зеркало из https://github.com/mozilla/gecko-dev.git
1470 строки
60 KiB
Python
1470 строки
60 KiB
Python
# This Source Code Form is subject to the terms of the Mozilla Public
|
|
# License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
|
|
|
from __future__ import absolute_import, print_function, unicode_literals
|
|
|
|
import argparse
|
|
import errno
|
|
import json
|
|
import logging
|
|
import os
|
|
import sys
|
|
import subprocess
|
|
|
|
from mach.decorators import (
|
|
CommandArgument,
|
|
CommandProvider,
|
|
Command,
|
|
SettingsProvider,
|
|
SubCommand,
|
|
)
|
|
|
|
from mozbuild.base import (
|
|
BuildEnvironmentNotFoundException,
|
|
MachCommandBase,
|
|
MachCommandConditions as conditions,
|
|
)
|
|
|
|
UNKNOWN_TEST = '''
|
|
I was unable to find tests from the given argument(s).
|
|
|
|
You should specify a test directory, filename, test suite name, or
|
|
abbreviation. If no arguments are given, there must be local file
|
|
changes and corresponding IMPACTED_TESTS annotations in moz.build
|
|
files relevant to those files.
|
|
|
|
It's possible my little brain doesn't know about the type of test you are
|
|
trying to execute. If you suspect this, please request support by filing
|
|
a bug at
|
|
https://bugzilla.mozilla.org/enter_bug.cgi?product=Testing&component=General.
|
|
'''.strip()
|
|
|
|
UNKNOWN_FLAVOR = '''
|
|
I know you are trying to run a %s%s test. Unfortunately, I can't run those
|
|
tests yet. Sorry!
|
|
'''.strip()
|
|
|
|
TEST_HELP = '''
|
|
Test or tests to run. Tests can be specified by filename, directory, suite
|
|
name or suite alias.
|
|
|
|
The following test suites and aliases are supported: {}
|
|
'''.strip()
|
|
|
|
|
|
@SettingsProvider
|
|
class TestConfig(object):
|
|
|
|
@classmethod
|
|
def config_settings(cls):
|
|
from mozlog.commandline import log_formatters
|
|
from mozlog.structuredlog import log_levels
|
|
format_desc = "The default format to use when running tests with `mach test`."
|
|
format_choices = log_formatters.keys()
|
|
level_desc = "The default log level to use when running tests with `mach test`."
|
|
level_choices = [l.lower() for l in log_levels]
|
|
return [
|
|
('test.format', 'string', format_desc, 'mach', {'choices': format_choices}),
|
|
('test.level', 'string', level_desc, 'info', {'choices': level_choices}),
|
|
]
|
|
|
|
|
|
def get_test_parser():
|
|
from mozlog.commandline import add_logging_group
|
|
from moztest.resolve import TEST_SUITES
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument('what', default=None, nargs='+',
|
|
help=TEST_HELP.format(', '.join(sorted(TEST_SUITES))))
|
|
parser.add_argument('extra_args', default=None, nargs=argparse.REMAINDER,
|
|
help="Extra arguments to pass to the underlying test command(s). "
|
|
"If an underlying command doesn't recognize the argument, it "
|
|
"will fail.")
|
|
parser.add_argument('--debugger', default=None, action='store',
|
|
nargs='?', help="Specify a debugger to use.")
|
|
add_logging_group(parser)
|
|
return parser
|
|
|
|
|
|
ADD_TEST_SUPPORTED_SUITES = ['mochitest-chrome', 'mochitest-plain', 'mochitest-browser-chrome',
|
|
'web-platform-tests-testharness', 'web-platform-tests-reftest',
|
|
'xpcshell']
|
|
ADD_TEST_SUPPORTED_DOCS = ['js', 'html', 'xhtml', 'xul']
|
|
|
|
SUITE_SYNONYMS = {
|
|
"wpt": "web-platform-tests-testharness",
|
|
"wpt-testharness": "web-platform-tests-testharness",
|
|
"wpt-reftest": "web-platform-tests-reftest"
|
|
}
|
|
|
|
MISSING_ARG = object()
|
|
|
|
|
|
def create_parser_addtest():
|
|
import addtest
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument('--suite',
|
|
choices=sorted(ADD_TEST_SUPPORTED_SUITES + SUITE_SYNONYMS.keys()),
|
|
help='suite for the test. '
|
|
'If you pass a `test` argument this will be determined '
|
|
'based on the filename and the folder it is in')
|
|
parser.add_argument('-o', '--overwrite',
|
|
action='store_true',
|
|
help='Overwrite an existing file if it exists.')
|
|
parser.add_argument('--doc',
|
|
choices=ADD_TEST_SUPPORTED_DOCS,
|
|
help='Document type for the test (if applicable).'
|
|
'If you pass a `test` argument this will be determined '
|
|
'based on the filename.')
|
|
parser.add_argument("-e", "--editor", action="store", nargs="?",
|
|
default=MISSING_ARG, help="Open the created file(s) in an editor; if a "
|
|
"binary is supplied it will be used otherwise the default editor for "
|
|
"your environment will be opened")
|
|
|
|
for base_suite in addtest.TEST_CREATORS:
|
|
cls = addtest.TEST_CREATORS[base_suite]
|
|
if hasattr(cls, "get_parser"):
|
|
group = parser.add_argument_group(base_suite)
|
|
cls.get_parser(group)
|
|
|
|
parser.add_argument('test',
|
|
nargs='?',
|
|
help=('Test to create.'))
|
|
return parser
|
|
|
|
|
|
@CommandProvider
|
|
class AddTest(MachCommandBase):
|
|
@Command('addtest', category='testing',
|
|
description='Generate tests based on templates',
|
|
parser=create_parser_addtest)
|
|
def addtest(self, suite=None, test=None, doc=None, overwrite=False,
|
|
editor=MISSING_ARG, **kwargs):
|
|
import addtest
|
|
from moztest.resolve import TEST_SUITES
|
|
|
|
if not suite and not test:
|
|
return create_parser_addtest().parse_args(["--help"])
|
|
|
|
if suite in SUITE_SYNONYMS:
|
|
suite = SUITE_SYNONYMS[suite]
|
|
|
|
if test:
|
|
if not overwrite and os.path.isfile(os.path.abspath(test)):
|
|
print("Error: can't generate a test that already exists:", test)
|
|
return 1
|
|
|
|
abs_test = os.path.abspath(test)
|
|
if doc is None:
|
|
doc = self.guess_doc(abs_test)
|
|
if suite is None:
|
|
guessed_suite, err = self.guess_suite(abs_test)
|
|
if err:
|
|
print(err)
|
|
return 1
|
|
suite = guessed_suite
|
|
|
|
else:
|
|
test = None
|
|
if doc is None:
|
|
doc = "html"
|
|
|
|
if not suite:
|
|
print("We couldn't automatically determine a suite. "
|
|
"Please specify `--suite` with one of the following options:\n{}\n"
|
|
"If you'd like to add support to a new suite, please file a bug "
|
|
"blocking https://bugzilla.mozilla.org/show_bug.cgi?id=1540285."
|
|
.format(ADD_TEST_SUPPORTED_SUITES))
|
|
return 1
|
|
|
|
if doc not in ADD_TEST_SUPPORTED_DOCS:
|
|
print("Error: invalid `doc`. Either pass in a test with a valid extension"
|
|
"({}) or pass in the `doc` argument".format(ADD_TEST_SUPPORTED_DOCS))
|
|
return 1
|
|
|
|
creator_cls = addtest.creator_for_suite(suite)
|
|
|
|
if creator_cls is None:
|
|
print("Sorry, `addtest` doesn't currently know how to add {}".format(suite))
|
|
return 1
|
|
|
|
creator = creator_cls(self.topsrcdir, test, suite, doc, **kwargs)
|
|
|
|
creator.check_args()
|
|
|
|
paths = []
|
|
added_tests = False
|
|
for path, template in creator:
|
|
if not template:
|
|
continue
|
|
added_tests = True
|
|
if (path):
|
|
paths.append(path)
|
|
print("Adding a test file at {} (suite `{}`)".format(path, suite))
|
|
|
|
try:
|
|
os.makedirs(os.path.dirname(path))
|
|
except OSError:
|
|
pass
|
|
|
|
with open(path, "w") as f:
|
|
f.write(template)
|
|
else:
|
|
# write to stdout if you passed only suite and doc and not a file path
|
|
print(template)
|
|
|
|
if not added_tests:
|
|
return 1
|
|
|
|
if test:
|
|
creator.update_manifest()
|
|
|
|
# Small hack, should really do this better
|
|
if suite.startswith("wpt-"):
|
|
suite = "web-platform-tests"
|
|
|
|
mach_command = TEST_SUITES[suite]["mach_command"]
|
|
print('Please make sure to add the new test to your commit. '
|
|
'You can now run the test with:\n ./mach {} {}'.format(mach_command, test))
|
|
|
|
if editor is not MISSING_ARG:
|
|
if editor is not None:
|
|
editor = editor
|
|
elif "VISUAL" in os.environ:
|
|
editor = os.environ["VISUAL"]
|
|
elif "EDITOR" in os.environ:
|
|
editor = os.environ["EDITOR"]
|
|
else:
|
|
print('Unable to determine editor; please specify a binary')
|
|
editor = None
|
|
|
|
proc = None
|
|
if editor:
|
|
import subprocess
|
|
proc = subprocess.Popen("%s %s" % (editor, " ".join(paths)), shell=True)
|
|
|
|
if proc:
|
|
proc.wait()
|
|
|
|
return 0
|
|
|
|
def guess_doc(self, abs_test):
|
|
filename = os.path.basename(abs_test)
|
|
return os.path.splitext(filename)[1].strip(".")
|
|
|
|
def guess_suite(self, abs_test):
|
|
# If you pass a abs_test, try to detect the type based on the name
|
|
# and folder. This detection can be skipped if you pass the `type` arg.
|
|
err = None
|
|
guessed_suite = None
|
|
parent = os.path.dirname(abs_test)
|
|
filename = os.path.basename(abs_test)
|
|
|
|
has_browser_ini = os.path.isfile(os.path.join(parent, "browser.ini"))
|
|
has_chrome_ini = os.path.isfile(os.path.join(parent, "chrome.ini"))
|
|
has_plain_ini = os.path.isfile(os.path.join(parent, "mochitest.ini"))
|
|
has_xpcshell_ini = os.path.isfile(os.path.join(parent, "xpcshell.ini"))
|
|
|
|
in_wpt_folder = abs_test.startswith(
|
|
os.path.abspath(os.path.join("testing", "web-platform")))
|
|
|
|
if in_wpt_folder:
|
|
guessed_suite = "web-platform-tests-testharness"
|
|
if "/css/" in abs_test:
|
|
guessed_suite = "web-platform-tests-reftest"
|
|
elif (filename.startswith("test_") and
|
|
has_xpcshell_ini and
|
|
self.guess_doc(abs_test) == "js"):
|
|
guessed_suite = "xpcshell"
|
|
else:
|
|
if filename.startswith("browser_") and has_browser_ini:
|
|
guessed_suite = "mochitest-browser-chrome"
|
|
elif filename.startswith("test_"):
|
|
if has_chrome_ini and has_plain_ini:
|
|
err = ("Error: directory contains both a chrome.ini and mochitest.ini. "
|
|
"Please set --suite=mochitest-chrome or --suite=mochitest-plain.")
|
|
elif has_chrome_ini:
|
|
guessed_suite = "mochitest-chrome"
|
|
elif has_plain_ini:
|
|
guessed_suite = "mochitest-plain"
|
|
return guessed_suite, err
|
|
|
|
|
|
@CommandProvider
|
|
class Test(MachCommandBase):
|
|
@Command('test', category='testing',
|
|
description='Run tests (detects the kind of test and runs it).',
|
|
parser=get_test_parser)
|
|
def test(self, what, extra_args, **log_args):
|
|
"""Run tests from names or paths.
|
|
|
|
mach test accepts arguments specifying which tests to run. Each argument
|
|
can be:
|
|
|
|
* The path to a test file
|
|
* A directory containing tests
|
|
* A test suite name
|
|
* An alias to a test suite name (codes used on TreeHerder)
|
|
|
|
If no input is provided, tests will be run based on files changed in
|
|
the local tree. Relevant tests, tags, or flavors are determined by
|
|
IMPACTED_TESTS annotations in moz.build files relevant to the
|
|
changed files.
|
|
|
|
When paths or directories are given, they are first resolved to test
|
|
files known to the build system.
|
|
|
|
If resolved tests belong to more than one test type/flavor/harness,
|
|
the harness for each relevant type/flavor will be invoked. e.g. if
|
|
you specify a directory with xpcshell and browser chrome mochitests,
|
|
both harnesses will be invoked.
|
|
"""
|
|
from mozlog.commandline import setup_logging
|
|
from mozlog.handlers import StreamHandler
|
|
from moztest.resolve import get_suite_definition, TestResolver, TEST_SUITES
|
|
|
|
resolver = self._spawn(TestResolver)
|
|
run_suites, run_tests = resolver.resolve_metadata(what)
|
|
|
|
if not run_suites and not run_tests:
|
|
print(UNKNOWN_TEST)
|
|
return 1
|
|
|
|
if log_args.get('debugger', None):
|
|
import mozdebug
|
|
if not mozdebug.get_debugger_info(log_args.get('debugger')):
|
|
sys.exit(1)
|
|
extra_args_debugger_notation = '='.join([
|
|
'--debugger',
|
|
log_args.get('debugger')
|
|
]).encode('ascii')
|
|
if extra_args:
|
|
extra_args.append(extra_args_debugger_notation)
|
|
else:
|
|
extra_args = [extra_args_debugger_notation]
|
|
|
|
# Create shared logger
|
|
format_args = {'level': self._mach_context.settings['test']['level']}
|
|
if not run_suites and len(run_tests) == 1:
|
|
format_args['verbose'] = True
|
|
format_args['compact'] = False
|
|
|
|
default_format = self._mach_context.settings['test']['format']
|
|
log = setup_logging('mach-test', log_args, {default_format: sys.stdout}, format_args)
|
|
for handler in log.handlers:
|
|
if isinstance(handler, StreamHandler):
|
|
handler.formatter.inner.summary_on_shutdown = True
|
|
|
|
status = None
|
|
for suite_name in run_suites:
|
|
suite = TEST_SUITES[suite_name]
|
|
kwargs = suite['kwargs']
|
|
kwargs['log'] = log
|
|
|
|
if 'mach_command' in suite:
|
|
res = self._mach_context.commands.dispatch(
|
|
suite['mach_command'], self._mach_context,
|
|
argv=extra_args, **kwargs)
|
|
if res:
|
|
status = res
|
|
|
|
buckets = {}
|
|
for test in run_tests:
|
|
key = (test['flavor'], test.get('subsuite', ''))
|
|
buckets.setdefault(key, []).append(test)
|
|
|
|
for (flavor, subsuite), tests in sorted(buckets.items()):
|
|
_, m = get_suite_definition(flavor, subsuite)
|
|
if 'mach_command' not in m:
|
|
substr = '-{}'.format(subsuite) if subsuite else ''
|
|
print(UNKNOWN_FLAVOR % (flavor, substr))
|
|
status = 1
|
|
continue
|
|
|
|
kwargs = dict(m['kwargs'])
|
|
kwargs['log'] = log
|
|
|
|
res = self._mach_context.commands.dispatch(
|
|
m['mach_command'], self._mach_context,
|
|
argv=extra_args, test_objects=tests, **kwargs)
|
|
if res:
|
|
status = res
|
|
|
|
log.shutdown()
|
|
return status
|
|
|
|
|
|
@CommandProvider
|
|
class MachCommands(MachCommandBase):
|
|
@Command('cppunittest', category='testing',
|
|
description='Run cpp unit tests (C++ tests).')
|
|
@CommandArgument('--enable-webrender', action='store_true', default=False,
|
|
dest='enable_webrender',
|
|
help='Enable the WebRender compositor in Gecko.')
|
|
@CommandArgument('test_files', nargs='*', metavar='N',
|
|
help='Test to run. Can be specified as one or more files or '
|
|
'directories, or omitted. If omitted, the entire test suite is '
|
|
'executed.')
|
|
def run_cppunit_test(self, **params):
|
|
from mozlog import commandline
|
|
|
|
log = params.get('log')
|
|
if not log:
|
|
log = commandline.setup_logging("cppunittest",
|
|
{},
|
|
{"tbpl": sys.stdout})
|
|
|
|
# See if we have crash symbols
|
|
symbols_path = os.path.join(self.distdir, 'crashreporter-symbols')
|
|
if not os.path.isdir(symbols_path):
|
|
symbols_path = None
|
|
|
|
# If no tests specified, run all tests in main manifest
|
|
tests = params['test_files']
|
|
if len(tests) == 0:
|
|
tests = [os.path.join(self.distdir, 'cppunittests')]
|
|
manifest_path = os.path.join(
|
|
self.topsrcdir, 'testing', 'cppunittest.ini')
|
|
else:
|
|
manifest_path = None
|
|
|
|
utility_path = self.bindir
|
|
|
|
if conditions.is_android(self):
|
|
from mozrunner.devices.android_device import (verify_android_device, InstallIntent)
|
|
verify_android_device(self, install=InstallIntent.NO)
|
|
return self.run_android_test(tests, symbols_path, manifest_path, log)
|
|
|
|
return self.run_desktop_test(tests, symbols_path, manifest_path,
|
|
utility_path, log)
|
|
|
|
def run_desktop_test(self, tests, symbols_path, manifest_path,
|
|
utility_path, log):
|
|
import runcppunittests as cppunittests
|
|
from mozlog import commandline
|
|
|
|
parser = cppunittests.CPPUnittestOptions()
|
|
commandline.add_logging_group(parser)
|
|
options, args = parser.parse_args()
|
|
|
|
options.symbols_path = symbols_path
|
|
options.manifest_path = manifest_path
|
|
options.utility_path = utility_path
|
|
options.xre_path = self.bindir
|
|
|
|
try:
|
|
result = cppunittests.run_test_harness(options, tests)
|
|
except Exception as e:
|
|
log.error("Caught exception running cpp unit tests: %s" % str(e))
|
|
result = False
|
|
raise
|
|
|
|
return 0 if result else 1
|
|
|
|
def run_android_test(self, tests, symbols_path, manifest_path, log):
|
|
import remotecppunittests as remotecppunittests
|
|
from mozlog import commandline
|
|
|
|
parser = remotecppunittests.RemoteCPPUnittestOptions()
|
|
commandline.add_logging_group(parser)
|
|
options, args = parser.parse_args()
|
|
|
|
if not options.adb_path:
|
|
from mozrunner.devices.android_device import get_adb_path
|
|
options.adb_path = get_adb_path(self)
|
|
options.symbols_path = symbols_path
|
|
options.manifest_path = manifest_path
|
|
options.xre_path = self.bindir
|
|
options.local_lib = self.bindir.replace('bin', 'fennec')
|
|
for file in os.listdir(os.path.join(self.topobjdir, "dist")):
|
|
if file.endswith(".apk") and file.startswith("fennec"):
|
|
options.local_apk = os.path.join(self.topobjdir, "dist", file)
|
|
log.info("using APK: " + options.local_apk)
|
|
break
|
|
|
|
try:
|
|
result = remotecppunittests.run_test_harness(options, tests)
|
|
except Exception as e:
|
|
log.error("Caught exception running cpp unit tests: %s" % str(e))
|
|
result = False
|
|
raise
|
|
|
|
return 0 if result else 1
|
|
|
|
|
|
def executable_name(name):
|
|
return name + '.exe' if sys.platform.startswith('win') else name
|
|
|
|
|
|
@CommandProvider
|
|
class SpiderMonkeyTests(MachCommandBase):
|
|
@Command('jstests', category='testing',
|
|
description='Run SpiderMonkey JS tests in the JS shell.')
|
|
@CommandArgument('--shell', help='The shell to be used')
|
|
@CommandArgument('params', nargs=argparse.REMAINDER,
|
|
help="Extra arguments to pass down to the test harness.")
|
|
def run_jstests(self, shell, params):
|
|
import subprocess
|
|
|
|
self.virtualenv_manager.ensure()
|
|
python = self.virtualenv_manager.python_path
|
|
|
|
js = shell or os.path.join(self.bindir, executable_name('js'))
|
|
jstest_cmd = [
|
|
python,
|
|
os.path.join(self.topsrcdir, 'js', 'src', 'tests', 'jstests.py'),
|
|
js,
|
|
'--jitflags=jstests',
|
|
] + params
|
|
|
|
return subprocess.call(jstest_cmd)
|
|
|
|
@Command('jit-test', category='testing',
|
|
description='Run SpiderMonkey jit-tests in the JS shell.')
|
|
@CommandArgument('--shell', help='The shell to be used')
|
|
@CommandArgument('params', nargs=argparse.REMAINDER,
|
|
help="Extra arguments to pass down to the test harness.")
|
|
def run_jittests(self, shell, params):
|
|
import subprocess
|
|
|
|
self.virtualenv_manager.ensure()
|
|
python = self.virtualenv_manager.python_path
|
|
|
|
js = shell or os.path.join(self.bindir, executable_name('js'))
|
|
jittest_cmd = [
|
|
python,
|
|
os.path.join(self.topsrcdir, 'js', 'src', 'jit-test', 'jit_test.py'),
|
|
js,
|
|
] + params
|
|
|
|
return subprocess.call(jittest_cmd)
|
|
|
|
@Command('jsapi-tests', category='testing',
|
|
description='Run SpiderMonkey JSAPI tests.')
|
|
@CommandArgument('test_name', nargs='?', metavar='N',
|
|
help='Test to run. Can be a prefix or omitted. If '
|
|
'omitted, the entire test suite is executed.')
|
|
def run_jsapitests(self, test_name=None):
|
|
import subprocess
|
|
|
|
jsapi_tests_cmd = [
|
|
os.path.join(self.bindir, executable_name('jsapi-tests'))
|
|
]
|
|
if test_name:
|
|
jsapi_tests_cmd.append(test_name)
|
|
|
|
return subprocess.call(jsapi_tests_cmd)
|
|
|
|
def run_check_js_msg(self):
|
|
import subprocess
|
|
|
|
self.virtualenv_manager.ensure()
|
|
python = self.virtualenv_manager.python_path
|
|
|
|
check_cmd = [
|
|
python,
|
|
os.path.join(self.topsrcdir, 'config', 'check_js_msg_encoding.py')
|
|
]
|
|
|
|
return subprocess.call(check_cmd)
|
|
|
|
@Command('check-spidermonkey', category='testing',
|
|
description='Run SpiderMonkey tests (JavaScript engine).')
|
|
@CommandArgument('--valgrind', action='store_true',
|
|
help='Run jit-test suite with valgrind flag')
|
|
def run_checkspidermonkey(self, valgrind=False):
|
|
print('Running jit-tests')
|
|
jittest_args = [
|
|
'--no-slow',
|
|
'--jitflags=all',
|
|
]
|
|
if valgrind:
|
|
jittest_args.append('--valgrind')
|
|
jittest_result = self.run_jittests(shell=None, params=jittest_args)
|
|
|
|
print('running jstests')
|
|
jstest_result = self.run_jstests(shell=None, params=[])
|
|
|
|
print('running jsapi-tests')
|
|
jsapi_tests_result = self.run_jsapitests(test_name=None)
|
|
|
|
print('running check-js-msg-encoding')
|
|
check_js_msg_result = self.run_check_js_msg()
|
|
|
|
return jittest_result and jstest_result and jsapi_tests_result and check_js_msg_result
|
|
|
|
|
|
def get_jsshell_parser():
|
|
from jsshell.benchmark import get_parser
|
|
return get_parser()
|
|
|
|
|
|
@CommandProvider
|
|
class JsShellTests(MachCommandBase):
|
|
@Command('jsshell-bench', category='testing',
|
|
parser=get_jsshell_parser,
|
|
description="Run benchmarks in the SpiderMonkey JS shell.")
|
|
def run_jsshelltests(self, **kwargs):
|
|
self._activate_virtualenv()
|
|
from jsshell import benchmark
|
|
return benchmark.run(**kwargs)
|
|
|
|
|
|
@CommandProvider
|
|
class CramTest(MachCommandBase):
|
|
@Command('cramtest', category='testing',
|
|
description="Mercurial style .t tests for command line applications.")
|
|
@CommandArgument('test_paths', nargs='*', metavar='N',
|
|
help="Test paths to run. Each path can be a test file or directory. "
|
|
"If omitted, the entire suite will be run.")
|
|
@CommandArgument('cram_args', nargs=argparse.REMAINDER,
|
|
help="Extra arguments to pass down to the cram binary. See "
|
|
"'./mach python -m cram -- -h' for a list of available options.")
|
|
def cramtest(self, cram_args=None, test_paths=None, test_objects=None):
|
|
self._activate_virtualenv()
|
|
import mozinfo
|
|
from manifestparser import TestManifest
|
|
|
|
if test_objects is None:
|
|
from moztest.resolve import TestResolver
|
|
resolver = self._spawn(TestResolver)
|
|
if test_paths:
|
|
# If we were given test paths, try to find tests matching them.
|
|
test_objects = resolver.resolve_tests(paths=test_paths, flavor='cram')
|
|
else:
|
|
# Otherwise just run everything in CRAMTEST_MANIFESTS
|
|
test_objects = resolver.resolve_tests(flavor='cram')
|
|
|
|
if not test_objects:
|
|
message = 'No tests were collected, check spelling of the test paths.'
|
|
self.log(logging.WARN, 'cramtest', {}, message)
|
|
return 1
|
|
|
|
mp = TestManifest()
|
|
mp.tests.extend(test_objects)
|
|
tests = mp.active_tests(disabled=False, **mozinfo.info)
|
|
|
|
python = self.virtualenv_manager.python_path
|
|
cmd = [python, '-m', 'cram'] + cram_args + [t['relpath'] for t in tests]
|
|
return subprocess.call(cmd, cwd=self.topsrcdir)
|
|
|
|
|
|
@CommandProvider
|
|
class TestInfoCommand(MachCommandBase):
|
|
from datetime import date, timedelta
|
|
|
|
@Command('test-info', category='testing',
|
|
description='Display historical test results.')
|
|
def test_info(self):
|
|
"""
|
|
All functions implemented as subcommands.
|
|
"""
|
|
|
|
@SubCommand('test-info', 'tests',
|
|
description='Display historical test result summary for named tests.')
|
|
@CommandArgument('test_names', nargs=argparse.REMAINDER,
|
|
help='Test(s) of interest.')
|
|
@CommandArgument('--branches',
|
|
default='mozilla-central,mozilla-inbound,autoland',
|
|
help='Report for named branches '
|
|
'(default: mozilla-central,mozilla-inbound,autoland)')
|
|
@CommandArgument('--start',
|
|
default=(date.today() - timedelta(7)
|
|
).strftime("%Y-%m-%d"),
|
|
help='Start date (YYYY-MM-DD)')
|
|
@CommandArgument('--end',
|
|
default=date.today().strftime("%Y-%m-%d"),
|
|
help='End date (YYYY-MM-DD)')
|
|
@CommandArgument('--show-info', action='store_true',
|
|
help='Retrieve and display general test information.')
|
|
@CommandArgument('--show-results', action='store_true',
|
|
help='Retrieve and display ActiveData test result summary.')
|
|
@CommandArgument('--show-durations', action='store_true',
|
|
help='Retrieve and display ActiveData test duration summary.')
|
|
@CommandArgument('--show-tasks', action='store_true',
|
|
help='Retrieve and display ActiveData test task names.')
|
|
@CommandArgument('--show-bugs', action='store_true',
|
|
help='Retrieve and display related Bugzilla bugs.')
|
|
@CommandArgument('--verbose', action='store_true',
|
|
help='Enable debug logging.')
|
|
def test_info_tests(self, **params):
|
|
from mozbuild.base import MozbuildObject
|
|
from mozfile import which
|
|
|
|
self.branches = params['branches']
|
|
self.start = params['start']
|
|
self.end = params['end']
|
|
self.show_info = params['show_info']
|
|
self.show_results = params['show_results']
|
|
self.show_durations = params['show_durations']
|
|
self.show_tasks = params['show_tasks']
|
|
self.show_bugs = params['show_bugs']
|
|
self.verbose = params['verbose']
|
|
|
|
if (not self.show_info and
|
|
not self.show_results and
|
|
not self.show_durations and
|
|
not self.show_tasks and
|
|
not self.show_bugs):
|
|
# by default, show everything
|
|
self.show_info = True
|
|
self.show_results = True
|
|
self.show_durations = True
|
|
self.show_tasks = True
|
|
self.show_bugs = True
|
|
|
|
here = os.path.abspath(os.path.dirname(__file__))
|
|
build_obj = MozbuildObject.from_environment(cwd=here)
|
|
|
|
self._hg = None
|
|
if conditions.is_hg(build_obj):
|
|
self._hg = which('hg')
|
|
if not self._hg:
|
|
raise OSError(errno.ENOENT, "Could not find 'hg' on PATH.")
|
|
|
|
self._git = None
|
|
if conditions.is_git(build_obj):
|
|
self._git = which('git')
|
|
if not self._git:
|
|
raise OSError(errno.ENOENT, "Could not find 'git' on PATH.")
|
|
|
|
for test_name in params['test_names']:
|
|
print("===== %s =====" % test_name)
|
|
self.test_name = test_name
|
|
if len(self.test_name) < 6:
|
|
print("'%s' is too short for a test name!" % self.test_name)
|
|
continue
|
|
self.set_test_name()
|
|
if self.show_bugs:
|
|
self.report_bugs()
|
|
self.set_activedata_test_name()
|
|
if self.show_results:
|
|
self.report_test_results()
|
|
if self.show_durations:
|
|
self.report_test_durations()
|
|
if self.show_tasks:
|
|
self.report_test_tasks()
|
|
|
|
def find_in_hg_or_git(self, test_name):
|
|
if self._hg:
|
|
cmd = [self._hg, 'files', '-I', test_name]
|
|
elif self._git:
|
|
cmd = [self._git, 'ls-files', test_name]
|
|
else:
|
|
return None
|
|
try:
|
|
out = subprocess.check_output(cmd).splitlines()
|
|
except subprocess.CalledProcessError:
|
|
out = None
|
|
return out
|
|
|
|
def set_test_name(self):
|
|
# Generating a unified report for a specific test is complicated
|
|
# by differences in the test name used in various data sources.
|
|
# Consider:
|
|
# - It is often convenient to request a report based only on
|
|
# a short file name, rather than the full path;
|
|
# - Bugs may be filed in bugzilla against a simple, short test
|
|
# name or the full path to the test;
|
|
# - In ActiveData, the full path is usually used, but sometimes
|
|
# also includes additional path components outside of the
|
|
# mercurial repo (common for reftests).
|
|
# This function attempts to find appropriate names for different
|
|
# queries based on the specified test name.
|
|
|
|
import posixpath
|
|
|
|
# full_test_name is full path to file in hg (or git)
|
|
self.full_test_name = None
|
|
out = self.find_in_hg_or_git(self.test_name)
|
|
if out and len(out) == 1:
|
|
self.full_test_name = out[0]
|
|
elif out and len(out) > 1:
|
|
print("Ambiguous test name specified. Found:")
|
|
for line in out:
|
|
print(line)
|
|
else:
|
|
out = self.find_in_hg_or_git('**/%s*' % self.test_name)
|
|
if out and len(out) == 1:
|
|
self.full_test_name = out[0]
|
|
elif out and len(out) > 1:
|
|
print("Ambiguous test name. Found:")
|
|
for line in out:
|
|
print(line)
|
|
if self.full_test_name:
|
|
self.full_test_name.replace(os.sep, posixpath.sep)
|
|
print("Found %s in source control." % self.full_test_name)
|
|
else:
|
|
print("Unable to validate test name '%s'!" % self.test_name)
|
|
self.full_test_name = self.test_name
|
|
|
|
# search for full_test_name in test manifests
|
|
from moztest.resolve import TestResolver, TestManifestLoader
|
|
here = os.path.abspath(os.path.dirname(__file__))
|
|
resolver = TestResolver.from_environment(cwd=here, loader_cls=TestManifestLoader)
|
|
relpath = self._wrap_path_argument(self.full_test_name).relpath()
|
|
tests = list(resolver.resolve_tests(paths=[relpath]))
|
|
if len(tests) == 1:
|
|
relpath = self._wrap_path_argument(tests[0]['manifest']).relpath()
|
|
print("%s found in manifest %s" % (self.full_test_name, relpath))
|
|
if tests[0].get('flavor'):
|
|
print(" flavor: %s" % tests[0]['flavor'])
|
|
if tests[0].get('skip-if'):
|
|
print(" skip-if: %s" % tests[0]['skip-if'])
|
|
if tests[0].get('fail-if'):
|
|
print(" fail-if: %s" % tests[0]['fail-if'])
|
|
elif len(tests) == 0:
|
|
print("%s not found in any test manifest!" % self.full_test_name)
|
|
else:
|
|
print("%s found in more than one manifest!" % self.full_test_name)
|
|
|
|
# short_name is full_test_name without path
|
|
self.short_name = None
|
|
name_idx = self.full_test_name.rfind('/')
|
|
if name_idx > 0:
|
|
self.short_name = self.full_test_name[name_idx + 1:]
|
|
if self.short_name and self.short_name == self.test_name:
|
|
self.short_name = None
|
|
|
|
if not (self.show_results or self.show_durations or self.show_tasks):
|
|
# no need to determine ActiveData name if not querying
|
|
return
|
|
|
|
def set_activedata_test_name(self):
|
|
import re
|
|
|
|
# activedata_test_name is name in ActiveData
|
|
self.activedata_test_name = None
|
|
simple_names = [
|
|
self.full_test_name,
|
|
self.test_name,
|
|
self.short_name
|
|
]
|
|
simple_names = [x for x in simple_names if x]
|
|
searches = [
|
|
{"in": {"result.test": simple_names}},
|
|
]
|
|
regex_names = [".*%s.*" % re.escape(x) for x in simple_names if x]
|
|
for r in regex_names:
|
|
searches.append({"regexp": {"result.test": r}})
|
|
query = {
|
|
"from": "unittest",
|
|
"format": "list",
|
|
"limit": 10,
|
|
"groupby": ["result.test"],
|
|
"where": {"and": [
|
|
{"or": searches},
|
|
{"in": {"build.branch": self.branches.split(',')}},
|
|
{"gt": {"run.timestamp": {"date": self.start}}},
|
|
{"lt": {"run.timestamp": {"date": self.end}}}
|
|
]}
|
|
}
|
|
print("Querying ActiveData...") # Following query can take a long time
|
|
data = self.submit(query)
|
|
if data and len(data) > 0:
|
|
self.activedata_test_name = [
|
|
d['result']['test']
|
|
for p in simple_names + regex_names
|
|
for d in data
|
|
if re.match(p + "$", d['result']['test'])
|
|
][0] # first match is best match
|
|
if self.activedata_test_name:
|
|
print("Found records matching '%s' in ActiveData." %
|
|
self.activedata_test_name)
|
|
else:
|
|
print("Unable to find matching records in ActiveData; using %s!" %
|
|
self.test_name)
|
|
self.activedata_test_name = self.test_name
|
|
|
|
def get_run_types(self, record):
|
|
types_label = ""
|
|
if 'run' in record and 'type' in record['run']:
|
|
run_types = record['run']['type']
|
|
run_types = run_types if isinstance(run_types, list) else [run_types]
|
|
fission = True if 'fis' in run_types else False
|
|
for run_type in run_types:
|
|
# chunked is not interesting
|
|
if run_type == 'chunked':
|
|
continue
|
|
# fission implies e10s
|
|
if fission and run_type == 'e10s':
|
|
continue
|
|
types_label += "-" + run_type
|
|
return types_label
|
|
|
|
def get_platform(self, record):
|
|
if 'platform' in record['build']:
|
|
platform = record['build']['platform']
|
|
else:
|
|
platform = "-"
|
|
tp = record['build']['type']
|
|
if type(tp) is list:
|
|
tp = "-".join(tp)
|
|
return "%s/%s%s:" % (platform, tp, self.get_run_types(record))
|
|
|
|
def submit(self, query):
|
|
import requests
|
|
import datetime
|
|
if self.verbose:
|
|
print(datetime.datetime.now())
|
|
print(json.dumps(query))
|
|
response = requests.post("http://activedata.allizom.org/query",
|
|
data=json.dumps(query),
|
|
stream=True)
|
|
if self.verbose:
|
|
print(datetime.datetime.now())
|
|
print(response)
|
|
response.raise_for_status()
|
|
data = response.json()["data"]
|
|
return data
|
|
|
|
def report_test_results(self):
|
|
# Report test pass/fail summary from ActiveData
|
|
query = {
|
|
"from": "unittest",
|
|
"format": "list",
|
|
"limit": 100,
|
|
"groupby": ["build.platform", "build.type", "run.type"],
|
|
"select": [
|
|
{"aggregate": "count"},
|
|
{
|
|
"name": "failures",
|
|
"value": {"case": [
|
|
{"when": {"eq": {"result.ok": "F"}}, "then": 1}
|
|
]},
|
|
"aggregate": "sum",
|
|
"default": 0
|
|
},
|
|
{
|
|
"name": "skips",
|
|
"value": {"case": [
|
|
{"when": {"eq": {"result.status": "SKIP"}}, "then": 1}
|
|
]},
|
|
"aggregate": "sum",
|
|
"default": 0
|
|
}
|
|
],
|
|
"where": {"and": [
|
|
{"eq": {"result.test": self.activedata_test_name}},
|
|
{"in": {"build.branch": self.branches.split(',')}},
|
|
{"gt": {"run.timestamp": {"date": self.start}}},
|
|
{"lt": {"run.timestamp": {"date": self.end}}}
|
|
]}
|
|
}
|
|
print("\nTest results for %s on %s between %s and %s" %
|
|
(self.activedata_test_name, self.branches, self.start, self.end))
|
|
data = self.submit(query)
|
|
if data and len(data) > 0:
|
|
data.sort(key=self.get_platform)
|
|
worst_rate = 0.0
|
|
worst_platform = None
|
|
total_runs = 0
|
|
total_failures = 0
|
|
for record in data:
|
|
platform = self.get_platform(record)
|
|
if platform.startswith("-"):
|
|
continue
|
|
runs = record['count']
|
|
total_runs = total_runs + runs
|
|
failures = record.get('failures', 0)
|
|
skips = record.get('skips', 0)
|
|
total_failures = total_failures + failures
|
|
rate = (float)(failures) / runs
|
|
if rate >= worst_rate:
|
|
worst_rate = rate
|
|
worst_platform = platform
|
|
worst_failures = failures
|
|
worst_runs = runs
|
|
print("%-40s %6d failures (%6d skipped) in %6d runs" % (
|
|
platform, failures, skips, runs))
|
|
print("\nTotal: %d failures in %d runs or %.3f failures/run" %
|
|
(total_failures, total_runs, (float)(total_failures) / total_runs))
|
|
if worst_failures > 0:
|
|
print("Worst rate on %s %d failures in %d runs or %.3f failures/run" %
|
|
(worst_platform, worst_failures, worst_runs, worst_rate))
|
|
else:
|
|
print("No test result data found.")
|
|
|
|
def report_test_durations(self):
|
|
# Report test durations summary from ActiveData
|
|
query = {
|
|
"from": "unittest",
|
|
"format": "list",
|
|
"limit": 100,
|
|
"groupby": ["build.platform", "build.type", "run.type"],
|
|
"select": [
|
|
{"value": "result.duration",
|
|
"aggregate": "average", "name": "average"},
|
|
{"value": "result.duration", "aggregate": "min", "name": "min"},
|
|
{"value": "result.duration", "aggregate": "max", "name": "max"},
|
|
{"aggregate": "count"}
|
|
],
|
|
"where": {"and": [
|
|
{"eq": {"result.ok": "T"}},
|
|
{"eq": {"result.test": self.activedata_test_name}},
|
|
{"in": {"build.branch": self.branches.split(',')}},
|
|
{"gt": {"run.timestamp": {"date": self.start}}},
|
|
{"lt": {"run.timestamp": {"date": self.end}}}
|
|
]}
|
|
}
|
|
data = self.submit(query)
|
|
print("\nTest durations for %s on %s between %s and %s" %
|
|
(self.activedata_test_name, self.branches, self.start, self.end))
|
|
if data and len(data) > 0:
|
|
data.sort(key=self.get_platform)
|
|
for record in data:
|
|
platform = self.get_platform(record)
|
|
if platform.startswith("-"):
|
|
continue
|
|
print("%-40s %6.2f s (%.2f s - %.2f s over %d runs)" % (
|
|
platform, record['average'], record['min'],
|
|
record['max'], record['count']))
|
|
else:
|
|
print("No test durations found.")
|
|
|
|
def report_test_tasks(self):
|
|
# Report test tasks summary from ActiveData
|
|
query = {
|
|
"from": "unittest",
|
|
"format": "list",
|
|
"limit": 1000,
|
|
"select": ["build.platform", "build.type", "run.type", "run.name"],
|
|
"where": {"and": [
|
|
{"eq": {"result.test": self.activedata_test_name}},
|
|
{"in": {"build.branch": self.branches.split(',')}},
|
|
{"gt": {"run.timestamp": {"date": self.start}}},
|
|
{"lt": {"run.timestamp": {"date": self.end}}}
|
|
]}
|
|
}
|
|
data = self.submit(query)
|
|
print("\nTest tasks for %s on %s between %s and %s" %
|
|
(self.activedata_test_name, self.branches, self.start, self.end))
|
|
if data and len(data) > 0:
|
|
data.sort(key=self.get_platform)
|
|
consolidated = {}
|
|
for record in data:
|
|
platform = self.get_platform(record)
|
|
if platform not in consolidated:
|
|
consolidated[platform] = {}
|
|
if record['run']['name'] in consolidated[platform]:
|
|
consolidated[platform][record['run']['name']] += 1
|
|
else:
|
|
consolidated[platform][record['run']['name']] = 1
|
|
for key in sorted(consolidated.keys()):
|
|
tasks = ""
|
|
for task in consolidated[key].keys():
|
|
if tasks:
|
|
tasks += "\n%-40s " % ""
|
|
tasks += task
|
|
tasks += " in %d runs" % consolidated[key][task]
|
|
print("%-40s %s" % (key, tasks))
|
|
else:
|
|
print("No test tasks found.")
|
|
|
|
def report_bugs(self):
|
|
# Report open bugs matching test name
|
|
import requests
|
|
search = self.full_test_name
|
|
if self.test_name:
|
|
search = '%s,%s' % (search, self.test_name)
|
|
if self.short_name:
|
|
search = '%s,%s' % (search, self.short_name)
|
|
payload = {'quicksearch': search,
|
|
'include_fields': 'id,summary'}
|
|
response = requests.get('https://bugzilla.mozilla.org/rest/bug',
|
|
payload)
|
|
response.raise_for_status()
|
|
json_response = response.json()
|
|
print("\nBugzilla quick search for '%s':" % search)
|
|
if 'bugs' in json_response:
|
|
for bug in json_response['bugs']:
|
|
print("Bug %s: %s" % (bug['id'], bug['summary']))
|
|
else:
|
|
print("No bugs found.")
|
|
|
|
@SubCommand('test-info', 'long-tasks',
|
|
description='Find tasks approaching their taskcluster max-run-time.')
|
|
@CommandArgument('--branches',
|
|
default='mozilla-central,mozilla-inbound,autoland',
|
|
help='Report for named branches '
|
|
'(default: mozilla-central,mozilla-inbound,autoland)')
|
|
@CommandArgument('--start',
|
|
default=(date.today() - timedelta(7)
|
|
).strftime("%Y-%m-%d"),
|
|
help='Start date (YYYY-MM-DD)')
|
|
@CommandArgument('--end',
|
|
default=date.today().strftime("%Y-%m-%d"),
|
|
help='End date (YYYY-MM-DD)')
|
|
@CommandArgument('--max-threshold-pct',
|
|
default=90.0,
|
|
help='Count tasks exceeding this percentage of max-run-time.')
|
|
@CommandArgument('--filter-threshold-pct',
|
|
default=0.5,
|
|
help='Report tasks exceeding this percentage of long tasks.')
|
|
@CommandArgument('--verbose', action='store_true',
|
|
help='Enable debug logging.')
|
|
def report_long_running_tasks(self, **params):
|
|
def get_long_running_ratio(record):
|
|
count = record['count']
|
|
tasks_gt_pct = record['tasks_gt_pct']
|
|
return count / tasks_gt_pct
|
|
|
|
branches = params['branches']
|
|
start = params['start']
|
|
end = params['end']
|
|
self.verbose = params['verbose']
|
|
threshold_pct = float(params['max_threshold_pct'])
|
|
filter_threshold_pct = float(params['filter_threshold_pct'])
|
|
|
|
# Search test durations in ActiveData for long-running tests
|
|
query = {
|
|
"from": "task",
|
|
"format": "list",
|
|
"groupby": ["run.name"],
|
|
"limit": 1000,
|
|
"select": [
|
|
{
|
|
"value": "task.maxRunTime",
|
|
"aggregate": "median",
|
|
"name": "max_run_time"
|
|
},
|
|
{
|
|
"aggregate": "count"
|
|
},
|
|
{
|
|
"value": {
|
|
"when": {
|
|
"gt": [
|
|
{
|
|
"div": ["action.duration", "task.maxRunTime"]
|
|
}, threshold_pct/100.0
|
|
]
|
|
},
|
|
"then": 1
|
|
},
|
|
"aggregate": "sum",
|
|
"name": "tasks_gt_pct"
|
|
},
|
|
],
|
|
"where": {"and": [
|
|
{"in": {"build.branch": branches.split(',')}},
|
|
{"gt": {"task.run.start_time": {"date": start}}},
|
|
{"lte": {"task.run.start_time": {"date": end}}},
|
|
{"eq": {"task.state": "completed"}},
|
|
]}
|
|
}
|
|
data = self.submit(query)
|
|
print("\nTasks nearing their max-run-time on %s between %s and %s" %
|
|
(branches, start, end))
|
|
if data and len(data) > 0:
|
|
filtered = []
|
|
for record in data:
|
|
if 'tasks_gt_pct' in record:
|
|
count = record['count']
|
|
tasks_gt_pct = record['tasks_gt_pct']
|
|
if float(tasks_gt_pct) / count > filter_threshold_pct / 100.0:
|
|
filtered.append(record)
|
|
filtered.sort(key=get_long_running_ratio)
|
|
if not filtered:
|
|
print("No long running tasks found.")
|
|
for record in filtered:
|
|
name = record['run']['name']
|
|
count = record['count']
|
|
max_run_time = record['max_run_time']
|
|
tasks_gt_pct = record['tasks_gt_pct']
|
|
print("%-55s: %d of %d runs (%.1f%%) exceeded %d%% of max-run-time (%d s)" %
|
|
(name, tasks_gt_pct, count, tasks_gt_pct * 100 / count,
|
|
threshold_pct, max_run_time))
|
|
else:
|
|
print("No tasks found.")
|
|
|
|
@SubCommand('test-info', 'report',
|
|
description='Generate a json report of test manifests and/or tests '
|
|
'categorized by Bugzilla component and optionally filtered '
|
|
'by path, component, and/or manifest annotations.')
|
|
@CommandArgument('--components', default=None,
|
|
help='Comma-separated list of Bugzilla components.'
|
|
' eg. Testing::General,Core::WebVR')
|
|
@CommandArgument('--flavor',
|
|
help='Limit results to tests of the specified flavor (eg. "xpcshell").')
|
|
@CommandArgument('--subsuite',
|
|
help='Limit results to tests of the specified subsuite (eg. "devtools").')
|
|
@CommandArgument('paths', nargs=argparse.REMAINDER,
|
|
help='File system paths of interest.')
|
|
@CommandArgument('--show-manifests', action='store_true',
|
|
help='Include test manifests in report.')
|
|
@CommandArgument('--show-tests', action='store_true',
|
|
help='Include individual tests in report.')
|
|
@CommandArgument('--show-summary', action='store_true',
|
|
help='Include summary in report.')
|
|
@CommandArgument('--show-annotations', action='store_true',
|
|
help='Include list of manifest annotation conditions in report.')
|
|
@CommandArgument('--filter-values',
|
|
help='Comma-separated list of value regular expressions to filter on; '
|
|
'displayed tests contain all specified values.')
|
|
@CommandArgument('--filter-keys',
|
|
help='Comma-separated list of test keys to filter on, '
|
|
'like "skip-if"; only these fields will be searched '
|
|
'for filter-values.')
|
|
@CommandArgument('--no-component-report', action='store_false',
|
|
dest="show_components", default=True,
|
|
help='Do not categorize by bugzilla component.')
|
|
@CommandArgument('--output-file',
|
|
help='Path to report file.')
|
|
def test_report(self, components, flavor, subsuite, paths,
|
|
show_manifests, show_tests, show_summary, show_annotations,
|
|
filter_values, filter_keys, show_components, output_file):
|
|
import mozpack.path as mozpath
|
|
import re
|
|
from mozbuild.build_commands import Build
|
|
from moztest.resolve import TestResolver, TestManifestLoader
|
|
|
|
def matches_filters(test):
|
|
'''
|
|
Return True if all of the requested filter_values are found in this test;
|
|
if filter_keys are specified, restrict search to those test keys.
|
|
'''
|
|
for value in filter_values:
|
|
value_found = False
|
|
for key in test:
|
|
if not filter_keys or key in filter_keys:
|
|
if re.search(value, test[key]):
|
|
value_found = True
|
|
break
|
|
if not value_found:
|
|
return False
|
|
return True
|
|
|
|
# Ensure useful report by default
|
|
if not show_manifests and not show_tests and not show_summary and not show_annotations:
|
|
show_manifests = True
|
|
show_summary = True
|
|
|
|
by_component = {}
|
|
if components:
|
|
components = components.split(',')
|
|
if filter_keys:
|
|
filter_keys = filter_keys.split(',')
|
|
if filter_values:
|
|
filter_values = filter_values.split(',')
|
|
else:
|
|
filter_values = []
|
|
display_keys = (filter_keys or []) + ['skip-if', 'fail-if', 'fails-if']
|
|
display_keys = set(display_keys)
|
|
|
|
try:
|
|
self.config_environment
|
|
except BuildEnvironmentNotFoundException:
|
|
print("Looks like configure has not run yet, running it now...")
|
|
builder = Build(self._mach_context)
|
|
builder.configure()
|
|
|
|
print("Finding tests...")
|
|
here = os.path.abspath(os.path.dirname(__file__))
|
|
resolver = TestResolver.from_environment(cwd=here, loader_cls=TestManifestLoader)
|
|
tests = list(resolver.resolve_tests(paths=paths, flavor=flavor,
|
|
subsuite=subsuite))
|
|
|
|
manifest_paths = set()
|
|
for t in tests:
|
|
if 'manifest' in t and t['manifest'] is not None:
|
|
manifest_paths.add(t['manifest'])
|
|
manifest_count = len(manifest_paths)
|
|
print("Resolver found {} tests, {} manifests".format(len(tests), manifest_count))
|
|
|
|
if show_manifests:
|
|
by_component['manifests'] = {}
|
|
manifest_paths = list(manifest_paths)
|
|
manifest_paths.sort()
|
|
relpaths = []
|
|
for manifest_path in manifest_paths:
|
|
relpath = mozpath.relpath(manifest_path, self.topsrcdir)
|
|
if mozpath.commonprefix((manifest_path, self.topsrcdir)) != self.topsrcdir:
|
|
continue
|
|
relpaths.append(relpath)
|
|
reader = self.mozbuild_reader(config_mode='empty')
|
|
files_info = reader.files_info(relpaths)
|
|
for manifest_path in manifest_paths:
|
|
relpath = mozpath.relpath(manifest_path, self.topsrcdir)
|
|
if mozpath.commonprefix((manifest_path, self.topsrcdir)) != self.topsrcdir:
|
|
continue
|
|
manifest_info = None
|
|
if relpath in files_info:
|
|
bug_component = files_info[relpath].get('BUG_COMPONENT')
|
|
key = "{}::{}".format(bug_component.product, bug_component.component)
|
|
if (not components) or (key in components):
|
|
manifest_info = {
|
|
'manifest': relpath,
|
|
'tests': 0,
|
|
'skipped': 0
|
|
}
|
|
rkey = key if show_components else 'all'
|
|
if rkey in by_component['manifests']:
|
|
by_component['manifests'][rkey].append(manifest_info)
|
|
else:
|
|
by_component['manifests'][rkey] = [manifest_info]
|
|
if manifest_info:
|
|
for t in tests:
|
|
if t['manifest'] == manifest_path:
|
|
manifest_info['tests'] += 1
|
|
if t.get('skip-if'):
|
|
manifest_info['skipped'] += 1
|
|
for key in by_component['manifests']:
|
|
by_component['manifests'][key].sort()
|
|
|
|
if show_tests:
|
|
by_component['tests'] = {}
|
|
|
|
if show_tests or show_summary or show_annotations:
|
|
test_count = 0
|
|
failed_count = 0
|
|
skipped_count = 0
|
|
annotation_count = 0
|
|
condition_count = 0
|
|
component_set = set()
|
|
relpaths = []
|
|
conditions = {}
|
|
known_unconditional_annotations = ['skip', 'fail', 'asserts', 'random']
|
|
known_conditional_annotations = ['skip-if', 'fail-if', 'run-if',
|
|
'fails-if', 'fuzzy-if', 'random-if', 'asserts-if']
|
|
for t in tests:
|
|
relpath = t.get('srcdir_relpath')
|
|
relpaths.append(relpath)
|
|
reader = self.mozbuild_reader(config_mode='empty')
|
|
files_info = reader.files_info(relpaths)
|
|
for t in tests:
|
|
if not matches_filters(t):
|
|
continue
|
|
if 'referenced-test' in t:
|
|
# Avoid double-counting reftests: disregard reference file entries
|
|
continue
|
|
if show_annotations:
|
|
for key in t:
|
|
if key in known_unconditional_annotations:
|
|
annotation_count += 1
|
|
if key in known_conditional_annotations:
|
|
annotation_count += 1
|
|
# Here 'key' is a manifest annotation type like 'skip-if' and t[key]
|
|
# is the associated condition. For example, the manifestparser
|
|
# manifest annotation, "skip-if = os == 'win'", is expected to be
|
|
# encoded as t['skip-if'] = "os == 'win'".
|
|
# To allow for reftest manifests, t[key] may have multiple entries
|
|
# separated by ';', each corresponding to a condition for that test
|
|
# and annotation type. For example,
|
|
# "skip-if(Android&&webrender) skip-if(OSX)", would be
|
|
# encoded as t['skip-if'] = "Android&&webrender;OSX".
|
|
annotation_conditions = t[key].split(';')
|
|
for condition in annotation_conditions:
|
|
condition_count += 1
|
|
# Trim reftest fuzzy-if ranges: everything after the first comma
|
|
# eg. "Android,0-2,1-3" -> "Android"
|
|
condition = condition.split(',')[0]
|
|
if condition not in conditions:
|
|
conditions[condition] = 0
|
|
conditions[condition] += 1
|
|
test_count += 1
|
|
relpath = t.get('srcdir_relpath')
|
|
if relpath in files_info:
|
|
bug_component = files_info[relpath].get('BUG_COMPONENT')
|
|
key = "{}::{}".format(bug_component.product, bug_component.component)
|
|
if (not components) or (key in components):
|
|
component_set.add(key)
|
|
test_info = {'test': relpath}
|
|
for test_key in display_keys:
|
|
value = t.get(test_key)
|
|
if value:
|
|
test_info[test_key] = value
|
|
if t.get('fail-if'):
|
|
failed_count += 1
|
|
if t.get('fails-if'):
|
|
failed_count += 1
|
|
if t.get('skip-if'):
|
|
skipped_count += 1
|
|
if show_tests:
|
|
rkey = key if show_components else 'all'
|
|
if rkey in by_component['tests']:
|
|
by_component['tests'][rkey].append(test_info)
|
|
else:
|
|
by_component['tests'][rkey] = [test_info]
|
|
if show_tests:
|
|
for key in by_component['tests']:
|
|
by_component['tests'][key].sort(key=lambda k: k['test'])
|
|
|
|
if show_summary:
|
|
by_component['summary'] = {}
|
|
by_component['summary']['components'] = len(component_set)
|
|
by_component['summary']['manifests'] = manifest_count
|
|
by_component['summary']['tests'] = test_count
|
|
by_component['summary']['failed tests'] = failed_count
|
|
by_component['summary']['skipped tests'] = skipped_count
|
|
|
|
if show_annotations:
|
|
by_component['annotations'] = {}
|
|
by_component['annotations']['total annotations'] = annotation_count
|
|
by_component['annotations']['total conditions'] = condition_count
|
|
by_component['annotations']['unique conditions'] = len(conditions)
|
|
by_component['annotations']['conditions'] = conditions
|
|
|
|
json_report = json.dumps(by_component, indent=2, sort_keys=True)
|
|
if output_file:
|
|
output_file = os.path.abspath(output_file)
|
|
output_dir = os.path.dirname(output_file)
|
|
if not os.path.isdir(output_dir):
|
|
os.makedirs(output_dir)
|
|
|
|
with open(output_file, 'w') as f:
|
|
f.write(json_report)
|
|
else:
|
|
print(json_report)
|
|
|
|
|
|
@CommandProvider
|
|
class RustTests(MachCommandBase):
|
|
@Command('rusttests', category='testing',
|
|
conditions=[conditions.is_non_artifact_build],
|
|
description="Run rust unit tests (via cargo test).")
|
|
def run_rusttests(self, **kwargs):
|
|
return self._mach_context.commands.dispatch('build', self._mach_context,
|
|
what=['pre-export',
|
|
'export',
|
|
'recurse_rusttests'])
|
|
|
|
|
|
@CommandProvider
|
|
class TestFluentMigration(MachCommandBase):
|
|
@Command('fluent-migration-test', category='testing',
|
|
description="Test Fluent migration recipes.")
|
|
@CommandArgument('test_paths', nargs='*', metavar='N',
|
|
help="Recipe paths to test.")
|
|
def run_migration_tests(self, test_paths=None, **kwargs):
|
|
if not test_paths:
|
|
test_paths = []
|
|
self._activate_virtualenv()
|
|
from test_fluent_migrations import fmt
|
|
rv = 0
|
|
with_context = []
|
|
for to_test in test_paths:
|
|
try:
|
|
context = fmt.inspect_migration(to_test)
|
|
for issue in context['issues']:
|
|
self.log(logging.ERROR, 'fluent-migration-test', {
|
|
'error': issue['msg'],
|
|
'file': to_test,
|
|
}, 'ERROR in {file}: {error}')
|
|
if context['issues']:
|
|
continue
|
|
with_context.append({
|
|
'to_test': to_test,
|
|
'references': context['references'],
|
|
})
|
|
except Exception as e:
|
|
self.log(logging.ERROR, 'fluent-migration-test', {
|
|
'error': str(e),
|
|
'file': to_test
|
|
}, 'ERROR in {file}: {error}')
|
|
rv |= 1
|
|
obj_dir = fmt.prepare_object_dir(self)
|
|
for context in with_context:
|
|
rv |= fmt.test_migration(self, obj_dir, **context)
|
|
return rv
|