v8/tools/clusterfuzz/v8_foozzie_test.py

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

376 lines
13 KiB
Python
Raw Normal View History

#!/usr/bin/env python
# Copyright 2016 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.
import os
import random
import subprocess
import sys
import unittest
import v8_commands
import v8_foozzie
import v8_fuzz_config
import v8_suppressions
try:
basestring
except NameError:
basestring = str
PYTHON3 = sys.version_info >= (3, 0)
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
FOOZZIE = os.path.join(BASE_DIR, 'v8_foozzie.py')
TEST_DATA = os.path.join(BASE_DIR, 'testdata')
KNOWN_BUILDS = [
'd8',
'clang_x86/d8',
'clang_x86_v8_arm/d8',
'clang_x64_v8_arm64/d8',
'clang_x64_pointer_compression/d8',
]
class ConfigTest(unittest.TestCase):
def testExperiments(self):
"""Test integrity of probabilities and configs."""
CONFIGS = v8_foozzie.CONFIGS
EXPERIMENTS = v8_fuzz_config.FOOZZIE_EXPERIMENTS
FLAGS = v8_fuzz_config.ADDITIONAL_FLAGS
# Probabilities add up to 100%.
first_is_int = lambda x: type(x[0]) == int
assert all(map(first_is_int, EXPERIMENTS))
assert sum(x[0] for x in EXPERIMENTS) == 100
# Configs used in experiments are defined.
assert all(map(lambda x: x[1] in CONFIGS, EXPERIMENTS))
assert all(map(lambda x: x[2] in CONFIGS, EXPERIMENTS))
# The last config item points to a known build configuration.
assert all(map(lambda x: x[3] in KNOWN_BUILDS, EXPERIMENTS))
# All flags have a probability.
first_is_float = lambda x: type(x[0]) == float
assert all(map(first_is_float, FLAGS))
first_between_0_and_1 = lambda x: x[0] > 0 and x[0] < 1
assert all(map(first_between_0_and_1, FLAGS))
# Test consistent flags.
second_is_string = lambda x: isinstance(x[1], basestring)
assert all(map(second_is_string, FLAGS))
# We allow spaces to separate more flags. We don't allow spaces in the flag
# value.
is_flag = lambda x: x.startswith('--')
all_parts_are_flags = lambda x: all(map(is_flag, x[1].split()))
assert all(map(all_parts_are_flags, FLAGS))
def testConfig(self):
"""Smoke test how to choose experiments."""
config = v8_fuzz_config.Config('foo', random.Random(42))
experiments = [
[25, 'ignition', 'jitless', 'd8'],
[75, 'ignition', 'ignition', 'clang_x86/d8'],
]
flags = [
[0.1, '--flag'],
[0.3, '--baz'],
[0.3, '--foo --bar'],
]
self.assertEqual(
[
'--first-config=ignition',
'--second-config=jitless',
'--second-d8=d8',
'--second-config-extra-flags=--baz',
'--second-config-extra-flags=--foo',
'--second-config-extra-flags=--bar',
],
config.choose_foozzie_flags(experiments, flags),
)
self.assertEqual(
[
'--first-config=ignition',
'--second-config=jitless',
'--second-d8=d8',
],
config.choose_foozzie_flags(experiments, flags),
)
class UnitTest(unittest.TestCase):
def testCluster(self):
crash_test_example_path = 'CrashTests/path/to/file.js'
self.assertEqual(
v8_foozzie.ORIGINAL_SOURCE_DEFAULT,
v8_foozzie.cluster_failures(''))
self.assertEqual(
v8_foozzie.ORIGINAL_SOURCE_CRASHTESTS,
v8_foozzie.cluster_failures(crash_test_example_path))
self.assertEqual(
'_o_O_',
v8_foozzie.cluster_failures(
crash_test_example_path,
known_failures={crash_test_example_path: '_o_O_'}))
self.assertEqual(
'980',
v8_foozzie.cluster_failures('v8/test/mjsunit/apply.js'))
def testDiff(self):
def diff_fun(one, two, skip=False):
suppress = v8_suppressions.get_suppression(skip)
return suppress.diff_lines(one.splitlines(), two.splitlines())
one = ''
two = ''
diff = None, None
self.assertEqual(diff, diff_fun(one, two))
one = 'a \n b\nc();'
two = 'a \n b\nc();'
diff = None, None
self.assertEqual(diff, diff_fun(one, two))
# Ignore line before caret and caret position.
one = """
undefined
weird stuff
^
somefile.js: TypeError: suppressed message
undefined
"""
two = """
undefined
other weird stuff
^
somefile.js: TypeError: suppressed message
undefined
"""
diff = None, None
self.assertEqual(diff, diff_fun(one, two))
one = """
Still equal
Extra line
"""
two = """
Still equal
"""
diff = '- Extra line', None
self.assertEqual(diff, diff_fun(one, two))
one = """
Still equal
"""
two = """
Still equal
Extra line
"""
diff = '+ Extra line', None
self.assertEqual(diff, diff_fun(one, two))
one = """
undefined
somefile.js: TypeError: undefined is not a constructor
"""
two = """
undefined
otherfile.js: TypeError: undefined is not a constructor
"""
diff = """- somefile.js: TypeError: undefined is not a constructor
+ otherfile.js: TypeError: undefined is not a constructor""", None
self.assertEqual(diff, diff_fun(one, two))
# Test that skipping suppressions works.
one = """
v8-foozzie source: foo
weird stuff
^
"""
two = """
v8-foozzie source: foo
other weird stuff
^
"""
self.assertEqual((None, 'foo'), diff_fun(one, two))
diff = ('- ^\n+ ^', 'foo')
self.assertEqual(diff, diff_fun(one, two, skip=True))
def testOutputCapping(self):
def output(stdout, is_crash):
exit_code = -1 if is_crash else 0
return v8_commands.Output(exit_code=exit_code, stdout=stdout, pid=0)
def check(stdout1, stdout2, is_crash1, is_crash2, capped_lines1,
capped_lines2):
output1 = output(stdout1, is_crash1)
output2 = output(stdout2, is_crash2)
self.assertEqual(
(capped_lines1, capped_lines2),
v8_suppressions.get_output_capped(output1, output2))
# No capping, already equal.
check('1\n2', '1\n2', True, True, '1\n2', '1\n2')
# No crash, no capping.
check('1\n2', '1\n2\n3', False, False, '1\n2', '1\n2\n3')
check('1\n2\n3', '1\n2', False, False, '1\n2\n3', '1\n2')
# Cap smallest if all runs crash.
check('1\n2', '1\n2\n3', True, True, '1\n2', '1\n2')
check('1\n2\n3', '1\n2', True, True, '1\n2', '1\n2')
check('1\n2', '1\n23', True, True, '1\n2', '1\n2')
check('1\n23', '1\n2', True, True, '1\n2', '1\n2')
# Cap the non-crashy run.
check('1\n2\n3', '1\n2', False, True, '1\n2', '1\n2')
check('1\n2', '1\n2\n3', True, False, '1\n2', '1\n2')
check('1\n23', '1\n2', False, True, '1\n2', '1\n2')
check('1\n2', '1\n23', True, False, '1\n2', '1\n2')
# The crashy run has more output.
check('1\n2\n3', '1\n2', True, False, '1\n2\n3', '1\n2')
check('1\n2', '1\n2\n3', False, True, '1\n2', '1\n2\n3')
check('1\n23', '1\n2', True, False, '1\n23', '1\n2')
check('1\n2', '1\n23', False, True, '1\n2', '1\n23')
# Keep output difference when capping.
check('1\n2', '3\n4\n5', True, True, '1\n2', '3\n4')
check('1\n2\n3', '4\n5', True, True, '1\n2', '4\n5')
check('12', '345', True, True, '12', '34')
check('123', '45', True, True, '12', '45')
def cut_verbose_output(stdout, n_comp):
# This removes the first lines containing d8 commands of `n_comp` comparison
# runs.
return '\n'.join(stdout.split('\n')[n_comp * 2:])
def run_foozzie(second_d8_dir, *extra_flags, **kwargs):
second_config = 'ignition_turbo'
if 'second_config' in kwargs:
second_config = 'jitless'
kwargs = {}
if PYTHON3:
kwargs['text'] = True
return subprocess.check_output([
sys.executable, FOOZZIE,
'--random-seed', '12345',
'--first-d8', os.path.join(TEST_DATA, 'baseline', 'd8.py'),
'--second-d8', os.path.join(TEST_DATA, second_d8_dir, 'd8.py'),
'--first-config', 'ignition',
'--second-config', second_config,
os.path.join(TEST_DATA, 'fuzz-123.js'),
] + list(extra_flags), **kwargs)
class SystemTest(unittest.TestCase):
"""This tests the whole correctness-fuzzing harness with fake build
artifacts.
Overview of fakes:
baseline: Example foozzie output including a syntax error.
[foozzie] Prioritize reporting differences with x64 if present tldr: This adds an on-demand comparison with x64 when a difference to a non-x64 build is detected. Normally foozzie compares the baseline build (just ignition), with two secondary builds. One, the default, always uses the shipping configuration, the second passes additional flags. Both can use a different architecture than the baseline build as well. Differences between ignition and turbofan are then often detected independent of the architectures used, but reported several times (for each compared architecture). This makes the reporting more specific, by running another build on demand that uses the baseline architecture, but otherwise the same configuration that showed a difference. If it shows the difference as well, the baseline architecture is used for the report. As a result only pure architecture differences will be reported with an architecture other than x64. This also adds some minor refactorings to reduce the code complexity when looping over comparisons. For testing this, the fake-d8s are extended with different behavior for different flags passed. We add two test cases for testing: x64 vs. ia32 with difference in x64 and ia32 x64 vs. ia32 with difference only in ia32 Bug: chromium:1196633 No-Try: true Test: tools/clusterfuzz/v8_foozzie_test.py Change-Id: Ic470ae8f0b37fb1628b32e4fafc0c39377e16f8c Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2897099 Reviewed-by: Liviu Rau <liviurau@chromium.org> Commit-Queue: Michael Achenbach <machenbach@chromium.org> Cr-Commit-Position: refs/heads/master@{#74612}
2021-05-17 13:29:26 +00:00
build1: Difference to baseline is a stack trace difference expected to
be suppressed.
build2: Difference to baseline is a non-suppressed output difference
causing the script to fail.
build3: As build1 but with an architecture difference as well.
"""
def testSyntaxErrorDiffPass(self):
stdout = run_foozzie('build1', '--skip-smoke-tests')
self.assertEqual('# V8 correctness - pass\n',
cut_verbose_output(stdout, 3))
# Default comparison includes suppressions.
self.assertIn('v8_suppressions.js', stdout)
# Default comparison doesn't include any specific mock files.
self.assertNotIn('v8_mock_archs.js', stdout)
self.assertNotIn('v8_mock_webassembly.js', stdout)
def testDifferentOutputFail(self):
with open(os.path.join(TEST_DATA, 'failure_output.txt')) as f:
expected_output = f.read()
with self.assertRaises(subprocess.CalledProcessError) as ctx:
run_foozzie('build2', '--skip-smoke-tests',
'--first-config-extra-flags=--flag1',
'--first-config-extra-flags=--flag2=0',
'--second-config-extra-flags=--flag3')
e = ctx.exception
self.assertEqual(v8_foozzie.RETURN_FAIL, e.returncode)
self.assertEqual(expected_output, cut_verbose_output(e.output, 2))
def testSmokeTest(self):
with open(os.path.join(TEST_DATA, 'smoke_test_output.txt')) as f:
expected_output = f.read()
with self.assertRaises(subprocess.CalledProcessError) as ctx:
run_foozzie('build2')
e = ctx.exception
self.assertEqual(v8_foozzie.RETURN_FAIL, e.returncode)
self.assertEqual(expected_output, e.output)
def testDifferentArch(self):
"""Test that the architecture-specific mocks are passed to both runs when
we use executables with different architectures.
"""
# Build 3 simulates x86, while the baseline is x64.
stdout = run_foozzie('build3', '--skip-smoke-tests')
lines = stdout.split('\n')
# TODO(machenbach): Don't depend on the command-lines being printed in
# particular lines.
self.assertIn('v8_mock_archs.js', lines[1])
self.assertIn('v8_mock_archs.js', lines[3])
[foozzie] Prioritize reporting differences with x64 if present tldr: This adds an on-demand comparison with x64 when a difference to a non-x64 build is detected. Normally foozzie compares the baseline build (just ignition), with two secondary builds. One, the default, always uses the shipping configuration, the second passes additional flags. Both can use a different architecture than the baseline build as well. Differences between ignition and turbofan are then often detected independent of the architectures used, but reported several times (for each compared architecture). This makes the reporting more specific, by running another build on demand that uses the baseline architecture, but otherwise the same configuration that showed a difference. If it shows the difference as well, the baseline architecture is used for the report. As a result only pure architecture differences will be reported with an architecture other than x64. This also adds some minor refactorings to reduce the code complexity when looping over comparisons. For testing this, the fake-d8s are extended with different behavior for different flags passed. We add two test cases for testing: x64 vs. ia32 with difference in x64 and ia32 x64 vs. ia32 with difference only in ia32 Bug: chromium:1196633 No-Try: true Test: tools/clusterfuzz/v8_foozzie_test.py Change-Id: Ic470ae8f0b37fb1628b32e4fafc0c39377e16f8c Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2897099 Reviewed-by: Liviu Rau <liviurau@chromium.org> Commit-Queue: Michael Achenbach <machenbach@chromium.org> Cr-Commit-Position: refs/heads/master@{#74612}
2021-05-17 13:29:26 +00:00
def testDifferentArchFailFirst(self):
"""Test that we re-test against x64. This tests the path that also fails
on x64 and then reports the error as x64.
"""
with open(os.path.join(TEST_DATA, 'failure_output_arch.txt')) as f:
expected_output = f.read()
# Build 3 simulates x86 and produces a difference on --bad-flag, but
# the baseline build shows the same difference when --bad-flag is passed.
with self.assertRaises(subprocess.CalledProcessError) as ctx:
run_foozzie('build3', '--skip-smoke-tests',
'--second-config-extra-flags=--bad-flag')
e = ctx.exception
self.assertEqual(v8_foozzie.RETURN_FAIL, e.returncode)
self.assertEqual(expected_output, cut_verbose_output(e.output, 3))
def testDifferentArchFailSecond(self):
"""As above, but we test the path that only fails in the second (ia32)
run and not with x64 and then reports the error as ia32.
"""
with open(os.path.join(TEST_DATA, 'failure_output_second.txt')) as f:
expected_output = f.read()
# Build 3 simulates x86 and produces a difference on --very-bad-flag,
# which the baseline build doesn't.
with self.assertRaises(subprocess.CalledProcessError) as ctx:
run_foozzie('build3', '--skip-smoke-tests',
'--second-config-extra-flags=--very-bad-flag')
e = ctx.exception
self.assertEqual(v8_foozzie.RETURN_FAIL, e.returncode)
self.assertEqual(expected_output, cut_verbose_output(e.output, 3))
def testJitless(self):
"""Test that webassembly is mocked out when comparing with jitless."""
stdout = run_foozzie(
'build1', '--skip-smoke-tests', second_config='jitless')
lines = stdout.split('\n')
# TODO(machenbach): Don't depend on the command-lines being printed in
# particular lines.
self.assertIn('v8_mock_webassembly.js', lines[1])
self.assertIn('v8_mock_webassembly.js', lines[3])
def testSkipSuppressions(self):
"""Test that the suppressions file is not passed when skipping
suppressions.
"""
# Compare baseline with baseline. This passes as there is no difference.
stdout = run_foozzie(
'baseline', '--skip-smoke-tests', '--skip-suppressions')
self.assertNotIn('v8_suppressions.js', stdout)
# Compare with a build that usually suppresses a difference. Now we fail
# since we skip suppressions.
with self.assertRaises(subprocess.CalledProcessError) as ctx:
run_foozzie(
'build1', '--skip-smoke-tests', '--skip-suppressions')
e = ctx.exception
self.assertEqual(v8_foozzie.RETURN_FAIL, e.returncode)
self.assertNotIn('v8_suppressions.js', e.output)
if __name__ == '__main__':
unittest.main()