describe_task(self, taskid):
result = 'ID %s' % taskid
if self.rev_prio_map:
result = result + (' pri %d' % self.rev_prio_map[taskid])
return result
def dump_prio(self, comment):
bb.debug(3, '%s (most important first):\n%s' %
(comment,
'\n'.join(['%d. %s' % (index + 1, self.describe_task(taskid)) for
index, taskid in enumerate(self.prio_map)])))
class RunQueueSchedulerSpeed(RunQueueScheduler):
"""
A scheduler optimised for speed. The priority map is sorted by task weight,
heavier weighted tasks (tasks needed by the most other tasks) are run first.
"""
name = "speed"
def __init__(self, runqueue, rqdata):
"""
The priority map is sorted by task weight.
"""
RunQueueScheduler.__init__(self, runqueue, rqdata)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
weights = {}
for tid in self.rqdata.runtaskentries:
weight = self.rqdata.runtaskentries[tid].weight
if not weight in weights:
weights[weight] = []
weights[weight].append(tid)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.prio_map = []
for weight in sorted(weights):
for w in weights[weight]:
self.prio_map.append(w)
self.prio_map.reverse()
class RunQueueSchedulerCompletion(RunQueueSchedulerSpeed):
"""
bitbake: runqueue.py: revised completion scheduler The idea is that tasks which complete building a recipe (like do_package_qa) are more important than tasks which start building new recipes (do_fetch) or those which increase disk usage (do_compile). Therefore tasks get ordered like this (most important first, do_rm_work before do_build because the enhanced rm_work.bbclass was used): 1. ID /work/poky/meta/recipes-support/popt/popt_1.16.bb:do_build 2. ID /work/poky/meta/recipes-core/readline/readline_6.3.bb:do_build 3. ID /work/poky/meta/recipes-connectivity/libnss-mdns/libnss-mdns_0.10.bb:do_build ... 464. ID /work/poky/meta/recipes-sato/images/core-image-sato.bb:do_build 465. ID /work/poky/meta/recipes-graphics/xorg-proto/inputproto_2.3.2.bb:do_rm_work 466. ID /work/poky/meta/recipes-devtools/python/python3_3.5.2.bb:do_rm_work 467. ID /work/poky/meta/recipes-core/packagegroups/packagegroup-base.bb:do_rm_work ... 3620. ID virtual:native:/work/poky/meta/recipes-extended/pbzip2/pbzip2_1.1.13.bb:do_install 3621. ID /work/poky/meta/recipes-devtools/qemu/qemu-helper-native_1.0.bb:do_install 3622. ID /work/poky/meta/recipes-core/zlib/zlib_1.2.8.bb:do_compile_ptest_base 3623. ID /work/poky/meta/recipes-extended/bzip2/bzip2_1.0.6.bb:do_compile_ptest_base ... 3645. ID /work/poky/meta/recipes-support/libevent/libevent_2.0.22.bb:do_compile_ptest_base 3646. ID /work/poky/meta/recipes-core/busybox/busybox_1.24.1.bb:do_compile_ptest_base 3647. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_uboot_mkimage 3648. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_sizecheck 3649. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_strip 3650. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_compile_kernelmodules 3651. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_shared_workdir 3652. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_kernel_link_images 3653. ID /work/poky/meta/recipes-devtools/quilt/quilt-native_0.64.bb:do_compile 3654. ID /work/poky/meta/recipes-extended/texinfo-dummy-native/texinfo-dummy-native.bb:do_compile ... The order of the same task between different recipes is the same as with the speed scheduler, i.e. more important recipes come first. (Bitbake rev: 70e297e5c285ce0a02e9efd3117ff62cdc77ec12) Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2017-01-19 19:35:16 +00:00
A scheduler optimised to complete .bb files as quickly as possible. The
priority map is sorted by task weight, but then reordered so once a given
bitbake: runqueue.py: revised completion scheduler The idea is that tasks which complete building a recipe (like do_package_qa) are more important than tasks which start building new recipes (do_fetch) or those which increase disk usage (do_compile). Therefore tasks get ordered like this (most important first, do_rm_work before do_build because the enhanced rm_work.bbclass was used): 1. ID /work/poky/meta/recipes-support/popt/popt_1.16.bb:do_build 2. ID /work/poky/meta/recipes-core/readline/readline_6.3.bb:do_build 3. ID /work/poky/meta/recipes-connectivity/libnss-mdns/libnss-mdns_0.10.bb:do_build ... 464. ID /work/poky/meta/recipes-sato/images/core-image-sato.bb:do_build 465. ID /work/poky/meta/recipes-graphics/xorg-proto/inputproto_2.3.2.bb:do_rm_work 466. ID /work/poky/meta/recipes-devtools/python/python3_3.5.2.bb:do_rm_work 467. ID /work/poky/meta/recipes-core/packagegroups/packagegroup-base.bb:do_rm_work ... 3620. ID virtual:native:/work/poky/meta/recipes-extended/pbzip2/pbzip2_1.1.13.bb:do_install 3621. ID /work/poky/meta/recipes-devtools/qemu/qemu-helper-native_1.0.bb:do_install 3622. ID /work/poky/meta/recipes-core/zlib/zlib_1.2.8.bb:do_compile_ptest_base 3623. ID /work/poky/meta/recipes-extended/bzip2/bzip2_1.0.6.bb:do_compile_ptest_base ... 3645. ID /work/poky/meta/recipes-support/libevent/libevent_2.0.22.bb:do_compile_ptest_base 3646. ID /work/poky/meta/recipes-core/busybox/busybox_1.24.1.bb:do_compile_ptest_base 3647. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_uboot_mkimage 3648. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_sizecheck 3649. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_strip 3650. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_compile_kernelmodules 3651. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_shared_workdir 3652. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_kernel_link_images 3653. ID /work/poky/meta/recipes-devtools/quilt/quilt-native_0.64.bb:do_compile 3654. ID /work/poky/meta/recipes-extended/texinfo-dummy-native/texinfo-dummy-native.bb:do_compile ... The order of the same task between different recipes is the same as with the speed scheduler, i.e. more important recipes come first. (Bitbake rev: 70e297e5c285ce0a02e9efd3117ff62cdc77ec12) Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2017-01-19 19:35:16 +00:00
.bb file starts to build, it's completed as quickly as possible by
running all tasks related to the same .bb file one after the after.
This works well where disk space is at a premium and classes like OE's
rm_work are in force.
"""
name = "completion"
def __init__(self, runqueue, rqdata):
bitbake: runqueue.py: revised completion scheduler The idea is that tasks which complete building a recipe (like do_package_qa) are more important than tasks which start building new recipes (do_fetch) or those which increase disk usage (do_compile). Therefore tasks get ordered like this (most important first, do_rm_work before do_build because the enhanced rm_work.bbclass was used): 1. ID /work/poky/meta/recipes-support/popt/popt_1.16.bb:do_build 2. ID /work/poky/meta/recipes-core/readline/readline_6.3.bb:do_build 3. ID /work/poky/meta/recipes-connectivity/libnss-mdns/libnss-mdns_0.10.bb:do_build ... 464. ID /work/poky/meta/recipes-sato/images/core-image-sato.bb:do_build 465. ID /work/poky/meta/recipes-graphics/xorg-proto/inputproto_2.3.2.bb:do_rm_work 466. ID /work/poky/meta/recipes-devtools/python/python3_3.5.2.bb:do_rm_work 467. ID /work/poky/meta/recipes-core/packagegroups/packagegroup-base.bb:do_rm_work ... 3620. ID virtual:native:/work/poky/meta/recipes-extended/pbzip2/pbzip2_1.1.13.bb:do_install 3621. ID /work/poky/meta/recipes-devtools/qemu/qemu-helper-native_1.0.bb:do_install 3622. ID /work/poky/meta/recipes-core/zlib/zlib_1.2.8.bb:do_compile_ptest_base 3623. ID /work/poky/meta/recipes-extended/bzip2/bzip2_1.0.6.bb:do_compile_ptest_base ... 3645. ID /work/poky/meta/recipes-support/libevent/libevent_2.0.22.bb:do_compile_ptest_base 3646. ID /work/poky/meta/recipes-core/busybox/busybox_1.24.1.bb:do_compile_ptest_base 3647. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_uboot_mkimage 3648. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_sizecheck 3649. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_strip 3650. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_compile_kernelmodules 3651. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_shared_workdir 3652. ID /work/poky/meta/recipes-kernel/linux/linux-yocto_4.8.bb:do_kernel_link_images 3653. ID /work/poky/meta/recipes-devtools/quilt/quilt-native_0.64.bb:do_compile 3654. ID /work/poky/meta/recipes-extended/texinfo-dummy-native/texinfo-dummy-native.bb:do_compile ... The order of the same task between different recipes is the same as with the speed scheduler, i.e. more important recipes come first. (Bitbake rev: 70e297e5c285ce0a02e9efd3117ff62cdc77ec12) Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2017-01-19 19:35:16 +00:00
super(RunQueueSchedulerCompletion, self).__init__(runqueue, rqdata)
# Extract list of tasks for each recipe, with tasks sorted
# ascending from "must run first" (typically do_fetch) to
# "runs last" (do_build). The speed scheduler prioritizes
# tasks that must run first before the ones that run later;
# this is what we depend on here.
task_lists = {}
for taskid in self.prio_map:
fn, taskname = taskid.rsplit(':', 1)
task_lists.setdefault(fn, []).append(taskname)
# Now unify the different task lists. The strategy is that
# common tasks get skipped and new ones get inserted after the
# preceeding common one(s) as they are found. Because task
# lists should differ only by their number of tasks, but not
# the ordering of the common tasks, this should result in a
# deterministic result that is a superset of the individual
# task ordering.
all_tasks = []
for recipe, new_tasks in task_lists.items():
index = 0
old_task = all_tasks[index] if index < len(all_tasks) else None
for new_task in new_tasks:
if old_task == new_task:
# Common task, skip it. This is the fast-path which
# avoids a full search.
index += 1
old_task = all_tasks[index] if index < len(all_tasks) else None
else:
try:
index = all_tasks.index(new_task)
# Already present, just not at the current
# place. We re-synchronized by changing the
# index so that it matches again. Now
# move on to the next existing task.
index += 1
old_task = all_tasks[index] if index < len(all_tasks) else None
except ValueError:
# Not present. Insert before old_task, which
# remains the same (but gets shifted back).
all_tasks.insert(index, new_task)
index += 1
bb.debug(3, 'merged task list: %s' % all_tasks)
# Now reverse the order so that tasks that finish the work on one
# recipe are considered more imporant (= come first). The ordering
# is now so that do_build is most important.
all_tasks.reverse()
# Group tasks of the same kind before tasks of less important
# kinds at the head of the queue (because earlier = lower
# priority number = runs earlier), while preserving the
# ordering by recipe. If recipe foo is more important than
# bar, then the goal is to work on foo's do_populate_sysroot
# before bar's do_populate_sysroot and on the more important
# tasks of foo before any of the less important tasks in any
# other recipe (if those other recipes are more important than
# foo).
#
# All of this only applies when tasks are runable. Explicit
# dependencies still override this ordering by priority.
#
# Here's an example why this priority re-ordering helps with
# minimizing disk usage. Consider a recipe foo with a higher
# priority than bar where foo DEPENDS on bar. Then the
# implicit rule (from base.bbclass) is that foo's do_configure
# depends on bar's do_populate_sysroot. This ensures that
# bar's do_populate_sysroot gets done first. Normally the
# tasks from foo would continue to run once that is done, and
# bar only gets completed and cleaned up later. By ordering
# bar's task that depend on bar's do_populate_sysroot before foo's
# do_configure, that problem gets avoided.
task_index = 0
self.dump_prio('original priorities')
for task in all_tasks:
for index in range(task_index, self.numTasks):
taskid = self.prio_map[index]
taskname = taskid.rsplit(':', 1)[1]
if taskname == task:
del self.prio_map[index]
self.prio_map.insert(task_index, taskid)
task_index += 1
self.dump_prio('completion priorities')
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
class RunTaskEntry(object):
def __init__(self):
self.depends = set()
self.revdeps = set()
self.hash = None
self.task = None
self.weight = 1
class RunQueueData:
"""
BitBake Run Queue implementation
"""
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def __init__(self, rq, cooker, cfgData, dataCaches, taskData, targets):
self.cooker = cooker
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.dataCaches = dataCaches
self.taskData = taskData
self.targets = targets
self.rq = rq
self.warn_multi_bb = False
self.stampwhitelist = cfgData.getVar("BB_STAMP_WHITELIST") or ""
self.multi_provider_whitelist = (cfgData.getVar("MULTI_PROVIDER_WHITELIST") or "").split()
self.setscenewhitelist = get_setscene_enforce_whitelist(cfgData)
self.setscenewhitelist_checked = False
self.setscene_enforce = (cfgData.getVar('BB_SETSCENE_ENFORCE') == "1")
self.init_progress_reporter = bb.progress.DummyMultiStageProcessProgressReporter()
self.reset()
def reset(self):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.runtaskentries = {}
def runq_depends_names(self, ids):
import re
ret = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for id in ids:
nam = os.path.basename(id)
nam = re.sub("_[^,]*,", ",", nam)
ret.extend([nam])
return ret
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def get_task_hash(self, tid):
return self.runtaskentries[tid].hash
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def get_user_idstring(self, tid, task_name_suffix = ""):
return tid + task_name_suffix
def get_short_user_idstring(self, task, task_name_suffix = ""):
(mc, fn, taskname, taskfn) = split_tid_mcfn(task)
pn = self.dataCaches[mc].pkg_fn[taskfn]
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
taskname = taskname_from_tid(task) + task_name_suffix
return "%s:%s" % (pn, taskname)
def circular_depchains_handler(self, tasks):
"""
Some tasks aren't buildable, likely due to circular dependency issues.
Identify the circular dependencies and print them in a user readable format.
"""
from copy import deepcopy
valid_chains = []
explored_deps = {}
msgs = []
def chain_reorder(chain):
"""
Reorder a dependency chain so the lowest task id is first
"""
lowest = 0
new_chain = []
for entry in range(len(chain)):
if chain[entry] < chain[lowest]:
lowest = entry
new_chain.extend(chain[lowest:])
new_chain.extend(chain[:lowest])
return new_chain
def chain_compare_equal(chain1, chain2):
"""
Compare two dependency chains and see if they're the same
"""
if len(chain1) != len(chain2):
return False
for index in range(len(chain1)):
if chain1[index] != chain2[index]:
return False
return True
def chain_array_contains(chain, chain_array):
"""
Return True if chain_array contains chain
"""
for ch in chain_array:
if chain_compare_equal(ch, chain):
return True
return False
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def find_chains(tid, prev_chain):
prev_chain.append(tid)
total_deps = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
total_deps.extend(self.runtaskentries[tid].revdeps)
for revdep in self.runtaskentries[tid].revdeps:
if revdep in prev_chain:
idx = prev_chain.index(revdep)
# To prevent duplicates, reorder the chain to start with the lowest taskid
# and search through an array of those we've already printed
chain = prev_chain[idx:]
new_chain = chain_reorder(chain)
if not chain_array_contains(new_chain, valid_chains):
valid_chains.append(new_chain)
msgs.append("Dependency loop #%d found:\n" % len(valid_chains))
for dep in new_chain:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
msgs.append(" Task %s (dependent Tasks %s)\n" % (dep, self.runq_depends_names(self.runtaskentries[dep].depends)))
msgs.append("\n")
if len(valid_chains) > 10:
msgs.append("Aborted dependency loops search after 10 matches.\n")
return msgs
continue
scan = False
if revdep not in explored_deps:
scan = True
elif revdep in explored_deps[revdep]:
scan = True
else:
for dep in prev_chain:
if dep in explored_deps[revdep]:
scan = True
if scan:
find_chains(revdep, copy.deepcopy(prev_chain))
for dep in explored_deps[revdep]:
if dep not in total_deps:
total_deps.append(dep)
explored_deps[tid] = total_deps
for task in tasks:
find_chains(task, [])
return msgs
def calculate_task_weights(self, endpoints):
"""
Calculate a number representing the "weight" of each task. Heavier weighted tasks
have more dependencies and hence should be executed sooner for maximum speed.
This function also sanity checks the task list finding tasks that are not
possible to execute due to circular dependencies.
"""
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
numTasks = len(self.runtaskentries)
weight = {}
deps_left = {}
task_done = {}
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
task_done[tid] = False
weight[tid] = 1
deps_left[tid] = len(self.runtaskentries[tid].revdeps)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in endpoints:
weight[tid] = 10
task_done[tid] = True
while True:
next_points = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in endpoints:
for revdep in self.runtaskentries[tid].depends:
weight[revdep] = weight[revdep] + weight[tid]
deps_left[revdep] = deps_left[revdep] - 1
if deps_left[revdep] == 0:
next_points.append(revdep)
task_done[revdep] = True
endpoints = next_points
if len(next_points) == 0:
break
# Circular dependency sanity check
problem_tasks = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
if task_done[tid] is False or deps_left[tid] != 0:
problem_tasks.append(tid)
logger.debug(2, "Task %s is not buildable", tid)
logger.debug(2, "(Complete marker was %s and the remaining dependency count was %s)\n", task_done[tid], deps_left[tid])
self.runtaskentries[tid].weight = weight[tid]
if problem_tasks:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
message = "%s unbuildable tasks were found.\n" % len(problem_tasks)
message = message + "These are usually caused by circular dependencies and any circular dependency chains found will be printed below. Increase the debug level to see a list of unbuildable tasks.\n\n"
message = message + "Identifying dependency loops (this may take a short while)...\n"
logger.error(message)
msgs = self.circular_depchains_handler(problem_tasks)
message = "\n"
for msg in msgs:
message = message + msg
bb.msg.fatal("RunQueue", message)
return weight
def prepare(self):
"""
Turn a set of taskData into a RunQueue and compute data needed
to optimise the execution order.
"""
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
runq_build = {}
recursivetasks = {}
recursiveitasks = {}
recursivetasksselfref = set()
taskData = self.taskData
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
found = False
for mc in self.taskData:
if len(taskData[mc].taskentries) > 0:
found = True
break
if not found:
# Nothing to do
return 0
self.init_progress_reporter.start()
self.init_progress_reporter.next_stage()
# Step A - Work out a list of tasks to run
#
# Taskdata gives us a list of possible providers for every build and run
# target ordered by priority. It also gives information on each of those
# providers.
#
# To create the actual list of tasks to execute we fix the list of
# providers and then resolve the dependencies into task IDs. This
# process is repeated for each type of dependency (tdepends, deptask,
# rdeptast, recrdeptask, idepends).
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def add_build_dependencies(depids, tasknames, depends, mc):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for depname in depids:
# Won't be in build_targets if ASSUME_PROVIDED
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if depname not in taskData[mc].build_targets or not taskData[mc].build_targets[depname]:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depdata = taskData[mc].build_targets[depname][0]
if depdata is None:
continue
for taskname in tasknames:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
t = depdata + ":" + taskname
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if t in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
depends.add(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def add_runtime_dependencies(depids, tasknames, depends, mc):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for depname in depids:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if depname not in taskData[mc].run_targets or not taskData[mc].run_targets[depname]:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depdata = taskData[mc].run_targets[depname][0]
if depdata is None:
continue
for taskname in tasknames:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
t = depdata + ":" + taskname
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if t in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
depends.add(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def add_resolved_dependencies(mc, fn, tasknames, depends):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for taskname in tasknames:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
tid = build_tid(mc, fn, taskname)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in self.runtaskentries:
depends.add(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for mc in taskData:
for tid in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
#runtid = build_tid(mc, fn, taskname)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
#logger.debug(2, "Processing %s,%s:%s", mc, fn, taskname)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depends = set()
task_deps = self.dataCaches[mc].task_deps[taskfn]
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.runtaskentries[tid] = RunTaskEntry()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if fn in taskData[mc].failed_fns:
continue
# Resolve task internal dependencies
#
# e.g. addtask before X after Y
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for t in taskData[mc].taskentries[tid].tdepends:
(_, depfn, deptaskname, _) = split_tid_mcfn(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depends.add(build_tid(mc, depfn, deptaskname))
# Resolve 'deptask' dependencies
#
# e.g. do_sometask[deptask] = "do_someothertask"
# (makes sure sometask runs after someothertask of all DEPENDS)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if 'deptask' in task_deps and taskname in task_deps['deptask']:
tasknames = task_deps['deptask'][taskname].split()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
add_build_dependencies(taskData[mc].depids[taskfn], tasknames, depends, mc)
# Resolve 'rdeptask' dependencies
#
# e.g. do_sometask[rdeptask] = "do_someothertask"
# (makes sure sometask runs after someothertask of all RDEPENDS)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if 'rdeptask' in task_deps and taskname in task_deps['rdeptask']:
tasknames = task_deps['rdeptask'][taskname].split()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
add_runtime_dependencies(taskData[mc].rdepids[taskfn], tasknames, depends, mc)
# Resolve inter-task dependencies
#
# e.g. do_sometask[depends] = "targetname:do_someothertask"
# (makes sure sometask runs after targetname's someothertask)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
idepends = taskData[mc].taskentries[tid].idepends
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for (depname, idependtask) in idepends:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if depname in taskData[mc].build_targets and taskData[mc].build_targets[depname] and not depname in taskData[mc].failed_deps:
# Won't be in build_targets if ASSUME_PROVIDED
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depdata = taskData[mc].build_targets[depname][0]
if depdata is not None:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
t = depdata + ":" + idependtask
depends.add(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if t not in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
bb.msg.fatal("RunQueue", "Task %s in %s depends upon non-existent task %s in %s" % (taskname, fn, idependtask, depdata))
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
irdepends = taskData[mc].taskentries[tid].irdepends
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for (depname, idependtask) in irdepends:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if depname in taskData[mc].run_targets:
# Won't be in run_targets if ASSUME_PROVIDED
if not taskData[mc].run_targets[depname]:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depdata = taskData[mc].run_targets[depname][0]
if depdata is not None:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
t = depdata + ":" + idependtask
depends.add(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if t not in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
bb.msg.fatal("RunQueue", "Task %s in %s rdepends upon non-existent task %s in %s" % (taskname, fn, idependtask, depdata))
# Resolve recursive 'recrdeptask' dependencies (Part A)
#
# e.g. do_sometask[recrdeptask] = "do_someothertask"
# (makes sure sometask runs after someothertask of all DEPENDS, RDEPENDS and intertask dependencies, recursively)
# We cover the recursive part of the dependencies below
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if 'recrdeptask' in task_deps and taskname in task_deps['recrdeptask']:
tasknames = task_deps['recrdeptask'][taskname].split()
recursivetasks[tid] = tasknames
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
add_build_dependencies(taskData[mc].depids[taskfn], tasknames, depends, mc)
add_runtime_dependencies(taskData[mc].rdepids[taskfn], tasknames, depends, mc)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if taskname in tasknames:
recursivetasksselfref.add(tid)
if 'recideptask' in task_deps and taskname in task_deps['recideptask']:
recursiveitasks[tid] = []
for t in task_deps['recideptask'][taskname].split():
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
newdep = build_tid(mc, fn, t)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
recursiveitasks[tid].append(newdep)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.runtaskentries[tid].depends = depends
#self.dump_data()
# Resolve recursive 'recrdeptask' dependencies (Part B)
#
# e.g. do_sometask[recrdeptask] = "do_someothertask"
# (makes sure sometask runs after someothertask of all DEPENDS, RDEPENDS and intertask dependencies, recursively)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
# We need to do this separately since we need all of runtaskentries[*].depends to be complete before this is processed
self.init_progress_reporter.next_stage(len(recursivetasks))
extradeps = {}
for taskcounter, tid in enumerate(recursivetasks):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
extradeps[tid] = set(self.runtaskentries[tid].depends)
tasknames = recursivetasks[tid]
seendeps = set()
def generate_recdeps(t):
newdeps = set()
(mc, fn, taskname, _) = split_tid_mcfn(t)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
add_resolved_dependencies(mc, fn, tasknames, newdeps)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
extradeps[tid].update(newdeps)
seendeps.add(t)
newdeps.add(t)
for i in newdeps:
if i not in self.runtaskentries:
# Not all recipes might have the recrdeptask task as a task
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
task = self.runtaskentries[i].task
for n in self.runtaskentries[i].depends:
if n not in seendeps:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
generate_recdeps(n)
generate_recdeps(tid)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in recursiveitasks:
for dep in recursiveitasks[tid]:
generate_recdeps(dep)
self.init_progress_reporter.update(taskcounter)
# Remove circular references so that do_a[recrdeptask] = "do_a do_b" can work
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in recursivetasks:
extradeps[tid].difference_update(recursivetasksselfref)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
task = self.runtaskentries[tid].task
# Add in extra dependencies
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in extradeps:
self.runtaskentries[tid].depends = extradeps[tid]
# Remove all self references
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in self.runtaskentries[tid].depends:
logger.debug(2, "Task %s contains self reference!", tid)
self.runtaskentries[tid].depends.remove(tid)
self.init_progress_reporter.next_stage()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
#self.dump_data()
# Step B - Mark all active tasks
#
# Start with the tasks we were asked to run and mark all dependencies
# as active too. If the task is to be 'forced', clear its stamp. Once
# all active tasks are marked, prune the ones we don't need.
logger.verbose("Marking Active Tasks")
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def mark_active(tid, depth):
"""
Mark an item as active along with its depends
(calls itself recursively)
"""
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in runq_build:
return
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
runq_build[tid] = 1
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
depends = self.runtaskentries[tid].depends
for depend in depends:
mark_active(depend, depth+1)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.target_tids = []
for (mc, target, task, fn) in self.targets:
if target not in taskData[mc].build_targets or not taskData[mc].build_targets[target]:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if target in taskData[mc].failed_deps:
continue
parents = False
if task.endswith('-'):
parents = True
task = task[:-1]
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if fn in taskData[mc].failed_fns:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
# fn already has mc prefix
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
tid = fn + ":" + task
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.target_tids.append(tid)
if tid not in taskData[mc].taskentries:
import difflib
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
tasks = []
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for x in taskData[mc].taskentries:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if x.startswith(fn + ":"):
tasks.append(taskname_from_tid(x))
close_matches = difflib.get_close_matches(task, tasks, cutoff=0.7)
if close_matches:
extra = ". Close matches:\n %s" % "\n ".join(close_matches)
else:
extra = ""
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
bb.msg.fatal("RunQueue", "Task %s does not exist for target %s (%s)%s" % (task, target, tid, extra))
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
# For tasks called "XXXX-", ony run their dependencies
if parents:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for i in self.runtaskentries[tid].depends:
mark_active(i, 1)
else:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
mark_active(tid, 1)
self.init_progress_reporter.next_stage()
# Step C - Prune all inactive tasks
#
# Once all active tasks are marked, prune the ones we don't need.
delcount = 0
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in list(self.runtaskentries.keys()):
if tid not in runq_build:
del self.runtaskentries[tid]
delcount += 1
self.init_progress_reporter.next_stage()
if self.cooker.configuration.runall is not None:
runall = "do_%s" % self.cooker.configuration.runall
runall_tids = { k: v for k, v in self.runtaskentries.items() if taskname_from_tid(k) == runall }
# re-run the mark_active and then drop unused tasks from new list
runq_build = {}
for tid in list(runall_tids):
mark_active(tid,1)
for tid in list(self.runtaskentries.keys()):
if tid not in runq_build:
del self.runtaskentries[tid]
delcount += 1
if len(self.runtaskentries) == 0:
bb.msg.fatal("RunQueue", "No remaining tasks to run for build target %s with runall %s" % (target, runall))
#
# Step D - Sanity checks and computation
#
# Check to make sure we still have tasks to run
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if len(self.runtaskentries) == 0:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if not taskData[''].abort:
bb.msg.fatal("RunQueue", "All buildable tasks have been run but the build is incomplete (--continue mode). Errors for the tasks that failed will have been printed above.")
else:
bb.msg.fatal("RunQueue", "No active tasks and not in --continue mode?! Please report this bug.")
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.verbose("Pruned %s inactive tasks, %s left", delcount, len(self.runtaskentries))
logger.verbose("Assign Weightings")
self.init_progress_reporter.next_stage()
# Generate a list of reverse dependencies to ease future calculations
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
for dep in self.runtaskentries[tid].depends:
self.runtaskentries[dep].revdeps.add(tid)
self.init_progress_reporter.next_stage()
# Identify tasks at the end of dependency chains
# Error on circular dependency loops (length two)
endpoints = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
revdeps = self.runtaskentries[tid].revdeps
if len(revdeps) == 0:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
endpoints.append(tid)
for dep in revdeps:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if dep in self.runtaskentries[tid].depends:
bb.msg.fatal("RunQueue", "Task %s has circular dependency on %s" % (tid, dep))
logger.verbose("Compute totals (have %s endpoint(s))", len(endpoints))
self.init_progress_reporter.next_stage()
# Calculate task weights
# Check of higher length circular dependencies
self.runq_weight = self.calculate_task_weights(endpoints)
self.init_progress_reporter.next_stage()
# Sanity Check - Check for multiple tasks building the same provider
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for mc in self.dataCaches:
prov_list = {}
seen_fn = []
for tid in self.runtaskentries:
(tidmc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if taskfn in seen_fn:
continue
if mc != tidmc:
continue
seen_fn.append(taskfn)
for prov in self.dataCaches[mc].fn_provides[taskfn]:
if prov not in prov_list:
prov_list[prov] = [taskfn]
elif taskfn not in prov_list[prov]:
prov_list[prov].append(taskfn)
for prov in prov_list:
if len(prov_list[prov]) < 2:
continue
if prov in self.multi_provider_whitelist:
continue
seen_pn = []
# If two versions of the same PN are being built its fatal, we don't support it.
for fn in prov_list[prov]:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
pn = self.dataCaches[mc].pkg_fn[fn]
if pn not in seen_pn:
seen_pn.append(pn)
else:
bb.fatal("Multiple versions of %s are due to be built (%s). Only one version of a given PN should be built in any given build. You likely need to set PREFERRED_VERSION_%s to select the correct version or don't depend on multiple versions." % (pn, " ".join(prov_list[prov]), pn))
msg = "Multiple .bb files are due to be built which each provide %s:\n %s" % (prov, "\n ".join(prov_list[prov]))
#
# Construct a list of things which uniquely depend on each provider
# since this may help the user figure out which dependency is triggering this warning
#
msg += "\nA list of tasks depending on these providers is shown and may help explain where the dependency comes from."
deplist = {}
commondeps = None
for provfn in prov_list[prov]:
deps = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
fn = fn_from_tid(tid)
if fn != provfn:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.runtaskentries[tid].revdeps:
fn = fn_from_tid(dep)
if fn == provfn:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
deps.add(dep)
if not commondeps:
commondeps = set(deps)
else:
commondeps &= deps
deplist[provfn] = deps
for provfn in deplist:
msg += "\n%s has unique dependees:\n %s" % (provfn, "\n ".join(deplist[provfn] - commondeps))
#
# Construct a list of provides and runtime providers for each recipe
# (rprovides has to cover RPROVIDES, PACKAGES, PACKAGES_DYNAMIC)
#
msg += "\nIt could be that one recipe provides something the other doesn't and should. The following provider and runtime provider differences may be helpful."
provide_results = {}
rprovide_results = {}
commonprovs = None
commonrprovs = None
for provfn in prov_list[prov]:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
provides = set(self.dataCaches[mc].fn_provides[provfn])
rprovides = set()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for rprovide in self.dataCaches[mc].rproviders:
if provfn in self.dataCaches[mc].rproviders[rprovide]:
rprovides.add(rprovide)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for package in self.dataCaches[mc].packages:
if provfn in self.dataCaches[mc].packages[package]:
rprovides.add(package)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for package in self.dataCaches[mc].packages_dynamic:
if provfn in self.dataCaches[mc].packages_dynamic[package]:
rprovides.add(package)
if not commonprovs:
commonprovs = set(provides)
else:
commonprovs &= provides
provide_results[provfn] = provides
if not commonrprovs:
commonrprovs = set(rprovides)
else:
commonrprovs &= rprovides
rprovide_results[provfn] = rprovides
#msg += "\nCommon provides:\n %s" % ("\n ".join(commonprovs))
#msg += "\nCommon rprovides:\n %s" % ("\n ".join(commonrprovs))
for provfn in prov_list[prov]:
msg += "\n%s has unique provides:\n %s" % (provfn, "\n ".join(provide_results[provfn] - commonprovs))
msg += "\n%s has unique rprovides:\n %s" % (provfn, "\n ".join(rprovide_results[provfn] - commonrprovs))
if self.warn_multi_bb:
logger.warning(msg)
else:
logger.error(msg)
self.init_progress_reporter.next_stage()
# Create a whitelist usable by the stamp checks
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.stampfnwhitelist = {}
for mc in self.taskData:
self.stampfnwhitelist[mc] = []
for entry in self.stampwhitelist.split():
if entry not in self.taskData[mc].build_targets:
continue
fn = self.taskData.build_targets[entry][0]
self.stampfnwhitelist[mc].append(fn)
self.init_progress_reporter.next_stage()
# Iterate over the task list looking for tasks with a 'setscene' function
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.runq_setscene_tids = []
if not self.cooker.configuration.nosetscene:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
(mc, fn, taskname, _) = split_tid_mcfn(tid)
setscenetid = tid + "_setscene"
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if setscenetid not in taskData[mc].taskentries:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.runq_setscene_tids.append(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def invalidate_task(tid, error_nostamp):
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
taskdep = self.dataCaches[mc].task_deps[taskfn]
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if fn + ":" + taskname not in taskData[mc].taskentries:
logger.warning("Task %s does not exist, invalidating this task will have no effect" % taskname)
if 'nostamp' in taskdep and taskname in taskdep['nostamp']:
if error_nostamp:
bb.fatal("Task %s is marked nostamp, cannot invalidate this task" % taskname)
else:
bb.debug(1, "Task %s is marked nostamp, cannot invalidate this task" % taskname)
else:
logger.verbose("Invalidate task %s, %s", taskname, fn)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
bb.parse.siggen.invalidate_task(taskname, self.dataCaches[mc], fn)
self.init_progress_reporter.next_stage()
2012-06-18 15:45:35 +00:00
# Invalidate task if force mode active
if self.cooker.configuration.force:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for tid in self.target_tids:
invalidate_task(tid, False)
# Invalidate task if invalidate mode active
if self.cooker.configuration.invalidate_stamp:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for tid in self.target_tids:
fn = fn_from_tid(tid)
for st in self.cooker.configuration.invalidate_stamp.split(','):
if not st.startswith("do_"):
st = "do_%s" % st
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
invalidate_task(fn + ":" + st, True)
2012-06-18 15:45:35 +00:00
self.init_progress_reporter.next_stage()
# Create and print to the logs a virtual/xxxx -> PN (fn) table
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for mc in taskData:
virtmap = taskData[mc].get_providermap(prefix="virtual/")
virtpnmap = {}
for v in virtmap:
virtpnmap[v] = self.dataCaches[mc].pkg_fn[virtmap[v]]
bb.debug(2, "%s resolved to: %s (%s)" % (v, virtpnmap[v], virtmap[v]))
if hasattr(bb.parse.siggen, "tasks_resolved"):
bb.parse.siggen.tasks_resolved(virtmap, virtpnmap, self.dataCaches[mc])
self.init_progress_reporter.next_stage()
# Iterate over the task list and call into the siggen code
dealtwith = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
todeal = set(self.runtaskentries)
while len(todeal) > 0:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in todeal.copy():
if len(self.runtaskentries[tid].depends - dealtwith) == 0:
dealtwith.add(tid)
todeal.remove(tid)
procdep = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.runtaskentries[tid].depends:
procdep.append(fn_from_tid(dep) + "." + taskname_from_tid(dep))
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.runtaskentries[tid].hash = bb.parse.siggen.get_taskhash(taskfn, taskname, procdep, self.dataCaches[mc])
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
task = self.runtaskentries[tid].task
bb.parse.siggen.writeout_file_checksum_cache()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
#self.dump_data()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
return len(self.runtaskentries)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def dump_data(self):
"""
Dump some debug information on the internal data structures
"""
logger.debug(3, "run_tasks:")
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.runtaskentries:
logger.debug(3, " %s: %s Deps %s RevDeps %s", tid,
self.runtaskentries[tid].weight,
self.runtaskentries[tid].depends,
self.runtaskentries[tid].revdeps)
class RunQueueWorker():
def __init__(self, process, pipe):
self.process = process
self.pipe = pipe
class RunQueue:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def __init__(self, cooker, cfgData, dataCaches, taskData, targets):
self.cooker = cooker
self.cfgData = cfgData
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rqdata = RunQueueData(self, cooker, cfgData, dataCaches, taskData, targets)
self.stamppolicy = cfgData.getVar("BB_STAMP_POLICY") or "perfile"
self.hashvalidate = cfgData.getVar("BB_HASHCHECK_FUNCTION") or None
self.setsceneverify = cfgData.getVar("BB_SETSCENE_VERIFY_FUNCTION2") or None
self.depvalidate = cfgData.getVar("BB_SETSCENE_DEPVALID") or None
self.state = runQueuePrepare
V5 Disk space monitoring Monitor disk availability and take action when the free disk space or amount of free inode is running low, it is enabled when BB_DISKMON_DIRS is set. * Variable meanings(from meta-yocto/conf/local.conf.sample): # Set the directories to monitor for disk usage, if more than one # directories are mounted in the same device, then only one directory # would be monitored since the monitor is based on the device. # The format is: # "action,directory,minimum_space,minimum_free_inode" # # The "action" must be set and should be one of: # ABORT: Immediately abort # STOPTASKS: The new tasks can't be executed any more, will stop the build # when the running tasks have been done. # WARN: show warnings (see BB_DISKMON_WARNINTERVAL for more information) # # The "directory" must be set, any directory is OK. # # Either "minimum_space" or "minimum_free_inode" (or both of them) # should be set, otherwise the monitor would not be enabled, # the unit can be G, M, K or none, but do NOT use GB, MB or KB # (B is not needed). #BB_DISKMON_DIRS = "STOPTASKS,${TMPDIR},1G,100K WARN,${SSTATE_DIR},1G,100K" # # Set disk space and inode interval (only works when the action is "WARN", # the unit can be G, M, or K, but do NOT use the GB, MB or KB # (B is not needed), the format is: # "disk_space_interval, disk_inode_interval", the default value is # "50M,5K" which means that it would warn when the free space is # lower than the minimum space(or inode), and would repeat the action # when the disk space reduces 50M (or the amount of inode reduces 5k) # again. #BB_DISKMON_WARNINTERVAL = "50M,5K" [YOCTO #1589] (Bitbake rev: 4d173d441d2beb8e6492b6b1842682f8cf32e6cc) Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2012-02-26 08:48:15 +00:00
# For disk space monitor
# Invoked at regular time intervals via the bitbake heartbeat event
# while the build is running. We generate a unique name for the handler
# here, just in case that there ever is more than one RunQueue instance,
# start the handler when reaching runQueueSceneRun, and stop it when
# done with the build.
V5 Disk space monitoring Monitor disk availability and take action when the free disk space or amount of free inode is running low, it is enabled when BB_DISKMON_DIRS is set. * Variable meanings(from meta-yocto/conf/local.conf.sample): # Set the directories to monitor for disk usage, if more than one # directories are mounted in the same device, then only one directory # would be monitored since the monitor is based on the device. # The format is: # "action,directory,minimum_space,minimum_free_inode" # # The "action" must be set and should be one of: # ABORT: Immediately abort # STOPTASKS: The new tasks can't be executed any more, will stop the build # when the running tasks have been done. # WARN: show warnings (see BB_DISKMON_WARNINTERVAL for more information) # # The "directory" must be set, any directory is OK. # # Either "minimum_space" or "minimum_free_inode" (or both of them) # should be set, otherwise the monitor would not be enabled, # the unit can be G, M, K or none, but do NOT use GB, MB or KB # (B is not needed). #BB_DISKMON_DIRS = "STOPTASKS,${TMPDIR},1G,100K WARN,${SSTATE_DIR},1G,100K" # # Set disk space and inode interval (only works when the action is "WARN", # the unit can be G, M, or K, but do NOT use the GB, MB or KB # (B is not needed), the format is: # "disk_space_interval, disk_inode_interval", the default value is # "50M,5K" which means that it would warn when the free space is # lower than the minimum space(or inode), and would repeat the action # when the disk space reduces 50M (or the amount of inode reduces 5k) # again. #BB_DISKMON_WARNINTERVAL = "50M,5K" [YOCTO #1589] (Bitbake rev: 4d173d441d2beb8e6492b6b1842682f8cf32e6cc) Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2012-02-26 08:48:15 +00:00
self.dm = monitordisk.diskMonitor(cfgData)
self.dm_event_handler_name = '_bb_diskmonitor_' + str(id(self))
self.dm_event_handler_registered = False
self.rqexe = None
self.worker = {}
self.fakeworker = {}
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
def _start_worker(self, mc, fakeroot = False, rqexec = None):
logger.debug(1, "Starting bitbake-worker")
magic = "decafbad"
if self.cooker.configuration.profile:
magic = "decafbadbad"
if fakeroot:
magic = magic + "beef"
mcdata = self.cooker.databuilder.mcdata[mc]
fakerootcmd = mcdata.getVar("FAKEROOTCMD")
fakerootenv = (mcdata.getVar("FAKEROOTBASEENV") or "").split()
env = os.environ.copy()
for key, value in (var.split('=') for var in fakerootenv):
env[key] = value
worker = subprocess.Popen([fakerootcmd, "bitbake-worker", magic], stdout=subprocess.PIPE, stdin=subprocess.PIPE, env=env)
else:
worker = subprocess.Popen(["bitbake-worker", magic], stdout=subprocess.PIPE, stdin=subprocess.PIPE)
bb.utils.nonblockingfd(worker.stdout)
workerpipe = runQueuePipe(worker.stdout, None, self.cfgData, self, rqexec)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
runqhash = {}
for tid in self.rqdata.runtaskentries:
runqhash[tid] = self.rqdata.runtaskentries[tid].hash
workerdata = {
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
"taskdeps" : self.rqdata.dataCaches[mc].task_deps,
"fakerootenv" : self.rqdata.dataCaches[mc].fakerootenv,
"fakerootdirs" : self.rqdata.dataCaches[mc].fakerootdirs,
"fakerootnoenv" : self.rqdata.dataCaches[mc].fakerootnoenv,
"sigdata" : bb.parse.siggen.get_taskdata(),
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
"runq_hash" : runqhash,
"logdefaultdebug" : bb.msg.loggerDefaultDebugLevel,
"logdefaultverbose" : bb.msg.loggerDefaultVerbose,
"logdefaultverboselogs" : bb.msg.loggerVerboseLogs,
"logdefaultdomain" : bb.msg.loggerDefaultDomains,
"prhost" : self.cooker.prhost,
"buildname" : self.cfgData.getVar("BUILDNAME"),
"date" : self.cfgData.getVar("DATE"),
"time" : self.cfgData.getVar("TIME"),
}
worker.stdin.write(b"<cookerconfig>" + pickle.dumps(self.cooker.configuration) + b"</cookerconfig>")
worker.stdin.write(b"<extraconfigdata>" + pickle.dumps(self.cooker.extraconfigdata) + b"</extraconfigdata>")
worker.stdin.write(b"<workerdata>" + pickle.dumps(workerdata) + b"</workerdata>")
worker.stdin.flush()
return RunQueueWorker(worker, workerpipe)
def _teardown_worker(self, worker):
if not worker:
return
logger.debug(1, "Teardown for bitbake-worker")
try:
worker.process.stdin.write(b"<quit></quit>")
worker.process.stdin.flush()
worker.process.stdin.close()
except IOError:
pass
while worker.process.returncode is None:
worker.pipe.read()
worker.process.poll()
while worker.pipe.read():
continue
worker.pipe.close()
def start_worker(self):
if self.worker:
self.teardown_workers()
self.teardown = False
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for mc in self.rqdata.dataCaches:
self.worker[mc] = self._start_worker(mc)
def start_fakeworker(self, rqexec, mc):
if not mc in self.fakeworker:
self.fakeworker[mc] = self._start_worker(mc, True, rqexec)
def teardown_workers(self):
self.teardown = True
for mc in self.worker:
self._teardown_worker(self.worker[mc])
self.worker = {}
for mc in self.fakeworker:
self._teardown_worker(self.fakeworker[mc])
self.fakeworker = {}
def read_workers(self):
for mc in self.worker:
self.worker[mc].pipe.read()
for mc in self.fakeworker:
self.fakeworker[mc].pipe.read()
def active_fds(self):
fds = []
for mc in self.worker:
fds.append(self.worker[mc].pipe.input)
for mc in self.fakeworker:
fds.append(self.fakeworker[mc].pipe.input)
return fds
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def check_stamp_task(self, tid, taskname = None, recurse = False, cache = None):
def get_timestamp(f):
try:
if not os.access(f, os.F_OK):
return None
return os.stat(f)[stat.ST_MTIME]
except:
return None
(mc, fn, tn, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if taskname is None:
taskname = tn
if self.stamppolicy == "perfile":
fulldeptree = False
else:
fulldeptree = True
stampwhitelist = []
if self.stamppolicy == "whitelist":
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
stampwhitelist = self.rqdata.stampfnwhitelist[mc]
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
stampfile = bb.build.stampfile(taskname, self.rqdata.dataCaches[mc], taskfn)
# If the stamp is missing, it's not current
if not os.access(stampfile, os.F_OK):
logger.debug(2, "Stampfile %s not available", stampfile)
return False
# If it's a 'nostamp' task, it's not current
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'nostamp' in taskdep and taskname in taskdep['nostamp']:
logger.debug(2, "%s.%s is nostamp\n", fn, taskname)
return False
if taskname != "do_setscene" and taskname.endswith("_setscene"):
return True
if cache is None:
cache = {}
iscurrent = True
t1 = get_timestamp(stampfile)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[tid].depends:
if iscurrent:
(mc2, fn2, taskname2, taskfn2) = split_tid_mcfn(dep)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
stampfile2 = bb.build.stampfile(taskname2, self.rqdata.dataCaches[mc2], taskfn2)
stampfile3 = bb.build.stampfile(taskname2 + "_setscene", self.rqdata.dataCaches[mc2], taskfn2)
t2 = get_timestamp(stampfile2)
t3 = get_timestamp(stampfile3)
if t3 and not t2:
continue
if t3 and t3 > t2:
continue
if fn == fn2 or (fulldeptree and fn2 not in stampwhitelist):
if not t2:
logger.debug(2, 'Stampfile %s does not exist', stampfile2)
iscurrent = False
break
if t1 < t2:
logger.debug(2, 'Stampfile %s < %s', stampfile, stampfile2)
iscurrent = False
break
if recurse and iscurrent:
if dep in cache:
iscurrent = cache[dep]
if not iscurrent:
logger.debug(2, 'Stampfile for dependency %s:%s invalid (cached)' % (fn2, taskname2))
else:
iscurrent = self.check_stamp_task(dep, recurse=True, cache=cache)
cache[dep] = iscurrent
if recurse:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
cache[tid] = iscurrent
return iscurrent
def _execute_runqueue(self):
"""
Run the tasks in a queue prepared by rqdata.prepare()
Upon failure, optionally try to recover the build using any alternate providers
(if the abort on failure configuration option isn't set)
"""
retval = True
if self.state is runQueuePrepare:
self.rqexe = RunQueueExecuteDummy(self)
# NOTE: if you add, remove or significantly refactor the stages of this
# process then you should recalculate the weightings here. This is quite
# easy to do - just change the next line temporarily to pass debug=True as
# the last parameter and you'll get a printout of the weightings as well
# as a map to the lines where next_stage() was called. Of course this isn't
# critical, but it helps to keep the progress reporting accurate.
self.rqdata.init_progress_reporter = bb.progress.MultiStageProcessProgressReporter(self.cooker.data,
"Initialising tasks",
[43, 967, 4, 3, 1, 5, 3, 7, 13, 1, 2, 1, 1, 246, 35, 1, 38, 1, 35, 2, 338, 204, 142, 3, 3, 37, 244])
if self.rqdata.prepare() == 0:
self.state = runQueueComplete
else:
self.state = runQueueSceneInit
self.rqdata.init_progress_reporter.next_stage()
# we are ready to run, emit dependency info to any UI or class which
# needs it
depgraph = self.cooker.buildDependTree(self, self.rqdata.taskData)
self.rqdata.init_progress_reporter.next_stage()
bb.event.fire(bb.event.DepTreeGenerated(depgraph), self.cooker.data)
if self.state is runQueueSceneInit:
dump = self.cooker.configuration.dump_signatures
if dump:
self.rqdata.init_progress_reporter.finish()
if 'printdiff' in dump:
invalidtasks = self.print_diffscenetasks()
self.dump_signatures(dump)
if 'printdiff' in dump:
self.write_diffscenetasks(invalidtasks)
self.state = runQueueComplete
else:
self.rqdata.init_progress_reporter.next_stage()
self.start_worker()
self.rqdata.init_progress_reporter.next_stage()
self.rqexe = RunQueueExecuteScenequeue(self)
if self.state is runQueueSceneRun:
if not self.dm_event_handler_registered:
res = bb.event.register(self.dm_event_handler_name,
lambda x: self.dm.check(self) if self.state in [runQueueSceneRun, runQueueRunning, runQueueCleanUp] else False,
('bb.event.HeartbeatEvent',))
self.dm_event_handler_registered = True
retval = self.rqexe.execute()
if self.state is runQueueRunInit:
if self.cooker.configuration.setsceneonly:
self.state = runQueueComplete
else:
# Just in case we didn't setscene
self.rqdata.init_progress_reporter.finish()
logger.info("Executing RunQueue Tasks")
self.rqexe = RunQueueExecuteTasks(self)
self.state = runQueueRunning
if self.state is runQueueRunning:
retval = self.rqexe.execute()
if self.state is runQueueCleanUp:
retval = self.rqexe.finish()
build_done = self.state is runQueueComplete or self.state is runQueueFailed
if build_done and self.dm_event_handler_registered:
bb.event.remove(self.dm_event_handler_name, None)
self.dm_event_handler_registered = False
if build_done and self.rqexe:
self.teardown_workers()
if self.rqexe.stats.failed:
logger.info("Tasks Summary: Attempted %d tasks of which %d didn't need to be rerun and %d failed.", self.rqexe.stats.completed + self.rqexe.stats.failed, self.rqexe.stats.skipped, self.rqexe.stats.failed)
else:
# Let's avoid the word "failed" if nothing actually did
logger.info("Tasks Summary: Attempted %d tasks of which %d didn't need to be rerun and all succeeded.", self.rqexe.stats.completed, self.rqexe.stats.skipped)
if self.state is runQueueFailed:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if not self.rqdata.taskData[''].tryaltconfigs:
raise bb.runqueue.TaskFailure(self.rqexe.failed_tids)
for tid in self.rqexe.failed_tids:
(mc, fn, tn, _) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rqdata.taskData[mc].fail_fn(fn)
self.rqdata.reset()
if self.state is runQueueComplete:
# All done
return False
# Loop
return retval
def execute_runqueue(self):
# Catch unexpected exceptions and ensure we exit when an error occurs, not loop.
try:
return self._execute_runqueue()
except bb.runqueue.TaskFailure:
raise
except SystemExit:
raise
except bb.BBHandledException:
try:
self.teardown_workers()
except:
pass
self.state = runQueueComplete
raise
except Exception as err:
logger.exception("An uncaught exception occurred in runqueue")
try:
self.teardown_workers()
except:
pass
self.state = runQueueComplete
raise
def finish_runqueue(self, now = False):
if not self.rqexe:
self.state = runQueueComplete
return
if now:
self.rqexe.finish_now()
else:
self.rqexe.finish()
def rq_dump_sigfn(self, fn, options):
bb_cache = bb.cache.NoCache(self.cooker.databuilder)
the_data = bb_cache.loadDataFull(fn, self.cooker.collection.get_file_appends(fn))
siggen = bb.parse.siggen
dataCaches = self.rqdata.dataCaches
siggen.dump_sigfn(fn, dataCaches, options)
def dump_signatures(self, options):
fns = set()
bb.note("Reparsing files to collect dependency data")
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
fn = fn_from_tid(tid)
fns.add(fn)
max_process = int(self.cfgData.getVar("BB_NUMBER_PARSE_THREADS") or os.cpu_count() or 1)
# We cannot use the real multiprocessing.Pool easily due to some local data
# that can't be pickled. This is a cheap multi-process solution.
launched = []
while fns:
if len(launched) < max_process:
p = Process(target=self.rq_dump_sigfn, args=(fns.pop(), options))
p.start()
launched.append(p)
for q in launched:
# The finished processes are joined when calling is_alive()
if not q.is_alive():
launched.remove(q)
for p in launched:
p.join()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
bb.parse.siggen.dump_sigs(self.rqdata.dataCaches, options)
return
def print_diffscenetasks(self):
valid = []
sq_hash = []
sq_hashfn = []
sq_fn = []
sq_taskname = []
sq_task = []
noexec = []
stamppresent = []
valid_new = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'noexec' in taskdep and taskname in taskdep['noexec']:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
noexec.append(tid)
continue
sq_fn.append(fn)
sq_hashfn.append(self.rqdata.dataCaches[mc].hashfn[taskfn])
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_hash.append(self.rqdata.runtaskentries[tid].hash)
sq_taskname.append(taskname)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_task.append(tid)
locs = { "sq_fn" : sq_fn, "sq_task" : sq_taskname, "sq_hash" : sq_hash, "sq_hashfn" : sq_hashfn, "d" : self.cooker.data }
try:
call = self.hashvalidate + "(sq_fn, sq_task, sq_hash, sq_hashfn, d, siginfo=True)"
valid = bb.utils.better_eval(call, locs)
# Handle version with no siginfo parameter
except TypeError:
call = self.hashvalidate + "(sq_fn, sq_task, sq_hash, sq_hashfn, d)"
valid = bb.utils.better_eval(call, locs)
for v in valid:
valid_new.add(sq_task[v])
# Tasks which are both setscene and noexec never care about dependencies
# We therefore find tasks which are setscene and noexec and mark their
# unique dependencies as valid.
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in noexec:
if tid not in self.rqdata.runq_setscene_tids:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[tid].depends:
hasnoexecparents = True
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep2 in self.rqdata.runtaskentries[dep].revdeps:
if dep2 in self.rqdata.runq_setscene_tids and dep2 in noexec:
continue
hasnoexecparents = False
break
if hasnoexecparents:
valid_new.add(dep)
invalidtasks = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
if tid not in valid_new and tid not in noexec:
invalidtasks.add(tid)
found = set()
processed = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in invalidtasks:
toprocess = set([tid])
while toprocess:
next = set()
for t in toprocess:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[t].depends:
if dep in invalidtasks:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
found.add(tid)
if dep not in processed:
processed.add(dep)
next.add(dep)
toprocess = next
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in found:
toprocess = set()
tasklist = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in invalidtasks.difference(found):
tasklist.append(tid)
if tasklist:
bb.plain("The differences between the current build and any cached tasks start at the following tasks:\n" + "\n".join(tasklist))
return invalidtasks.difference(found)
def write_diffscenetasks(self, invalidtasks):
# Define recursion callback
def recursecb(key, hash1, hash2):
hashes = [hash1, hash2]
hashfiles = bb.siggen.find_siginfo(key, None, hashes, self.cfgData)
recout = []
if len(hashfiles) == 2:
out2 = bb.siggen.compare_sigfiles(hashfiles[hash1], hashfiles[hash2], recursecb)
recout.extend(list(' ' + l for l in out2))
else:
recout.append("Unable to find matching sigdata for %s with hashes %s or %s" % (key, hash1, hash2))
return recout
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in invalidtasks:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
h = self.rqdata.runtaskentries[tid].hash
matches = bb.siggen.find_siginfo(pn, taskname, [], self.cfgData)
match = None
for m in matches:
if h in m:
match = m
if match is None:
bb.fatal("Can't find a task we're supposed to have written out? (hash: %s)?" % h)
matches = {k : v for k, v in iter(matches.items()) if h not in k}
if matches:
latestmatch = sorted(matches.keys(), key=lambda f: matches[f])[-1]
prevh = __find_md5__.search(latestmatch).group(0)
output = bb.siggen.compare_sigfiles(latestmatch, match, recursecb)
bb.plain("\nTask %s:%s couldn't be used from the cache because:\n We need hash %s, closest matching task was %s\n " % (pn, taskname, h, prevh) + '\n '.join(output))
class RunQueueExecute:
def __init__(self, rq):
self.rq = rq
self.cooker = rq.cooker
self.cfgData = rq.cfgData
self.rqdata = rq.rqdata
self.number_tasks = int(self.cfgData.getVar("BB_NUMBER_THREADS") or 1)
self.scheduler = self.cfgData.getVar("BB_SCHEDULER") or "speed"
self.runq_buildable = set()
self.runq_running = set()
self.runq_complete = set()
self.build_stamps = {}
self.build_stamps2 = []
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.failed_tids = []
self.stampcache = {}
for mc in rq.worker:
rq.worker[mc].pipe.setrunqueueexec(self)
for mc in rq.fakeworker:
rq.fakeworker[mc].pipe.setrunqueueexec(self)
if self.number_tasks <= 0:
bb.fatal("Invalid BB_NUMBER_THREADS %s" % self.number_tasks)
def runqueue_process_waitpid(self, task, status):
# self.build_stamps[pid] may not exist when use shared work directory.
if task in self.build_stamps:
self.build_stamps2.remove(self.build_stamps[task])
del self.build_stamps[task]
if status != 0:
self.task_fail(task, status)
else:
self.task_complete(task)
return True
def finish_now(self):
for mc in self.rq.worker:
try:
self.rq.worker[mc].process.stdin.write(b"<finishnow></finishnow>")
self.rq.worker[mc].process.stdin.flush()
except IOError:
# worker must have died?
pass
for mc in self.rq.fakeworker:
try:
self.rq.fakeworker[mc].process.stdin.write(b"<finishnow></finishnow>")
self.rq.fakeworker[mc].process.stdin.flush()
except IOError:
# worker must have died?
pass
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if len(self.failed_tids) != 0:
self.rq.state = runQueueFailed
return
self.rq.state = runQueueComplete
return
def finish(self):
self.rq.state = runQueueCleanUp
if self.stats.active > 0:
bb.event.fire(runQueueExitWait(self.stats.active), self.cfgData)
self.rq.read_workers()
return self.rq.active_fds()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if len(self.failed_tids) != 0:
self.rq.state = runQueueFailed
return True
self.rq.state = runQueueComplete
return True
def check_dependencies(self, task, taskdeps, setscene = False):
if not self.rq.depvalidate:
return False
taskdata = {}
taskdeps.add(task)
for dep in taskdeps:
(mc, fn, taskname, taskfn) = split_tid_mcfn(dep)
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
taskdata[dep] = [pn, taskname, fn]
call = self.rq.depvalidate + "(task, taskdata, notneeded, d)"
locs = { "task" : task, "taskdata" : taskdata, "notneeded" : self.scenequeue_notneeded, "d" : self.cooker.data }
valid = bb.utils.better_eval(call, locs)
return valid
class RunQueueExecuteDummy(RunQueueExecute):
def __init__(self, rq):
self.rq = rq
self.stats = RunQueueStats(0)
def finish(self):
self.rq.state = runQueueComplete
return
class RunQueueExecuteTasks(RunQueueExecute):
def __init__(self, rq):
RunQueueExecute.__init__(self, rq)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.stats = RunQueueStats(len(self.rqdata.runtaskentries))
self.stampcache = {}
initial_covered = self.rq.scenequeue_covered.copy()
# Mark initial buildable tasks
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
if len(self.rqdata.runtaskentries[tid].depends) == 0:
self.runq_buildable.add(tid)
if len(self.rqdata.runtaskentries[tid].revdeps) > 0 and self.rqdata.runtaskentries[tid].revdeps.issubset(self.rq.scenequeue_covered):
self.rq.scenequeue_covered.add(tid)
found = True
while found:
found = False
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
if tid in self.rq.scenequeue_covered:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(1, 'Considering %s: %s' % (tid, str(self.rqdata.runtaskentries[tid].revdeps)))
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if len(self.rqdata.runtaskentries[tid].revdeps) > 0 and self.rqdata.runtaskentries[tid].revdeps.issubset(self.rq.scenequeue_covered):
if tid in self.rq.scenequeue_notcovered:
continue
found = True
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.rq.scenequeue_covered.add(tid)
logger.debug(1, 'Skip list (pre setsceneverify) %s', sorted(self.rq.scenequeue_covered))
# Allow the metadata to elect for setscene tasks to run anyway
covered_remove = set()
if self.rq.setsceneverify:
invalidtasks = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
tasknames = {}
fns = {}
for tid in self.rqdata.runtaskentries:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
fns[tid] = taskfn
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
tasknames[tid] = taskname
if 'noexec' in taskdep and taskname in taskdep['noexec']:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(tid, taskname + "_setscene", cache=self.stampcache):
logger.debug(2, 'Setscene stamp current for task %s', tid)
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(tid, taskname, recurse = True, cache=self.stampcache):
logger.debug(2, 'Normal stamp current for task %s', tid)
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
invalidtasks.append(tid)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
call = self.rq.setsceneverify + "(covered, tasknames, fns, d, invalidtasks=invalidtasks)"
locs = { "covered" : self.rq.scenequeue_covered, "tasknames" : tasknames, "fns" : fns, "d" : self.cooker.data, "invalidtasks" : invalidtasks }
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
covered_remove = bb.utils.better_eval(call, locs)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
def removecoveredtask(tid):
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskname = taskname + '_setscene'
bb.build.del_stamp(taskname, self.rqdata.dataCaches[mc], taskfn)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.rq.scenequeue_covered.remove(tid)
toremove = covered_remove
for task in toremove:
logger.debug(1, 'Not skipping task %s due to setsceneverify', task)
while toremove:
covered_remove = []
for task in toremove:
removecoveredtask(task)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for deptask in self.rqdata.runtaskentries[task].depends:
if deptask not in self.rq.scenequeue_covered:
continue
if deptask in toremove or deptask in covered_remove or deptask in initial_covered:
continue
logger.debug(1, 'Task %s depends on task %s so not skipping' % (task, deptask))
covered_remove.append(deptask)
toremove = covered_remove
logger.debug(1, 'Full skip list %s', self.rq.scenequeue_covered)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
for mc in self.rqdata.dataCaches:
target_pairs = []
for tid in self.rqdata.target_tids:
(tidmc, fn, taskname, _) = split_tid_mcfn(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if tidmc == mc:
target_pairs.append((fn, taskname))
event.fire(bb.event.StampUpdate(target_pairs, self.rqdata.dataCaches[mc].stamp), self.cfgData)
schedulers = self.get_schedulers()
for scheduler in schedulers:
if self.scheduler == scheduler.name:
self.sched = scheduler(self, self.rqdata)
logger.debug(1, "Using runqueue scheduler '%s'", scheduler.name)
break
else:
bb.fatal("Invalid scheduler '%s'. Available schedulers: %s" %
(self.scheduler, ", ".join(obj.name for obj in schedulers)))
def get_schedulers(self):
schedulers = set(obj for obj in globals().values()
if type(obj) is type and
issubclass(obj, RunQueueScheduler))
user_schedulers = self.cfgData.getVar("BB_SCHEDULERS")
if user_schedulers:
for sched in user_schedulers.split():
if not "." in sched:
bb.note("Ignoring scheduler '%s' from BB_SCHEDULERS: not an import" % sched)
continue
modname, name = sched.rsplit(".", 1)
try:
module = __import__(modname, fromlist=(name,))
except ImportError as exc:
logger.critical("Unable to import scheduler '%s' from '%s': %s" % (name, modname, exc))
raise SystemExit(1)
else:
schedulers.add(getattr(module, name))
return schedulers
def setbuildable(self, task):
self.runq_buildable.add(task)
self.sched.newbuilable(task)
def task_completeoutright(self, task):
"""
Mark a task as completed
Look at the reverse dependencies and mark any task with
completed dependencies as buildable
"""
self.runq_complete.add(task)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for revdep in self.rqdata.runtaskentries[task].revdeps:
if revdep in self.runq_running:
continue
if revdep in self.runq_buildable:
continue
alldeps = 1
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[revdep].depends:
if dep not in self.runq_complete:
alldeps = 0
if alldeps == 1:
self.setbuildable(revdep)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
fn = fn_from_tid(revdep)
taskname = taskname_from_tid(revdep)
logger.debug(1, "Marking task %s as buildable", revdep)
def task_complete(self, task):
self.stats.taskCompleted()
bb.event.fire(runQueueTaskCompleted(task, self.stats, self.rq), self.cfgData)
self.task_completeoutright(task)
def task_fail(self, task, exitcode):
"""
Called when a task has failed
Updates the state engine with the failure
"""
self.stats.taskFailed()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.failed_tids.append(task)
bb.event.fire(runQueueTaskFailed(task, self.stats, exitcode, self.rq), self.cfgData)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if self.rqdata.taskData[''].abort:
self.rq.state = runQueueCleanUp
def task_skip(self, task, reason):
self.runq_running.add(task)
self.setbuildable(task)
bb.event.fire(runQueueTaskSkipped(task, self.stats, self.rq, reason), self.cfgData)
self.task_completeoutright(task)
self.stats.taskCompleted()
self.stats.taskSkipped()
def execute(self):
"""
Run the tasks in a queue prepared by rqdata.prepare()
"""
if self.rqdata.setscenewhitelist and not self.rqdata.setscenewhitelist_checked:
self.rqdata.setscenewhitelist_checked = True
# Check tasks that are going to run against the whitelist
def check_norun_task(tid, showerror=False):
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
# Ignore covered tasks
if tid in self.rq.scenequeue_covered:
return False
# Ignore stamped tasks
if self.rq.check_stamp_task(tid, taskname, cache=self.stampcache):
return False
# Ignore noexec tasks
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'noexec' in taskdep and taskname in taskdep['noexec']:
return False
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
if not check_setscene_enforce_whitelist(pn, taskname, self.rqdata.setscenewhitelist):
if showerror:
if tid in self.rqdata.runq_setscene_tids:
logger.error('Task %s.%s attempted to execute unexpectedly and should have been setscened' % (pn, taskname))
else:
logger.error('Task %s.%s attempted to execute unexpectedly' % (pn, taskname))
return True
return False
# Look to see if any tasks that we think shouldn't run are going to
unexpected = False
for tid in self.rqdata.runtaskentries:
if check_norun_task(tid):
unexpected = True
break
if unexpected:
# Run through the tasks in the rough order they'd have executed and print errors
# (since the order can be useful - usually missing sstate for the last few tasks
# is the cause of the problem)
task = self.sched.next()
while task is not None:
check_norun_task(task, showerror=True)
self.task_skip(task, 'Setscene enforcement check')
task = self.sched.next()
self.rq.state = runQueueCleanUp
return True
self.rq.read_workers()
if self.stats.total == 0:
# nothing to do
self.rq.state = runQueueCleanUp
task = self.sched.next()
if task is not None:
(mc, fn, taskname, taskfn) = split_tid_mcfn(task)
if task in self.rq.scenequeue_covered:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(2, "Setscene covered task %s", task)
self.task_skip(task, "covered")
return True
if self.rq.check_stamp_task(task, taskname, cache=self.stampcache):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(2, "Stamp current task %s", task)
self.task_skip(task, "existing")
return True
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'noexec' in taskdep and taskname in taskdep['noexec']:
startevent = runQueueTaskStarted(task, self.stats, self.rq,
noexec=True)
bb.event.fire(startevent, self.cfgData)
self.runq_running.add(task)
self.stats.taskActive()
if not (self.cooker.configuration.dry_run or self.rqdata.setscene_enforce):
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
bb.build.make_stamp(taskname, self.rqdata.dataCaches[mc], taskfn)
self.task_complete(task)
return True
else:
startevent = runQueueTaskStarted(task, self.stats, self.rq)
bb.event.fire(startevent, self.cfgData)
taskdepdata = self.build_taskdepdata(task)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'fakeroot' in taskdep and taskname in taskdep['fakeroot'] and not (self.cooker.configuration.dry_run or self.rqdata.setscene_enforce):
if not mc in self.rq.fakeworker:
try:
self.rq.start_fakeworker(self, mc)
except OSError as exc:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.critical("Failed to spawn fakeroot worker to run %s: %s" % (task, str(exc)))
self.rq.state = runQueueFailed
self.stats.taskFailed()
return True
self.rq.fakeworker[mc].process.stdin.write(b"<runtask>" + pickle.dumps((taskfn, task, taskname, False, self.cooker.collection.get_file_appends(fn), taskdepdata, self.rqdata.setscene_enforce)) + b"</runtask>")
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rq.fakeworker[mc].process.stdin.flush()
else:
self.rq.worker[mc].process.stdin.write(b"<runtask>" + pickle.dumps((taskfn, task, taskname, False, self.cooker.collection.get_file_appends(taskfn), taskdepdata, self.rqdata.setscene_enforce)) + b"</runtask>")
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rq.worker[mc].process.stdin.flush()
self.build_stamps[task] = bb.build.stampfile(taskname, self.rqdata.dataCaches[mc], taskfn, noextra=True)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.build_stamps2.append(self.build_stamps[task])
self.runq_running.add(task)
self.stats.taskActive()
if self.stats.active < self.number_tasks:
return True
if self.stats.active > 0:
self.rq.read_workers()
return self.rq.active_fds()
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if len(self.failed_tids) != 0:
self.rq.state = runQueueFailed
return True
# Sanity Checks
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for task in self.rqdata.runtaskentries:
if task not in self.runq_buildable:
logger.error("Task %s never buildable!", task)
if task not in self.runq_running:
logger.error("Task %s never ran!", task)
if task not in self.runq_complete:
logger.error("Task %s never completed!", task)
self.rq.state = runQueueComplete
return True
def build_taskdepdata(self, task):
taskdepdata = {}
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
next = self.rqdata.runtaskentries[task].depends
next.add(task)
while next:
additional = []
for revdep in next:
(mc, fn, taskname, taskfn) = split_tid_mcfn(revdep)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
deps = self.rqdata.runtaskentries[revdep].depends
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
provides = self.rqdata.dataCaches[mc].fn_provides[taskfn]
taskhash = self.rqdata.runtaskentries[revdep].hash
taskdepdata[revdep] = [pn, taskname, fn, deps, provides, taskhash]
for revdep2 in deps:
if revdep2 not in taskdepdata:
additional.append(revdep2)
next = additional
#bb.note("Task %s: " % task + str(taskdepdata).replace("], ", "],\n"))
return taskdepdata
class RunQueueExecuteScenequeue(RunQueueExecute):
def __init__(self, rq):
RunQueueExecute.__init__(self, rq)
self.scenequeue_covered = set()
self.scenequeue_notcovered = set()
self.scenequeue_notneeded = set()
# If we don't have any setscene functions, skip this step
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if len(self.rqdata.runq_setscene_tids) == 0:
rq.scenequeue_covered = set()
rq.state = runQueueRunInit
return
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.stats = RunQueueStats(len(self.rqdata.runq_setscene_tids))
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_revdeps = {}
sq_revdeps_new = {}
sq_revdeps_squash = {}
self.sq_harddeps = {}
self.stamps = {}
# We need to construct a dependency graph for the setscene functions. Intermediate
# dependencies between the setscene tasks only complicate the code. This code
# therefore aims to collapse the huge runqueue dependency tree into a smaller one
# only containing the setscene functions.
self.rqdata.init_progress_reporter.next_stage()
# First process the chains up to the first setscene task.
endpoints = {}
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
sq_revdeps[tid] = copy.copy(self.rqdata.runtaskentries[tid].revdeps)
sq_revdeps_new[tid] = set()
if (len(sq_revdeps[tid]) == 0) and tid not in self.rqdata.runq_setscene_tids:
#bb.warn("Added endpoint %s" % (tid))
endpoints[tid] = set()
self.rqdata.init_progress_reporter.next_stage()
# Secondly process the chains between setscene tasks.
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runq_setscene_tids:
#bb.warn("Added endpoint 2 %s" % (tid))
for dep in self.rqdata.runtaskentries[tid].depends:
if tid in sq_revdeps[dep]:
sq_revdeps[dep].remove(tid)
if dep not in endpoints:
endpoints[dep] = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
#bb.warn(" Added endpoint 3 %s" % (dep))
endpoints[dep].add(tid)
self.rqdata.init_progress_reporter.next_stage()
def process_endpoints(endpoints):
newendpoints = {}
for point, task in endpoints.items():
tasks = set()
if task:
tasks |= task
if sq_revdeps_new[point]:
tasks |= sq_revdeps_new[point]
sq_revdeps_new[point] = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if point in self.rqdata.runq_setscene_tids:
sq_revdeps_new[point] = tasks
tasks = set()
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[point].depends:
if point in sq_revdeps[dep]:
sq_revdeps[dep].remove(point)
if tasks:
sq_revdeps_new[dep] |= tasks
if len(sq_revdeps[dep]) == 0 and dep not in self.rqdata.runq_setscene_tids:
newendpoints[dep] = task
if len(newendpoints) != 0:
process_endpoints(newendpoints)
process_endpoints(endpoints)
self.rqdata.init_progress_reporter.next_stage()
# Build a list of setscene tasks which are "unskippable"
# These are direct endpoints referenced by the build
endpoints2 = {}
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_revdeps2 = {}
sq_revdeps_new2 = {}
def process_endpoints2(endpoints):
newendpoints = {}
for point, task in endpoints.items():
tasks = set([point])
if task:
tasks |= task
if sq_revdeps_new2[point]:
tasks |= sq_revdeps_new2[point]
sq_revdeps_new2[point] = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if point in self.rqdata.runq_setscene_tids:
sq_revdeps_new2[point] = tasks
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in self.rqdata.runtaskentries[point].depends:
if point in sq_revdeps2[dep]:
sq_revdeps2[dep].remove(point)
if tasks:
sq_revdeps_new2[dep] |= tasks
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if (len(sq_revdeps2[dep]) == 0 or len(sq_revdeps_new2[dep]) != 0) and dep not in self.rqdata.runq_setscene_tids:
newendpoints[dep] = tasks
if len(newendpoints) != 0:
process_endpoints2(newendpoints)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runtaskentries:
sq_revdeps2[tid] = copy.copy(self.rqdata.runtaskentries[tid].revdeps)
sq_revdeps_new2[tid] = set()
if (len(sq_revdeps2[tid]) == 0) and tid not in self.rqdata.runq_setscene_tids:
endpoints2[tid] = set()
process_endpoints2(endpoints2)
self.unskippable = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runq_setscene_tids:
if sq_revdeps_new2[tid]:
self.unskippable.append(tid)
self.rqdata.init_progress_reporter.next_stage(len(self.rqdata.runtaskentries))
for taskcounter, tid in enumerate(self.rqdata.runtaskentries):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if tid in self.rqdata.runq_setscene_tids:
deps = set()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for dep in sq_revdeps_new[tid]:
deps.add(dep)
sq_revdeps_squash[tid] = deps
elif len(sq_revdeps_new[tid]) != 0:
bb.msg.fatal("RunQueue", "Something went badly wrong during scenequeue generation, aborting. Please report this problem.")
self.rqdata.init_progress_reporter.update(taskcounter)
self.rqdata.init_progress_reporter.next_stage()
# Resolve setscene inter-task dependencies
# e.g. do_sometask_setscene[depends] = "targetname:do_someothertask_setscene"
# Note that anything explicitly depended upon will have its reverse dependencies removed to avoid circular dependencies
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.rqdata.runq_setscene_tids:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
realtid = tid + "_setscene"
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
idepends = self.rqdata.taskData[mc].taskentries[realtid].idepends
self.stamps[tid] = bb.build.stampfile(taskname + "_setscene", self.rqdata.dataCaches[mc], taskfn, noextra=True)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for (depname, idependtask) in idepends:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if depname not in self.rqdata.taskData[mc].build_targets:
continue
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
depfn = self.rqdata.taskData[mc].build_targets[depname][0]
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if depfn is None:
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
deptid = depfn + ":" + idependtask.replace("_setscene", "")
if deptid not in self.rqdata.runtaskentries:
bb.msg.fatal("RunQueue", "Task %s depends upon non-existent task %s:%s" % (realtid, depfn, idependtask))
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if not deptid in self.sq_harddeps:
self.sq_harddeps[deptid] = set()
self.sq_harddeps[deptid].add(tid)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_revdeps_squash[tid].add(deptid)
# Have to zero this to avoid circular dependencies
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_revdeps_squash[deptid] = set()
self.rqdata.init_progress_reporter.next_stage()
for task in self.sq_harddeps:
for dep in self.sq_harddeps[task]:
sq_revdeps_squash[dep].add(task)
self.rqdata.init_progress_reporter.next_stage()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
#for tid in sq_revdeps_squash:
# for dep in sq_revdeps_squash[tid]:
# data = data + "\n %s" % dep
# bb.warn("Task %s_setscene: is %s " % (tid, data
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.sq_deps = {}
self.sq_revdeps = sq_revdeps_squash
self.sq_revdeps2 = copy.deepcopy(self.sq_revdeps)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.sq_revdeps:
self.sq_deps[tid] = set()
for tid in self.sq_revdeps:
for dep in self.sq_revdeps[tid]:
self.sq_deps[dep].add(tid)
self.rqdata.init_progress_reporter.next_stage()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.sq_revdeps:
if len(self.sq_revdeps[tid]) == 0:
self.runq_buildable.add(tid)
self.rqdata.init_progress_reporter.finish()
self.outrightfail = []
if self.rq.hashvalidate:
sq_hash = []
sq_hashfn = []
sq_fn = []
sq_taskname = []
sq_task = []
noexec = []
stamppresent = []
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.sq_revdeps:
(mc, fn, taskname, taskfn) = split_tid_mcfn(tid)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'noexec' in taskdep and taskname in taskdep['noexec']:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
noexec.append(tid)
self.task_skip(tid)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
bb.build.make_stamp(taskname + "_setscene", self.rqdata.dataCaches[mc], taskfn)
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(tid, taskname + "_setscene", cache=self.stampcache):
logger.debug(2, 'Setscene stamp current for task %s', tid)
stamppresent.append(tid)
self.task_skip(tid)
continue
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(tid, taskname, recurse = True, cache=self.stampcache):
logger.debug(2, 'Normal stamp current for task %s', tid)
stamppresent.append(tid)
self.task_skip(tid)
continue
sq_fn.append(fn)
sq_hashfn.append(self.rqdata.dataCaches[mc].hashfn[taskfn])
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_hash.append(self.rqdata.runtaskentries[tid].hash)
sq_taskname.append(taskname)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
sq_task.append(tid)
call = self.rq.hashvalidate + "(sq_fn, sq_task, sq_hash, sq_hashfn, d)"
locs = { "sq_fn" : sq_fn, "sq_task" : sq_taskname, "sq_hash" : sq_hash, "sq_hashfn" : sq_hashfn, "d" : self.cooker.data }
valid = bb.utils.better_eval(call, locs)
valid_new = stamppresent
for v in valid:
valid_new.append(sq_task[v])
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for tid in self.sq_revdeps:
if tid not in valid_new and tid not in noexec:
logger.debug(2, 'No package found, so skipping setscene task %s', tid)
self.outrightfail.append(tid)
logger.info('Executing SetScene Tasks')
self.rq.state = runQueueSceneRun
def scenequeue_updatecounters(self, task, fail = False):
for dep in self.sq_deps[task]:
if fail and task in self.sq_harddeps and dep in self.sq_harddeps[task]:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(2, "%s was unavailable and is a hard dependency of %s so skipping" % (task, dep))
self.scenequeue_updatecounters(dep, fail)
continue
if task not in self.sq_revdeps2[dep]:
# May already have been removed by the fail case above
continue
self.sq_revdeps2[dep].remove(task)
if len(self.sq_revdeps2[dep]) == 0:
self.runq_buildable.add(dep)
def task_completeoutright(self, task):
"""
Mark a task as completed
Look at the reverse dependencies and mark any task with
completed dependencies as buildable
"""
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(1, 'Found task %s which could be accelerated', task)
self.scenequeue_covered.add(task)
self.scenequeue_updatecounters(task)
def check_taskfail(self, task):
if self.rqdata.setscenewhitelist:
realtask = task.split('_setscene')[0]
(mc, fn, taskname, taskfn) = split_tid_mcfn(realtask)
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
if not check_setscene_enforce_whitelist(pn, taskname, self.rqdata.setscenewhitelist):
logger.error('Task %s.%s failed' % (pn, taskname + "_setscene"))
self.rq.state = runQueueCleanUp
def task_complete(self, task):
self.stats.taskCompleted()
bb.event.fire(sceneQueueTaskCompleted(task, self.stats, self.rq), self.cfgData)
self.task_completeoutright(task)
def task_fail(self, task, result):
self.stats.taskFailed()
bb.event.fire(sceneQueueTaskFailed(task, self.stats, result, self), self.cfgData)
self.scenequeue_notcovered.add(task)
self.scenequeue_updatecounters(task, True)
self.check_taskfail(task)
def task_failoutright(self, task):
self.runq_running.add(task)
self.runq_buildable.add(task)
self.stats.taskCompleted()
self.stats.taskSkipped()
self.scenequeue_notcovered.add(task)
self.scenequeue_updatecounters(task, True)
def task_skip(self, task):
self.runq_running.add(task)
self.runq_buildable.add(task)
self.task_completeoutright(task)
self.stats.taskCompleted()
self.stats.taskSkipped()
def execute(self):
"""
Run the tasks in a queue prepared by prepare_runqueue
"""
self.rq.read_workers()
task = None
if self.stats.active < self.number_tasks:
# Find the next setscene to run
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
for nexttask in self.rqdata.runq_setscene_tids:
if nexttask in self.runq_buildable and nexttask not in self.runq_running and self.stamps[nexttask] not in self.build_stamps.values():
if nexttask in self.unskippable:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(2, "Setscene task %s is unskippable" % nexttask)
if nexttask not in self.unskippable and len(self.sq_revdeps[nexttask]) > 0 and self.sq_revdeps[nexttask].issubset(self.scenequeue_covered) and self.check_dependencies(nexttask, self.sq_revdeps[nexttask], True):
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
fn = fn_from_tid(nexttask)
foundtarget = False
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if nexttask in self.rqdata.target_tids:
foundtarget = True
if not foundtarget:
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
logger.debug(2, "Skipping setscene for task %s" % nexttask)
self.task_skip(nexttask)
self.scenequeue_notneeded.add(nexttask)
return True
if nexttask in self.outrightfail:
self.task_failoutright(nexttask)
return True
task = nexttask
break
if task is not None:
(mc, fn, taskname, taskfn) = split_tid_mcfn(task)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskname = taskname + "_setscene"
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(task, taskname_from_tid(task), recurse = True, cache=self.stampcache):
logger.debug(2, 'Stamp for underlying task %s is current, so skipping setscene variant', task)
self.task_failoutright(task)
return True
if self.cooker.configuration.force:
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
if task in self.rqdata.target_tids:
self.task_failoutright(task)
return True
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
if self.rq.check_stamp_task(task, taskname, cache=self.stampcache):
logger.debug(2, 'Setscene stamp current task %s, so skip it and its dependencies', task)
self.task_skip(task)
return True
startevent = sceneQueueTaskStarted(task, self.stats, self.rq)
bb.event.fire(startevent, self.cfgData)
taskdepdata = self.build_taskdepdata(task)
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
taskdep = self.rqdata.dataCaches[mc].task_deps[taskfn]
if 'fakeroot' in taskdep and taskname in taskdep['fakeroot'] and not self.cooker.configuration.dry_run:
if not mc in self.rq.fakeworker:
self.rq.start_fakeworker(self, mc)
self.rq.fakeworker[mc].process.stdin.write(b"<runtask>" + pickle.dumps((taskfn, task, taskname, True, self.cooker.collection.get_file_appends(taskfn), taskdepdata, False)) + b"</runtask>")
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rq.fakeworker[mc].process.stdin.flush()
else:
self.rq.worker[mc].process.stdin.write(b"<runtask>" + pickle.dumps((taskfn, task, taskname, True, self.cooker.collection.get_file_appends(taskfn), taskdepdata, False)) + b"</runtask>")
bitbake: bitbake: Initial multi-config support This patch adds the notion of supporting multiple configurations within a single build. To enable it, set a line in local.conf like: BBMULTICONFIG = "configA configB configC" This would tell bitbake that before it parses the base configuration, it should load conf/configA.conf and so on for each different configuration. These would contain lines like: MACHINE = "A" or other variables which can be set which can be built in the same build directory (or change TMPDIR not to conflict). One downside I've already discovered is that if we want to inherit this file right at the start of parsing, the only place you can put the configurations is in "cwd", since BBPATH isn't constructed until the layers are parsed and therefore using it as a preconf file isn't possible unless its located there. Execution of these targets takes the form "bitbake multiconfig:configA:core-image-minimal core-image-sato" so similar to our virtclass approach for native/nativesdk/multilib using BBCLASSEXTEND. Implementation wise, the implication is that instead of tasks being uniquely referenced with "recipename/fn:task" it now needs to be "configuration:recipename:task". We already started using "virtual" filenames for recipes when we implemented BBCLASSEXTEND and this patch adds a new prefix to these, "multiconfig:<configname>:" and hence avoid changes to a large part of the codebase thanks to this. databuilder has an internal array of data stores and uses the right one depending on the supplied virtual filename. That trick allows us to use the existing parsing code including the multithreading mostly unchanged as well as most of the cache code. For recipecache, we end up with a dict of these accessed by multiconfig (mc). taskdata and runqueue can only cope with one recipecache so for taskdata, we pass in each recipecache and have it compute the result and end up with an array of taskdatas. We can only have one runqueue so there extensive changes there. This initial implementation has some drawbacks: a) There are no inter-multi-configuration dependencies as yet b) There are no sstate optimisations. This means if the build uses the same object twice in say two different TMPDIRs, it will either load from an existing sstate cache at the start or build it twice. We can then in due course look at ways in which it would only build it once and then reuse it. This will likely need significant changes to the way sstate currently works to make that possible. (Bitbake rev: 5287991691578825c847bac2368e9b51c0ede3f0) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-08-16 16:47:06 +00:00
self.rq.worker[mc].process.stdin.flush()
self.build_stamps[task] = bb.build.stampfile(taskname, self.rqdata.dataCaches[mc], taskfn, noextra=True)
self.build_stamps2.append(self.build_stamps[task])
self.runq_running.add(task)
self.stats.taskActive()
if self.stats.active < self.number_tasks:
return True
if self.stats.active > 0:
self.rq.read_workers()
return self.rq.active_fds()
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
#for tid in self.sq_revdeps:
# if tid not in self.runq_running:
# buildable = tid in self.runq_buildable
# revdeps = self.sq_revdeps[tid]
# bb.warn("Found we didn't run %s %s %s" % (tid, buildable, str(revdeps)))
self.rq.scenequeue_covered = self.scenequeue_covered
self.rq.scenequeue_notcovered = self.scenequeue_notcovered
logger.debug(1, 'We can skip tasks %s', sorted(self.rq.scenequeue_covered))
self.rq.state = runQueueRunInit
completeevent = sceneQueueComplete(self.stats, self.rq)
bb.event.fire(completeevent, self.cfgData)
return True
def runqueue_process_waitpid(self, task, status):
RunQueueExecute.runqueue_process_waitpid(self, task, status)
def build_taskdepdata(self, task):
def getsetscenedeps(tid):
deps = set()
(mc, fn, taskname, _) = split_tid_mcfn(tid)
realtid = tid + "_setscene"
idepends = self.rqdata.taskData[mc].taskentries[realtid].idepends
for (depname, idependtask) in idepends:
if depname not in self.rqdata.taskData[mc].build_targets:
continue
depfn = self.rqdata.taskData[mc].build_targets[depname][0]
if depfn is None:
continue
deptid = depfn + ":" + idependtask.replace("_setscene", "")
deps.add(deptid)
return deps
taskdepdata = {}
next = getsetscenedeps(task)
next.add(task)
while next:
additional = []
for revdep in next:
(mc, fn, taskname, taskfn) = split_tid_mcfn(revdep)
pn = self.rqdata.dataCaches[mc].pkg_fn[taskfn]
deps = getsetscenedeps(revdep)
provides = self.rqdata.dataCaches[mc].fn_provides[taskfn]
taskhash = self.rqdata.runtaskentries[revdep].hash
taskdepdata[revdep] = [pn, taskname, fn, deps, provides, taskhash]
for revdep2 in deps:
if revdep2 not in taskdepdata:
additional.append(revdep2)
next = additional
#bb.note("Task %s: " % task + str(taskdepdata).replace("], ", "],\n"))
return taskdepdata
class TaskFailure(Exception):
"""
Exception raised when a task in a runqueue fails
"""
def __init__(self, x):
self.args = x
class runQueueExitWait(bb.event.Event):
"""
Event when waiting for task processes to exit
"""
def __init__(self, remain):
self.remain = remain
self.message = "Waiting for %s active tasks to finish" % remain
bb.event.Event.__init__(self)
class runQueueEvent(bb.event.Event):
"""
Base runQueue event class
"""
def __init__(self, task, stats, rq):
self.taskid = task
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.taskstring = task
self.taskname = taskname_from_tid(task)
self.taskfile = fn_from_tid(task)
self.taskhash = rq.rqdata.get_task_hash(task)
self.stats = stats.copy()
bb.event.Event.__init__(self)
class sceneQueueEvent(runQueueEvent):
"""
Base sceneQueue event class
"""
def __init__(self, task, stats, rq, noexec=False):
runQueueEvent.__init__(self, task, stats, rq)
bitbake: taskdata/runqueue: Rewrite without use of ID indirection I'm not sure what possesed me when I wrote this code originally but its indirection of everyting to use numeric IDs and position dependent lists is horrific. Given the way python internals work, its completely and utterly pointless from performance perspective. It also makes the code hard to understand and debug since any numeric ID has to be translated into something human readable. The hard part is that the IDs are infectous and spread from taskdata into runqueue and even partly into cooker for the dependency graph processing. The only real way to deal with this is to convert everything to use a more sane data structure. This patch: * Uses "<fn>:<taskname>" as the ID for tasks rather than a number * Changes to dict() based structures rather than position dependent lists * Drops the build name, runtime name and filename ID indexes On the most part there shouldn't be user visible changes. Sadly we did leak datastructures to the setscene verify function which has to be rewritten. To handle this, the variable name used to specifiy the version changes from BB_SETSCENE_VERIFY_FUNCTION to BB_SETSCENE_VERIFY_FUNCTION2 allowing multiple versions of bitbake to work with suitably written metadata. Anyone with custom schedulers may also need to change them. I believe the benefits in code readability and easier debugging far outweigh those issues though. It also means we have a saner codebase to add multiconfig support on top of. During development, I did have some of the original code coexisting with the new data stores to allow comparision of the data and check it was working correcty, particuarly for taskdata. I have also compared task-depends.dot files before and after the change. There should be no functionality changes in this patch, its purely a data structure change and that is visible in the patch. (Bitbake rev: 2c88afb60da54e58f555411a7bd7b006b0c29306) Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
2016-06-12 22:55:48 +00:00
self.taskstring = task + "_setscene"
self.taskname = taskname_from_tid(task) + "_setscene"
self.taskfile = fn_from_tid(task)
self.taskhash = rq.rqdata.get_task_hash(task)
class runQueueTaskStarted(runQueueEvent):
"""
Event notifying a task was started
"""
def __init__(self, task, stats, rq, noexec=False):
runQueueEvent.__init__(self, task, stats, rq)
self.noexec = noexec
class sceneQueueTaskStarted(sceneQueueEvent):
"""
Event notifying a setscene task was started
"""
def __init__(self, task, stats, rq, noexec=False):
sceneQueueEvent.__init__(self, task, stats, rq)
self.noexec = noexec
class runQueueTaskFailed(runQueueEvent):
"""
Event notifying a task failed
"""
def __init__(self, task, stats, exitcode, rq):
runQueueEvent.__init__(self, task, stats, rq)
self.exitcode = exitcode
class sceneQueueTaskFailed(sceneQueueEvent):
"""
Event notifying a setscene task failed
"""
def __init__(self, task, stats, exitcode, rq):
sceneQueueEvent.__init__(self, task, stats, rq)
self.exitcode = exitcode
class sceneQueueComplete(sceneQueueEvent):
"""
Event when all the sceneQueue tasks are complete
"""
def __init__(self, stats, rq):
self.stats = stats.copy()
bb.event.Event.__init__(self)
class runQueueTaskCompleted(runQueueEvent):
"""
Event notifying a task completed
"""
class sceneQueueTaskCompleted(sceneQueueEvent):
"""
Event notifying a setscene task completed
"""
class runQueueTaskSkipped(runQueueEvent):
"""
Event notifying a task was skipped
"""
def __init__(self, task, stats, rq, reason):
runQueueEvent.__init__(self, task, stats, rq)
self.reason = reason
class runQueuePipe():
"""
Abstraction for a pipe between a worker thread and the server
"""
def __init__(self, pipein, pipeout, d, rq, rqexec):
self.input = pipein
if pipeout:
pipeout.close()
bb.utils.nonblockingfd(self.input)
self.queue = b""
self.d = d
self.rq = rq
self.rqexec = rqexec
def setrunqueueexec(self, rqexec):
self.rqexec = rqexec
def read(self):
for workers, name in [(self.rq.worker, "Worker"), (self.rq.fakeworker, "Fakeroot")]:
for worker in workers.values():
worker.process.poll()
if worker.process.returncode is not None and not self.rq.teardown:
bb.error("%s process (%s) exited unexpectedly (%s), shutting down..." % (name, worker.process.pid, str(worker.process.returncode)))
self.rq.finish_runqueue(True)
start = len(self.queue)
try:
self.queue = self.queue + (self.input.read(102400) or b"")
except (OSError, IOError) as e:
if e.errno != errno.EAGAIN:
raise
end = len(self.queue)
found = True
while found and len(self.queue):
found = False
index = self.queue.find(b"</event>")
while index != -1 and self.queue.startswith(b"<event>"):
try:
event = pickle.loads(self.queue[7:index])
except ValueError as e:
bb.msg.fatal("RunQueue", "failed load pickle '%s': '%s'" % (e, self.queue[7:index]))
bb.event.fire_from_worker(event, self.d)
found = True
self.queue = self.queue[index+8:]
index = self.queue.find(b"</event>")
index = self.queue.find(b"</exitcode>")
while index != -1 and self.queue.startswith(b"<exitcode>"):
try:
task, status = pickle.loads(self.queue[10:index])
except ValueError as e:
bb.msg.fatal("RunQueue", "failed load pickle '%s': '%s'" % (e, self.queue[10:index]))
self.rqexec.runqueue_process_waitpid(task, status)
found = True
self.queue = self.queue[index+11:]
index = self.queue.find(b"</exitcode>")
return (end > start)
def close(self):
while self.read():
continue
if len(self.queue) > 0:
print("Warning, worker left partial message: %s" % self.queue)
self.input.close()
def get_setscene_enforce_whitelist(d):
if d.getVar('BB_SETSCENE_ENFORCE') != '1':
return None
whitelist = (d.getVar("BB_SETSCENE_ENFORCE_WHITELIST") or "").split()
outlist = []
for item in whitelist[:]:
if item.startswith('%:'):
for target in sys.argv[1:]:
if not target.startswith('-'):
outlist.append(target.split(':')[0] + ':' + item.split(':')[1])
else:
outlist.append(item)
return outlist
def check_setscene_enforce_whitelist(pn, taskname, whitelist):
import fnmatch
if whitelist:
item = '%s:%s' % (pn, taskname)
for whitelist_item in whitelist:
if fnmatch.fnmatch(item, whitelist_item):
return True
return False
return True