|
|
|
#!/usr/bin/env python3
|
|
|
|
# Copyright (c) 2017 The Chromium Authors. All rights reserved.
|
|
|
|
# Use of this source code is governed by a BSD-style license that can be
|
|
|
|
# found in the LICENSE file.
|
|
|
|
"""
|
|
|
|
This script (intended to be invoked by autoninja or autoninja.bat) detects
|
|
|
|
whether a build is accelerated using a service like goma. If so, it runs with a
|
|
|
|
large -j value, and otherwise it chooses a small one. This auto-adjustment
|
|
|
|
makes using remote build acceleration simpler and safer, and avoids errors that
|
|
|
|
can cause slow goma builds or swap-storms on unaccelerated builds.
|
|
|
|
|
|
|
|
autoninja tries to detect relevant build settings such as use_remoteexec, and it
|
|
|
|
does handle import statements, but it can't handle conditional setting of build
|
|
|
|
settings.
|
|
|
|
"""
|
|
|
|
|
Reland "Check whether goma is running when it is enabled"
This reverts commit 428143ee24c5f084c8dfb38cd17f07b4f7ba9bf7.
Reason for revert: Fixing the issues revealed by the original change by
avoiding python3 and by checking for the existence of gomacc[.exe]
before running it.
This also relands 2241db8a1f6c05325c8a0c9eb5426d96cd6fa984 - "Avoid
capture_output to support Python 3.6", to simplify relanding and any
possible reverts.
Original change's description:
> Revert "Check whether goma is running when it is enabled"
>
> This reverts commit b7ddc5a0091bcd4d070fcd91027d7099338e84b9.
>
> Reason for revert:
> This broke the builder where depot_tools is not in PATH.
> https://logs.chromium.org/logs/infra-internal/buildbucket/cr-buildbucket.appspot.com/8858077852309878080/+/u/build/stdout
>
> Original change's description:
> > Check whether goma is running when it is enabled
> >
> > One of the mistakes one can make when running ninja is having goma
> > enabled (use_goma=true in args.gn) but not having goma running. This can
> > lead to ~1,000 failed compile steps, which is messy.
> >
> > This change teaches autoninja.py to check whether goma is running. If
> > not then it tells autoninja to just print a warning message. The
> > check costs roughly 30 ms which seems reasonable.
> >
> > In fact, because this change also switches away from vpython (necessary
> > to use python3 to use subprocess.run) it actually runs about 600 ms
> > _faster_ than before this change.
> >
> > If build acceleration is requested through use_rbe then no checking for
> > whether the service is running is done. That could be added in the
> > future.
> >
> > autoninja.py could auto-start goma but that is error prone and has
> > limited additional value.
> >
> > This was tested on Linux, OSX, and Windows.
> >
> > Bug: 868590, b/174673874
> > Change-Id: Ie773e574878471e5136b9b82d52f86af3d848318
> > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2627014
> > Commit-Queue: Bruce Dawson <brucedawson@chromium.org>
> > Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
>
> TBR=yyanagisawa@google.com,dpranke@google.com,brucedawson@chromium.org,sanfin@chromium.org,infra-scoped@luci-project-accounts.iam.gserviceaccount.com
>
> Change-Id: I57a6c73ea853259f3d1ec7ad0ce51e495acc96db
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: 868590
> Bug: b/174673874
> Bug: 1167064
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2632018
> Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
> Commit-Queue: Yoshisato Yanagisawa <yyanagisawa@google.com>
TBR=yyanagisawa@google.com,dpranke@google.com,brucedawson@chromium.org,sanfin@chromium.org,infra-scoped@luci-project-accounts.iam.gserviceaccount.com
# Not skipping CQ checks because this is a reland.
Bug: 868590
Bug: b/174673874
Bug: 1167064
Change-Id: I8aa6830259bc18f8e7926cd0bf5c62e671c74a2d
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2634201
Reviewed-by: Bruce Dawson <brucedawson@chromium.org>
Reviewed-by: Dirk Pranke <dpranke@google.com>
Reviewed-by: Fumitoshi Ukai <ukai@google.com>
Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
Commit-Queue: Bruce Dawson <brucedawson@chromium.org>
4 years ago
|
|
|
import multiprocessing
|
|
|
|
import os
|
|
|
|
import platform
|
|
|
|
import re
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
import shlex
|
Reland "Check whether goma is running when it is enabled"
This reverts commit 428143ee24c5f084c8dfb38cd17f07b4f7ba9bf7.
Reason for revert: Fixing the issues revealed by the original change by
avoiding python3 and by checking for the existence of gomacc[.exe]
before running it.
This also relands 2241db8a1f6c05325c8a0c9eb5426d96cd6fa984 - "Avoid
capture_output to support Python 3.6", to simplify relanding and any
possible reverts.
Original change's description:
> Revert "Check whether goma is running when it is enabled"
>
> This reverts commit b7ddc5a0091bcd4d070fcd91027d7099338e84b9.
>
> Reason for revert:
> This broke the builder where depot_tools is not in PATH.
> https://logs.chromium.org/logs/infra-internal/buildbucket/cr-buildbucket.appspot.com/8858077852309878080/+/u/build/stdout
>
> Original change's description:
> > Check whether goma is running when it is enabled
> >
> > One of the mistakes one can make when running ninja is having goma
> > enabled (use_goma=true in args.gn) but not having goma running. This can
> > lead to ~1,000 failed compile steps, which is messy.
> >
> > This change teaches autoninja.py to check whether goma is running. If
> > not then it tells autoninja to just print a warning message. The
> > check costs roughly 30 ms which seems reasonable.
> >
> > In fact, because this change also switches away from vpython (necessary
> > to use python3 to use subprocess.run) it actually runs about 600 ms
> > _faster_ than before this change.
> >
> > If build acceleration is requested through use_rbe then no checking for
> > whether the service is running is done. That could be added in the
> > future.
> >
> > autoninja.py could auto-start goma but that is error prone and has
> > limited additional value.
> >
> > This was tested on Linux, OSX, and Windows.
> >
> > Bug: 868590, b/174673874
> > Change-Id: Ie773e574878471e5136b9b82d52f86af3d848318
> > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2627014
> > Commit-Queue: Bruce Dawson <brucedawson@chromium.org>
> > Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
>
> TBR=yyanagisawa@google.com,dpranke@google.com,brucedawson@chromium.org,sanfin@chromium.org,infra-scoped@luci-project-accounts.iam.gserviceaccount.com
>
> Change-Id: I57a6c73ea853259f3d1ec7ad0ce51e495acc96db
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: 868590
> Bug: b/174673874
> Bug: 1167064
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2632018
> Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
> Commit-Queue: Yoshisato Yanagisawa <yyanagisawa@google.com>
TBR=yyanagisawa@google.com,dpranke@google.com,brucedawson@chromium.org,sanfin@chromium.org,infra-scoped@luci-project-accounts.iam.gserviceaccount.com
# Not skipping CQ checks because this is a reland.
Bug: 868590
Bug: b/174673874
Bug: 1167064
Change-Id: I8aa6830259bc18f8e7926cd0bf5c62e671c74a2d
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2634201
Reviewed-by: Bruce Dawson <brucedawson@chromium.org>
Reviewed-by: Dirk Pranke <dpranke@google.com>
Reviewed-by: Fumitoshi Ukai <ukai@google.com>
Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@google.com>
Commit-Queue: Bruce Dawson <brucedawson@chromium.org>
4 years ago
|
|
|
import subprocess
|
|
|
|
import sys
|
|
|
|
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
import autosiso
|
|
|
|
import ninja
|
|
|
|
import ninja_reclient
|
|
|
|
import siso
|
|
|
|
|
|
|
|
if sys.platform in ['darwin', 'linux']:
|
|
|
|
import resource
|
|
|
|
|
|
|
|
SCRIPT_DIR = os.path.dirname(os.path.realpath(__file__))
|
|
|
|
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
# See [1] and [2] for the painful details of this next section, which handles
|
|
|
|
# escaping command lines so that they can be copied and pasted into a cmd
|
|
|
|
# window.
|
|
|
|
#
|
|
|
|
# pylint: disable=line-too-long
|
|
|
|
# [1] https://learn.microsoft.com/en-us/archive/blogs/twistylittlepassagesallalike/everyone-quotes-command-line-arguments-the-wrong-way # noqa
|
|
|
|
# [2] https://web.archive.org/web/20150815000000*/https://www.microsoft.com/resources/documentation/windows/xp/all/proddocs/en-us/set.mspx # noqa
|
|
|
|
_UNSAFE_FOR_CMD = set('^<>&|()%')
|
|
|
|
_ALL_META_CHARS = _UNSAFE_FOR_CMD.union(set('"'))
|
|
|
|
|
|
|
|
|
|
|
|
def _quote_for_cmd(arg):
|
|
|
|
# First, escape the arg so that CommandLineToArgvW will parse it properly.
|
|
|
|
if arg == '' or ' ' in arg or '"' in arg:
|
|
|
|
quote_re = re.compile(r'(\\*)"')
|
|
|
|
arg = '"%s"' % (quote_re.sub(lambda mo: 2 * mo.group(1) + '\\"', arg))
|
|
|
|
|
|
|
|
# Then check to see if the arg contains any metacharacters other than
|
|
|
|
# double quotes; if it does, quote everything (including the double
|
|
|
|
# quotes) for safety.
|
|
|
|
if any(a in _UNSAFE_FOR_CMD for a in arg):
|
|
|
|
arg = ''.join('^' + a if a in _ALL_META_CHARS else a for a in arg)
|
|
|
|
return arg
|
|
|
|
|
|
|
|
|
|
|
|
def _print_cmd(cmd):
|
|
|
|
shell_quoter = shlex.quote
|
|
|
|
if sys.platform.startswith('win'):
|
|
|
|
shell_quoter = _quote_for_cmd
|
|
|
|
print(*[shell_quoter(arg) for arg in cmd], file=sys.stderr)
|
|
|
|
|
|
|
|
|
|
|
|
def _gn_lines(output_dir, path):
|
|
|
|
"""
|
|
|
|
Generator function that returns args.gn lines one at a time, following
|
|
|
|
import directives as needed.
|
|
|
|
"""
|
|
|
|
import_re = re.compile(r'\s*import\("(.*)"\)')
|
|
|
|
with open(path, encoding='utf-8') as f:
|
|
|
|
for line in f:
|
|
|
|
match = import_re.match(line)
|
|
|
|
if match:
|
|
|
|
raw_import_path = match.groups()[0]
|
|
|
|
if raw_import_path[:2] == "//":
|
|
|
|
import_path = os.path.normpath(
|
|
|
|
os.path.join(output_dir, '..', '..',
|
|
|
|
raw_import_path[2:]))
|
|
|
|
else:
|
|
|
|
import_path = os.path.normpath(
|
|
|
|
os.path.join(os.path.dirname(path), raw_import_path))
|
|
|
|
for import_line in _gn_lines(output_dir, import_path):
|
|
|
|
yield import_line
|
|
|
|
else:
|
|
|
|
yield line
|
|
|
|
|
|
|
|
|
|
|
|
def main(args):
|
|
|
|
# The -t tools are incompatible with -j
|
|
|
|
t_specified = False
|
|
|
|
j_specified = False
|
|
|
|
offline = False
|
|
|
|
output_dir = '.'
|
|
|
|
input_args = args
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
summarize_build = os.environ.get('NINJA_SUMMARIZE_BUILD') == '1'
|
|
|
|
# On Windows the autoninja.bat script passes along the arguments enclosed in
|
|
|
|
# double quotes. This prevents multiple levels of parsing of the special '^'
|
|
|
|
# characters needed when compiling a single file but means that this script
|
|
|
|
# gets called with a single argument containing all of the actual arguments,
|
|
|
|
# separated by spaces. When this case is detected we need to do argument
|
|
|
|
# splitting ourselves. This means that arguments containing actual spaces
|
|
|
|
# are not supported by autoninja, but that is not a real limitation.
|
|
|
|
if (sys.platform.startswith('win') and len(args) == 2
|
|
|
|
and input_args[1].count(' ') > 0):
|
|
|
|
input_args = args[:1] + args[1].split()
|
|
|
|
|
|
|
|
# Ninja uses getopt_long, which allow to intermix non-option arguments.
|
|
|
|
# To leave non supported parameters untouched, we do not use getopt.
|
|
|
|
for index, arg in enumerate(input_args[1:]):
|
|
|
|
if arg.startswith('-j'):
|
|
|
|
j_specified = True
|
|
|
|
if arg.startswith('-t'):
|
|
|
|
t_specified = True
|
|
|
|
if arg == '-C':
|
|
|
|
# + 1 to get the next argument and +1 because we trimmed off
|
|
|
|
# input_args[0]
|
|
|
|
output_dir = input_args[index + 2]
|
|
|
|
elif arg.startswith('-C'):
|
|
|
|
# Support -Cout/Default
|
|
|
|
output_dir = arg[2:]
|
|
|
|
elif arg in ('-o', '--offline'):
|
|
|
|
offline = True
|
|
|
|
elif arg == '-h':
|
|
|
|
print('autoninja: Use -o/--offline to temporary disable goma.',
|
|
|
|
file=sys.stderr)
|
|
|
|
print(file=sys.stderr)
|
|
|
|
|
|
|
|
use_goma = False
|
|
|
|
use_remoteexec = False
|
|
|
|
use_rbe = False
|
|
|
|
use_siso = False
|
|
|
|
|
|
|
|
# Attempt to auto-detect remote build acceleration. We support gn-based
|
|
|
|
# builds, where we look for args.gn in the build tree, and cmake-based
|
|
|
|
# builds where we look for rules.ninja.
|
|
|
|
if os.path.exists(os.path.join(output_dir, 'args.gn')):
|
|
|
|
for line in _gn_lines(output_dir, os.path.join(output_dir, 'args.gn')):
|
|
|
|
# use_goma, use_remoteexec, or use_rbe will activate build
|
|
|
|
# acceleration.
|
|
|
|
#
|
|
|
|
# This test can match multi-argument lines. Examples of this
|
|
|
|
# are: is_debug=false use_goma=true is_official_build=false
|
|
|
|
# use_goma=false# use_goma=true This comment is ignored
|
|
|
|
#
|
|
|
|
# Anything after a comment is not consider a valid argument.
|
|
|
|
line_without_comment = line.split('#')[0]
|
|
|
|
if re.search(r'(^|\s)(use_goma)\s*=\s*true($|\s)',
|
|
|
|
line_without_comment):
|
|
|
|
use_goma = True
|
|
|
|
continue
|
|
|
|
if re.search(r'(^|\s)(use_remoteexec)\s*=\s*true($|\s)',
|
|
|
|
line_without_comment):
|
|
|
|
use_remoteexec = True
|
|
|
|
continue
|
|
|
|
if re.search(r'(^|\s)(use_rbe)\s*=\s*true($|\s)',
|
|
|
|
line_without_comment):
|
|
|
|
use_rbe = True
|
|
|
|
continue
|
|
|
|
if re.search(r'(^|\s)(use_siso)\s*=\s*true($|\s)',
|
|
|
|
line_without_comment):
|
|
|
|
use_siso = True
|
|
|
|
continue
|
|
|
|
|
|
|
|
siso_marker = os.path.join(output_dir, '.siso_deps')
|
|
|
|
if use_siso:
|
|
|
|
# autosiso generates a .ninja_log file so the mere existence of a
|
|
|
|
# .ninja_log file doesn't imply that a ninja build was done. However
|
|
|
|
# if there is a .ninja_log but no .siso_deps then that implies a
|
|
|
|
# ninja build.
|
|
|
|
ninja_marker = os.path.join(output_dir, '.ninja_log')
|
|
|
|
if os.path.exists(ninja_marker) and not os.path.exists(siso_marker):
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
print('Run gn clean before switching from ninja to siso in %s' %
|
|
|
|
output_dir,
|
|
|
|
file=sys.stderr)
|
|
|
|
return 1
|
|
|
|
if use_goma:
|
|
|
|
print('Siso does not support Goma.', file=sys.stderr)
|
|
|
|
print('Do not use use_siso=true and use_goma=true',
|
|
|
|
file=sys.stderr)
|
|
|
|
return 1
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
if use_remoteexec:
|
|
|
|
return autosiso.main(['autosiso'] + input_args[1:])
|
|
|
|
return siso.main(['siso', 'ninja'] + input_args[1:])
|
|
|
|
|
|
|
|
if os.path.exists(siso_marker):
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
print('Run gn clean before switching from siso to ninja in %s' %
|
|
|
|
output_dir,
|
|
|
|
file=sys.stderr)
|
|
|
|
return 1
|
|
|
|
|
|
|
|
else:
|
|
|
|
for relative_path in [
|
|
|
|
'', # GN keeps them in the root of output_dir
|
|
|
|
'CMakeFiles'
|
|
|
|
]:
|
|
|
|
path = os.path.join(output_dir, relative_path, 'rules.ninja')
|
|
|
|
if os.path.exists(path):
|
|
|
|
with open(path, encoding='utf-8') as file_handle:
|
|
|
|
for line in file_handle:
|
|
|
|
if re.match(r'^\s*command\s*=\s*\S+gomacc', line):
|
|
|
|
use_goma = True
|
|
|
|
break
|
|
|
|
|
|
|
|
# Strip -o/--offline so ninja doesn't see them.
|
|
|
|
input_args = [arg for arg in input_args if arg not in ('-o', '--offline')]
|
|
|
|
|
|
|
|
# If GOMA_DISABLED is set to "true", "t", "yes", "y", or "1"
|
|
|
|
# (case-insensitive) then gomacc will use the local compiler instead of
|
|
|
|
# doing a goma compile. This is convenient if you want to briefly disable
|
|
|
|
# goma. It avoids having to rebuild the world when transitioning between
|
|
|
|
# goma/non-goma builds. However, it is not as fast as doing a "normal"
|
|
|
|
# non-goma build because an extra process is created for each compile step.
|
|
|
|
# Checking this environment variable ensures that autoninja uses an
|
|
|
|
# appropriate -j value in this situation.
|
|
|
|
goma_disabled_env = os.environ.get('GOMA_DISABLED', '0').lower()
|
|
|
|
if offline or goma_disabled_env in ['true', 't', 'yes', 'y', '1']:
|
|
|
|
use_goma = False
|
|
|
|
|
|
|
|
if use_goma:
|
|
|
|
gomacc_file = 'gomacc.exe' if sys.platform.startswith(
|
|
|
|
'win') else 'gomacc'
|
|
|
|
goma_dir = os.environ.get('GOMA_DIR',
|
|
|
|
os.path.join(SCRIPT_DIR, '.cipd_bin'))
|
|
|
|
gomacc_path = os.path.join(goma_dir, gomacc_file)
|
|
|
|
# Don't invoke gomacc if it doesn't exist.
|
|
|
|
if os.path.exists(gomacc_path):
|
|
|
|
# Check to make sure that goma is running. If not, don't start the
|
|
|
|
# build.
|
|
|
|
status = subprocess.call([gomacc_path, 'port'],
|
|
|
|
stdout=subprocess.PIPE,
|
|
|
|
stderr=subprocess.PIPE,
|
|
|
|
shell=False)
|
|
|
|
if status == 1:
|
|
|
|
print(
|
|
|
|
'Goma is not running. Use "goma_ctl ensure_start" to start '
|
|
|
|
'it.',
|
|
|
|
file=sys.stderr)
|
|
|
|
if sys.platform.startswith('win'):
|
|
|
|
# Set an exit code of 1 in the batch file.
|
|
|
|
print('cmd "/c exit 1"')
|
|
|
|
else:
|
|
|
|
# Set an exit code of 1 by executing 'false' in the bash
|
|
|
|
# script.
|
|
|
|
print('false')
|
|
|
|
sys.exit(1)
|
|
|
|
|
|
|
|
# A large build (with or without goma) tends to hog all system resources.
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
# Depending on the operating system, we might have mechanisms available
|
|
|
|
# to run at a lower priority, which improves this situation.
|
|
|
|
if os.environ.get('NINJA_BUILD_IN_BACKGROUND') == '1':
|
|
|
|
if sys.platform in ['darwin', 'linux']:
|
|
|
|
# nice-level 10 is usually considered a good default for background
|
|
|
|
# tasks. The niceness is inherited by child processes, so we can
|
|
|
|
# just set it here for us and it'll apply to the build tool we
|
|
|
|
# spawn later.
|
|
|
|
os.nice(10)
|
|
|
|
|
|
|
|
# Tell goma or reclient to do local compiles.
|
|
|
|
if offline:
|
|
|
|
os.environ['RBE_remote_disabled'] = '1'
|
|
|
|
os.environ['GOMA_DISABLED'] = '1'
|
|
|
|
|
|
|
|
# On macOS and most Linux distributions, the default limit of open file
|
|
|
|
# descriptors is too low (256 and 1024, respectively).
|
|
|
|
# This causes a large j value to result in 'Too many open files' errors.
|
|
|
|
# Check whether the limit can be raised to a large enough value. If yes,
|
|
|
|
# use `ulimit -n .... &&` as a prefix to increase the limit when running
|
|
|
|
# ninja.
|
|
|
|
if sys.platform in ['darwin', 'linux']:
|
|
|
|
# Increase the number of allowed open file descriptors to the maximum.
|
|
|
|
fileno_limit, hard_limit = resource.getrlimit(resource.RLIMIT_NOFILE)
|
|
|
|
if fileno_limit < hard_limit:
|
|
|
|
try:
|
|
|
|
resource.setrlimit(resource.RLIMIT_NOFILE,
|
|
|
|
(hard_limit, hard_limit))
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
except Exception:
|
|
|
|
pass
|
|
|
|
fileno_limit, hard_limit = resource.getrlimit(
|
|
|
|
resource.RLIMIT_NOFILE)
|
|
|
|
|
|
|
|
# Call ninja.py so that it can find ninja binary installed by DEPS or one in
|
|
|
|
# PATH.
|
|
|
|
ninja_path = os.path.join(SCRIPT_DIR, 'ninja.py')
|
|
|
|
# If using remoteexec, use ninja_reclient.py which wraps ninja.py with
|
|
|
|
# starting and stopping reproxy.
|
|
|
|
if use_remoteexec:
|
|
|
|
ninja_path = os.path.join(SCRIPT_DIR, 'ninja_reclient.py')
|
|
|
|
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
args = [sys.executable, ninja_path] + input_args[1:]
|
|
|
|
|
|
|
|
num_cores = multiprocessing.cpu_count()
|
|
|
|
if not j_specified and not t_specified:
|
|
|
|
if not offline and (use_goma or use_remoteexec or use_rbe):
|
|
|
|
args.append('-j')
|
|
|
|
default_core_multiplier = 80
|
|
|
|
if platform.machine() in ('x86_64', 'AMD64'):
|
|
|
|
# Assume simultaneous multithreading and therefore half as many
|
|
|
|
# cores as logical processors.
|
|
|
|
num_cores //= 2
|
|
|
|
|
|
|
|
core_multiplier = int(
|
|
|
|
os.environ.get('NINJA_CORE_MULTIPLIER',
|
|
|
|
default_core_multiplier))
|
|
|
|
j_value = num_cores * core_multiplier
|
|
|
|
|
|
|
|
core_limit = int(os.environ.get('NINJA_CORE_LIMIT', j_value))
|
|
|
|
j_value = min(j_value, core_limit)
|
|
|
|
|
|
|
|
# On Windows, a -j higher than 1000 doesn't improve build times.
|
|
|
|
# On macOS, ninja is limited to at most FD_SETSIZE (1024) open file
|
|
|
|
# descriptors.
|
|
|
|
if sys.platform in ['darwin', 'win32']:
|
|
|
|
j_value = min(j_value, 1000)
|
|
|
|
|
|
|
|
# Use a j value that reliably works with the open file descriptors
|
|
|
|
# limit.
|
|
|
|
if sys.platform in ['darwin', 'linux']:
|
|
|
|
j_value = min(j_value, int(fileno_limit * 0.8))
|
|
|
|
|
|
|
|
args.append('%d' % j_value)
|
|
|
|
else:
|
|
|
|
j_value = num_cores
|
|
|
|
# Ninja defaults to |num_cores + 2|
|
|
|
|
j_value += int(os.environ.get('NINJA_CORE_ADDITION', '2'))
|
|
|
|
args.append('-j')
|
|
|
|
args.append('%d' % j_value)
|
|
|
|
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
if summarize_build:
|
|
|
|
# Enable statistics collection in Ninja.
|
|
|
|
args += ['-d', 'stats']
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
# Print the command-line to reassure the user that the right settings
|
|
|
|
# are being used.
|
|
|
|
_print_cmd(args)
|
|
|
|
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
if use_remoteexec:
|
|
|
|
return ninja_reclient.main(args[1:])
|
|
|
|
return ninja.main(args[1:])
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
Execute Ninja / Siso directly from autoninja.py.
Instead of printing a command-line, we just directly call into the respective main functions from Python. This saves spawning another interpreter and prevents things that can go wrong from having to quote, unquote, split and tunnel arguments through shells.
Part of my bigger auto{ninja,siso} refactoring.
Tested:
- Handling of the ^^ suffix on Windows still works correctly.
- Handling of error codes - i.e.; making sure
"autoninja base_unittests && base_unittests.exe" behaves properly
in the success/failure case.
- Make sure the command prompt title is reliably reset on exit.
I tested autoninja with all combinations of these:
- Host platform: Linux, macOS, Windows
- Remote GN args: <none>, use_goma=true, use_remoteexec=true
- Siso GN args: <none>, use_siso=true
- Targets: base, ../../base/types/expected_macros_unittest.cc^ (on Linux) and ../../base/types/expected_macros_unittest.cc^^ (on Windows)
R=brucedawson@chromium.org, jwata@google.com, tikuta@chromium.org
Bug: b/293657720
Change-Id: I275a775fdc5abb6555f79d4beab76cd0914d4bd6
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/4924185
Reviewed-by: Takuto Ikuta <tikuta@chromium.org>
Reviewed-by: Junji Watanabe <jwata@google.com>
Commit-Queue: Philipp Wollermann <philwo@chromium.org>
2 years ago
|
|
|
try:
|
|
|
|
sys.exit(main(sys.argv))
|
|
|
|
except KeyboardInterrupt:
|
|
|
|
sys.exit(1)
|