v8/tools/unittests/run_tests_test.py
Michael Achenbach fb0a2ea25e Revert "Reland "[build] Add V8-specific dcheck_always_on""
This reverts commit 67960ba110.

Reason for revert:
This has been properly fixed by https://crrev.com/c/3053740.
Now dcheck_always_on already defaults to false for subprojects
like V8 and no other switch is required. The switch didn't fully
work anyways due to https://crbug.com/1231890.

Original change's description:
> Reland "[build] Add V8-specific dcheck_always_on"
>
> This is a reland of cecc666f4d
>
> Depends on:
> https://crrev.com/c/3043611
>
> Original change's description:
> > [build] Add V8-specific dcheck_always_on
> >
> > This makes the V8 dcheck control independent of Chromium's and
> > prepares switching Chromium's default behavior without affecting V8
> > developers or builders.
> >
> > Preparation for: https://crrev.com/c/2893204
> >
> > Bug: chromium:1225701
> > Change-Id: I520b96019b04196f4420716ff3500ebd6c21666f
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3038528
> > Reviewed-by: Leszek Swirski <leszeks@chromium.org>
> > Commit-Queue: Michael Achenbach <machenbach@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#75827}
>
> Bug: chromium:1225701
> Change-Id: I56568b78592addba01793d2d14f768c9ee10103d
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3041670
> Reviewed-by: Liviu Rau <liviurau@chromium.org>
> Commit-Queue: Michael Achenbach <machenbach@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#75839}

Bug: chromium:1225701, chromium:1231890
Change-Id: I7e27f5774d8e162977f30f685da4b15dadcc1084
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3055294
Reviewed-by: Liviu Rau <liviurau@chromium.org>
Commit-Queue: Michael Achenbach <machenbach@chromium.org>
Cr-Commit-Position: refs/heads/master@{#75935}
2021-07-27 11:14:57 +00:00

675 lines
23 KiB
Python
Executable File

#!/usr/bin/env python
# Copyright 2017 the V8 project authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
"""
Global system tests for V8 test runners and fuzzers.
This hooks up the framework under tools/testrunner testing high-level scenarios
with different test suite extensions and build configurations.
"""
# TODO(machenbach): Mock out util.GuessOS to make these tests really platform
# independent.
# TODO(machenbach): Move coverage recording to a global test entry point to
# include other unittest suites in the coverage report.
# TODO(machenbach): Coverage data from multiprocessing doesn't work.
# TODO(majeski): Add some tests for the fuzzers.
# for py2/py3 compatibility
from __future__ import print_function
import collections
import contextlib
import json
import os
import shutil
import subprocess
import sys
import tempfile
import unittest
from cStringIO import StringIO
TOOLS_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
TEST_DATA_ROOT = os.path.join(TOOLS_ROOT, 'unittests', 'testdata')
RUN_TESTS_PY = os.path.join(TOOLS_ROOT, 'run-tests.py')
Result = collections.namedtuple(
'Result', ['stdout', 'stderr', 'returncode'])
Result.__str__ = lambda self: (
'\nReturncode: %s\nStdout:\n%s\nStderr:\n%s\n' %
(self.returncode, self.stdout, self.stderr))
@contextlib.contextmanager
def temp_dir():
"""Wrapper making a temporary directory available."""
path = None
try:
path = tempfile.mkdtemp('v8_test_')
yield path
finally:
if path:
shutil.rmtree(path)
@contextlib.contextmanager
def temp_base(baseroot='testroot1'):
"""Wrapper that sets up a temporary V8 test root.
Args:
baseroot: The folder with the test root blueprint. Relevant files will be
copied to the temporary test root, to guarantee a fresh setup with no
dirty state.
"""
basedir = os.path.join(TEST_DATA_ROOT, baseroot)
with temp_dir() as tempbase:
builddir = os.path.join(tempbase, 'out', 'build')
testroot = os.path.join(tempbase, 'test')
os.makedirs(builddir)
shutil.copy(os.path.join(basedir, 'v8_build_config.json'), builddir)
shutil.copy(os.path.join(basedir, 'd8_mocked.py'), builddir)
for suite in os.listdir(os.path.join(basedir, 'test')):
os.makedirs(os.path.join(testroot, suite))
for entry in os.listdir(os.path.join(basedir, 'test', suite)):
shutil.copy(
os.path.join(basedir, 'test', suite, entry),
os.path.join(testroot, suite))
yield tempbase
@contextlib.contextmanager
def capture():
"""Wrapper that replaces system stdout/stderr an provides the streams."""
oldout = sys.stdout
olderr = sys.stderr
try:
stdout=StringIO()
stderr=StringIO()
sys.stdout = stdout
sys.stderr = stderr
yield stdout, stderr
finally:
sys.stdout = oldout
sys.stderr = olderr
def run_tests(basedir, *args, **kwargs):
"""Executes the test runner with captured output."""
with capture() as (stdout, stderr):
sys_args = ['--command-prefix', sys.executable] + list(args)
if kwargs.get('infra_staging', False):
sys_args.append('--infra-staging')
else:
sys_args.append('--no-infra-staging')
code = standard_runner.StandardTestRunner(basedir=basedir).execute(sys_args)
return Result(stdout.getvalue(), stderr.getvalue(), code)
def override_build_config(basedir, **kwargs):
"""Override the build config with new values provided as kwargs."""
path = os.path.join(basedir, 'out', 'build', 'v8_build_config.json')
with open(path) as f:
config = json.load(f)
config.update(kwargs)
with open(path, 'w') as f:
json.dump(config, f)
class SystemTest(unittest.TestCase):
@classmethod
def setUpClass(cls):
# Try to set up python coverage and run without it if not available.
cls._cov = None
try:
import coverage
if int(coverage.__version__.split('.')[0]) < 4:
cls._cov = None
print('Python coverage version >= 4 required.')
raise ImportError()
cls._cov = coverage.Coverage(
source=([os.path.join(TOOLS_ROOT, 'testrunner')]),
omit=['*unittest*', '*__init__.py'],
)
cls._cov.exclude('raise NotImplementedError')
cls._cov.exclude('if __name__ == .__main__.:')
cls._cov.exclude('except TestRunnerError:')
cls._cov.exclude('except KeyboardInterrupt:')
cls._cov.exclude('if options.verbose:')
cls._cov.exclude('if verbose:')
cls._cov.exclude('pass')
cls._cov.exclude('assert False')
cls._cov.start()
except ImportError:
print('Running without python coverage.')
sys.path.append(TOOLS_ROOT)
global standard_runner
from testrunner import standard_runner
global num_fuzzer
from testrunner import num_fuzzer
from testrunner.local import command
from testrunner.local import pool
command.setup_testing()
pool.setup_testing()
@classmethod
def tearDownClass(cls):
if cls._cov:
cls._cov.stop()
print('')
print(cls._cov.report(show_missing=True))
def testPass(self):
"""Test running only passing tests in two variants.
Also test printing durations.
"""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default,stress',
'--time',
'sweet/bananas',
'sweet/raspberries',
)
self.assertIn('sweet/bananas default: PASS', result.stdout, result)
# TODO(majeski): Implement for test processors
# self.assertIn('Total time:', result.stderr, result)
# self.assertIn('sweet/bananas', result.stderr, result)
self.assertEqual(0, result.returncode, result)
def testPassHeavy(self):
"""Test running with some tests marked heavy."""
with temp_base(baseroot='testroot3') as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=nooptimization',
'-j2',
'sweet',
)
self.assertIn('7 tests ran', result.stdout, result)
self.assertEqual(0, result.returncode, result)
def testShardedProc(self):
with temp_base() as basedir:
for shard in [1, 2]:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default,stress',
'--shard-count=2',
'--shard-run=%d' % shard,
'sweet/blackberries',
'sweet/raspberries',
infra_staging=False,
)
# One of the shards gets one variant of each test.
self.assertIn('2 tests ran', result.stdout, result)
if shard == 1:
self.assertIn('sweet/raspberries default', result.stdout, result)
self.assertIn('sweet/raspberries stress', result.stdout, result)
self.assertEqual(0, result.returncode, result)
else:
self.assertIn(
'sweet/blackberries default: FAIL', result.stdout, result)
self.assertIn(
'sweet/blackberries stress: FAIL', result.stdout, result)
self.assertEqual(1, result.returncode, result)
@unittest.skip("incompatible with test processors")
def testSharded(self):
"""Test running a particular shard."""
with temp_base() as basedir:
for shard in [1, 2]:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default,stress',
'--shard-count=2',
'--shard-run=%d' % shard,
'sweet/bananas',
'sweet/raspberries',
)
# One of the shards gets one variant of each test.
self.assertIn('Running 2 tests', result.stdout, result)
self.assertIn('sweet/bananas', result.stdout, result)
self.assertIn('sweet/raspberries', result.stdout, result)
self.assertEqual(0, result.returncode, result)
def testFail(self):
"""Test running only failing tests in two variants."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default,stress',
'sweet/strawberries',
infra_staging=False,
)
self.assertIn('sweet/strawberries default: FAIL', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def check_cleaned_json_output(
self, expected_results_name, actual_json, basedir):
# Check relevant properties of the json output.
with open(actual_json) as f:
json_output = json.load(f)
# Replace duration in actual output as it's non-deterministic. Also
# replace the python executable prefix as it has a different absolute
# path dependent on where this runs.
def replace_variable_data(data):
data['duration'] = 1
data['command'] = ' '.join(
['/usr/bin/python'] + data['command'].split()[1:])
data['command'] = data['command'].replace(basedir + '/', '')
for data in json_output['slowest_tests']:
replace_variable_data(data)
for data in json_output['results']:
replace_variable_data(data)
json_output['duration_mean'] = 1
# We need lexicographic sorting here to avoid non-deterministic behaviour
# The original sorting key is duration, but in our fake test we have
# non-deterministic durations before we reset them to 1
json_output['slowest_tests'].sort(key= lambda x: str(x))
with open(os.path.join(TEST_DATA_ROOT, expected_results_name)) as f:
expected_test_results = json.load(f)
pretty_json = json.dumps(json_output, indent=2, sort_keys=True)
msg = None # Set to pretty_json for bootstrapping.
self.assertDictEqual(json_output, expected_test_results, msg)
def testFailWithRerunAndJSON(self):
"""Test re-running a failing test and output to json."""
with temp_base() as basedir:
json_path = os.path.join(basedir, 'out.json')
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'--rerun-failures-count=2',
'--random-seed=123',
'--json-test-results', json_path,
'sweet/strawberries',
infra_staging=False,
)
self.assertIn('sweet/strawberries default: FAIL', result.stdout, result)
# With test processors we don't count reruns as separated failures.
# TODO(majeski): fix it?
self.assertIn('1 tests failed', result.stdout, result)
self.assertEqual(0, result.returncode, result)
# TODO(majeski): Previously we only reported the variant flags in the
# flags field of the test result.
# After recent changes we report all flags, including the file names.
# This is redundant to the command. Needs investigation.
self.maxDiff = None
self.check_cleaned_json_output(
'expected_test_results1.json', json_path, basedir)
def testFlakeWithRerunAndJSON(self):
"""Test re-running a failing test and output to json."""
with temp_base(baseroot='testroot2') as basedir:
json_path = os.path.join(basedir, 'out.json')
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'--rerun-failures-count=2',
'--random-seed=123',
'--json-test-results', json_path,
'sweet',
infra_staging=False,
)
self.assertIn('sweet/bananaflakes default: FAIL PASS', result.stdout, result)
self.assertIn('=== sweet/bananaflakes (flaky) ===', result.stdout, result)
self.assertIn('1 tests failed', result.stdout, result)
self.assertIn('1 tests were flaky', result.stdout, result)
self.assertEqual(0, result.returncode, result)
self.maxDiff = None
self.check_cleaned_json_output(
'expected_test_results2.json', json_path, basedir)
def testAutoDetect(self):
"""Fake a build with several auto-detected options.
Using all those options at once doesn't really make much sense. This is
merely for getting coverage.
"""
with temp_base() as basedir:
override_build_config(
basedir, dcheck_always_on=True, is_asan=True, is_cfi=True,
is_msan=True, is_tsan=True, is_ubsan_vptr=True, target_cpu='x86',
v8_enable_i18n_support=False, v8_target_cpu='x86',
v8_enable_verify_csa=False, v8_enable_lite_mode=False,
v8_enable_pointer_compression=False,
v8_enable_pointer_compression_shared_cage=False)
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'sweet/bananas',
)
expect_text = (
'>>> Autodetected:\n'
'asan\n'
'cfi_vptr\n'
'dcheck_always_on\n'
'msan\n'
'no_i18n\n'
'tsan\n'
'ubsan_vptr\n'
'webassembly\n'
'>>> Running tests for ia32.release')
self.assertIn(expect_text, result.stdout, result)
self.assertEqual(0, result.returncode, result)
# TODO(machenbach): Test some more implications of the auto-detected
# options, e.g. that the right env variables are set.
def testSkips(self):
"""Test skipping tests in status file for a specific variant."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=nooptimization',
'sweet/strawberries',
infra_staging=False,
)
self.assertIn('0 tests ran', result.stdout, result)
self.assertEqual(2, result.returncode, result)
def testRunSkips(self):
"""Inverse the above. Test parameter to keep running skipped tests."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=nooptimization',
'--run-skipped',
'sweet/strawberries',
)
self.assertIn('1 tests failed', result.stdout, result)
self.assertIn('1 tests ran', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def testDefault(self):
"""Test using default test suites, though no tests are run since they don't
exist in a test setting.
"""
with temp_base() as basedir:
result = run_tests(
basedir,
infra_staging=False,
)
self.assertIn('0 tests ran', result.stdout, result)
self.assertEqual(2, result.returncode, result)
def testNoBuildConfig(self):
"""Test failing run when build config is not found."""
with temp_dir() as basedir:
result = run_tests(basedir)
self.assertIn('Failed to load build config', result.stdout, result)
self.assertEqual(5, result.returncode, result)
def testInconsistentArch(self):
"""Test failing run when attempting to wrongly override the arch."""
with temp_base() as basedir:
result = run_tests(basedir, '--arch=ia32')
self.assertIn(
'--arch value (ia32) inconsistent with build config (x64).',
result.stdout, result)
self.assertEqual(5, result.returncode, result)
def testWrongVariant(self):
"""Test using a bogus variant."""
with temp_base() as basedir:
result = run_tests(basedir, '--variants=meh')
self.assertEqual(5, result.returncode, result)
def testModeFromBuildConfig(self):
"""Test auto-detection of mode from build config."""
with temp_base() as basedir:
result = run_tests(basedir, '--outdir=out/build', 'sweet/bananas')
self.assertIn('Running tests for x64.release', result.stdout, result)
self.assertEqual(0, result.returncode, result)
@unittest.skip("not available with test processors")
def testReport(self):
"""Test the report feature.
This also exercises various paths in statusfile logic.
"""
with temp_base() as basedir:
result = run_tests(
basedir,
'--variants=default',
'sweet',
'--report',
)
self.assertIn(
'3 tests are expected to fail that we should fix',
result.stdout, result)
self.assertEqual(1, result.returncode, result)
@unittest.skip("not available with test processors")
def testWarnUnusedRules(self):
"""Test the unused-rules feature."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--variants=default,nooptimization',
'sweet',
'--warn-unused',
)
self.assertIn( 'Unused rule: carrots', result.stdout, result)
self.assertIn( 'Unused rule: regress/', result.stdout, result)
self.assertEqual(1, result.returncode, result)
@unittest.skip("not available with test processors")
def testCatNoSources(self):
"""Test printing sources, but the suite's tests have none available."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--variants=default',
'sweet/bananas',
'--cat',
)
self.assertIn('begin source: sweet/bananas', result.stdout, result)
self.assertIn('(no source available)', result.stdout, result)
self.assertEqual(0, result.returncode, result)
def testPredictable(self):
"""Test running a test in verify-predictable mode.
The test will fail because of missing allocation output. We verify that and
that the predictable flags are passed and printed after failure.
"""
with temp_base() as basedir:
override_build_config(basedir, v8_enable_verify_predictable=True)
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'sweet/bananas',
infra_staging=False,
)
self.assertIn('1 tests ran', result.stdout, result)
self.assertIn('sweet/bananas default: FAIL', result.stdout, result)
self.assertIn('Test had no allocation output', result.stdout, result)
self.assertIn('--predictable --verify-predictable', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def testSlowArch(self):
"""Test timeout factor manipulation on slow architecture."""
with temp_base() as basedir:
override_build_config(basedir, v8_target_cpu='arm64')
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'sweet/bananas',
)
# TODO(machenbach): We don't have a way for testing if the correct
# timeout was used.
self.assertEqual(0, result.returncode, result)
def testRandomSeedStressWithDefault(self):
"""Test using random-seed-stress feature has the right number of tests."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'--random-seed-stress-count=2',
'sweet/bananas',
infra_staging=False,
)
self.assertIn('2 tests ran', result.stdout, result)
self.assertEqual(0, result.returncode, result)
def testRandomSeedStressWithSeed(self):
"""Test using random-seed-stress feature passing a random seed."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'--random-seed-stress-count=2',
'--random-seed=123',
'sweet/strawberries',
)
self.assertIn('2 tests ran', result.stdout, result)
# We use a failing test so that the command is printed and we can verify
# that the right random seed was passed.
self.assertIn('--random-seed=123', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def testSpecificVariants(self):
"""Test using NO_VARIANTS modifiers in status files skips the desire tests.
The test runner cmd line configures 4 tests to run (2 tests * 2 variants).
But the status file applies a modifier to each skipping one of the
variants.
"""
with temp_base() as basedir:
override_build_config(basedir, is_asan=True)
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default,stress',
'sweet/bananas',
'sweet/raspberries',
)
# Both tests are either marked as running in only default or only
# slow variant.
self.assertIn('2 tests ran', result.stdout, result)
self.assertEqual(0, result.returncode, result)
def testStatusFilePresubmit(self):
"""Test that the fake status file is well-formed."""
with temp_base() as basedir:
from testrunner.local import statusfile
self.assertTrue(statusfile.PresubmitCheck(
os.path.join(basedir, 'test', 'sweet', 'sweet.status')))
def testDotsProgress(self):
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=dots',
'sweet/cherries',
'sweet/bananas',
'--no-sorting', '-j1', # make results order deterministic
infra_staging=False,
)
self.assertIn('2 tests ran', result.stdout, result)
self.assertIn('F.', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def testMonoProgress(self):
self._testCompactProgress('mono')
def testColorProgress(self):
self._testCompactProgress('color')
def _testCompactProgress(self, name):
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=%s' % name,
'sweet/cherries',
'sweet/bananas',
infra_staging=False,
)
if name == 'color':
expected = ('\033[34m% 28\033[0m|'
'\033[32m+ 1\033[0m|'
'\033[31m- 1\033[0m]: Done')
else:
expected = '% 28|+ 1|- 1]: Done'
self.assertIn(expected, result.stdout)
self.assertIn('sweet/cherries', result.stdout)
self.assertIn('sweet/bananas', result.stdout)
self.assertEqual(1, result.returncode, result)
def testExitAfterNFailures(self):
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--exit-after-n-failures=2',
'-j1',
'sweet/mangoes', # PASS
'sweet/strawberries', # FAIL
'sweet/blackberries', # FAIL
'sweet/raspberries', # should not run
)
self.assertIn('sweet/mangoes default: PASS', result.stdout, result)
self.assertIn('sweet/strawberries default: FAIL', result.stdout, result)
self.assertIn('Too many failures, exiting...', result.stdout, result)
self.assertIn('sweet/blackberries default: FAIL', result.stdout, result)
self.assertNotIn('sweet/raspberries', result.stdout, result)
self.assertIn('2 tests failed', result.stdout, result)
self.assertIn('3 tests ran', result.stdout, result)
self.assertEqual(1, result.returncode, result)
def testNumFuzzer(self):
sys_args = ['--command-prefix', sys.executable, '--outdir', 'out/build']
with temp_base() as basedir:
with capture() as (stdout, stderr):
code = num_fuzzer.NumFuzzer(basedir=basedir).execute(sys_args)
result = Result(stdout.getvalue(), stderr.getvalue(), code)
self.assertEqual(0, result.returncode, result)
def testRunnerFlags(self):
"""Test that runner-specific flags are passed to tests."""
with temp_base() as basedir:
result = run_tests(
basedir,
'--progress=verbose',
'--variants=default',
'--random-seed=42',
'sweet/bananas',
'-v',
)
self.assertIn(
'--test bananas --random-seed=42 --nohard-abort --testing-d8-test-runner',
result.stdout, result)
self.assertEqual(0, result.returncode, result)
if __name__ == '__main__':
unittest.main()