v8/tools/testrunner/standard_runner.py
Eric Holk 4d4a3294b9 [test] Reverse sense of wasm_traps variant
D8 enables the Wasm trap handler by default now, but we need to make sure the
older bounds check case still gets test coverage too, as bounds checks will
continue to be a supported configuration.

Cq-Include-Trybots: luci.v8.try:v8_linux_noi18n_rel_ng
Change-Id: I5b0bdded6929a9b3a8480e87d038398b8d2a0fd8
Reviewed-on: https://chromium-review.googlesource.com/1048835
Reviewed-by: Michael Achenbach <machenbach@chromium.org>
Commit-Queue: Eric Holk <eholk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#53078}
2018-05-08 17:49:14 +00:00

353 lines
13 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.
import os
import re
import sys
# Adds testrunner to the path hence it has to be imported at the beggining.
import base_runner
from testrunner.local import utils
from testrunner.local.variants import ALL_VARIANTS
from testrunner.objects import predictable
from testrunner.testproc.execution import ExecutionProc
from testrunner.testproc.filter import StatusFileFilterProc, NameFilterProc
from testrunner.testproc.loader import LoadProc
from testrunner.testproc.progress import ResultsTracker, TestsCounter
from testrunner.testproc.seed import SeedProc
from testrunner.testproc.variant import VariantProc
from testrunner.utils import random_utils
ARCH_GUESS = utils.DefaultArch()
VARIANTS = ["default"]
MORE_VARIANTS = [
"nooptimization",
"stress",
"stress_background_compile",
"stress_incremental_marking",
]
VARIANT_ALIASES = {
# The default for developer workstations.
"dev": VARIANTS,
# Additional variants, run on all bots.
"more": MORE_VARIANTS,
# Shortcut for the two above ("more" first - it has the longer running tests).
"exhaustive": MORE_VARIANTS + VARIANTS,
# Additional variants, run on a subset of bots.
"extra": ["future", "liftoff", "no_wasm_traps", "trusted"],
}
GC_STRESS_FLAGS = ["--gc-interval=500", "--stress-compaction",
"--concurrent-recompilation-queue-length=64",
"--concurrent-recompilation-delay=500",
"--concurrent-recompilation"]
RANDOM_GC_STRESS_FLAGS = ["--random-gc-interval=5000",
"--stress-compaction-random"]
PREDICTABLE_WRAPPER = os.path.join(
base_runner.BASE_DIR, 'tools', 'predictable_wrapper.py')
class StandardTestRunner(base_runner.BaseTestRunner):
def __init__(self, *args, **kwargs):
super(StandardTestRunner, self).__init__(*args, **kwargs)
self.sancov_dir = None
self._variants = None
def _get_default_suite_names(self):
return ['default']
def _add_parser_options(self, parser):
parser.add_option("--novfp3",
help="Indicates that V8 was compiled without VFP3"
" support",
default=False, action="store_true")
# Variants
parser.add_option("--no-variants", "--novariants",
help="Deprecated. "
"Equivalent to passing --variants=default",
default=False, dest="no_variants", action="store_true")
parser.add_option("--variants",
help="Comma-separated list of testing variants;"
" default: \"%s\"" % ",".join(VARIANTS))
parser.add_option("--exhaustive-variants",
default=False, action="store_true",
help="Deprecated. "
"Equivalent to passing --variants=exhaustive")
# Filters
parser.add_option("--slow-tests", default="dontcare",
help="Regard slow tests (run|skip|dontcare)")
parser.add_option("--pass-fail-tests", default="dontcare",
help="Regard pass|fail tests (run|skip|dontcare)")
parser.add_option("--quickcheck", default=False, action="store_true",
help=("Quick check mode (skip slow tests)"))
parser.add_option("--dont-skip-slow-simulator-tests",
help="Don't skip more slow tests when using a"
" simulator.",
default=False, action="store_true",
dest="dont_skip_simulator_slow_tests")
# Stress modes
parser.add_option("--gc-stress",
help="Switch on GC stress mode",
default=False, action="store_true")
parser.add_option("--random-gc-stress",
help="Switch on random GC stress mode",
default=False, action="store_true")
parser.add_option("--random-seed-stress-count", default=1, type="int",
dest="random_seed_stress_count",
help="Number of runs with different random seeds. Only "
"with test processors: 0 means infinite "
"generation.")
# Noop
parser.add_option("--cfi-vptr",
help="Run tests with UBSAN cfi_vptr option.",
default=False, action="store_true")
parser.add_option("--infra-staging", help="Use new test runner features",
dest='infra_staging', default=None,
action="store_true")
parser.add_option("--no-infra-staging",
help="Opt out of new test runner features",
dest='infra_staging', default=None,
action="store_false")
parser.add_option("--no-sorting", "--nosorting",
help="Don't sort tests according to duration of last"
" run.",
default=False, dest="no_sorting", action="store_true")
parser.add_option("--no-presubmit", "--nopresubmit",
help='Skip presubmit checks (deprecated)',
default=False, dest="no_presubmit", action="store_true")
# Unimplemented for test processors
parser.add_option("--sancov-dir",
help="Directory where to collect coverage data")
parser.add_option("--cat", help="Print the source of the tests",
default=False, action="store_true")
parser.add_option("--flakiness-results",
help="Path to a file for storing flakiness json.")
parser.add_option("--time", help="Print timing information after running",
default=False, action="store_true")
parser.add_option("--warn-unused", help="Report unused rules",
default=False, action="store_true")
parser.add_option("--report", default=False, action="store_true",
help="Print a summary of the tests to be run")
def _process_options(self, options):
if options.sancov_dir:
self.sancov_dir = options.sancov_dir
if not os.path.exists(self.sancov_dir):
print("sancov-dir %s doesn't exist" % self.sancov_dir)
raise base_runner.TestRunnerError()
if options.gc_stress:
options.extra_flags += GC_STRESS_FLAGS
if options.random_gc_stress:
options.extra_flags += RANDOM_GC_STRESS_FLAGS
if self.build_config.asan:
options.extra_flags.append("--invoke-weak-callbacks")
options.extra_flags.append("--omit-quit")
if options.novfp3:
options.extra_flags.append("--noenable-vfp3")
if options.no_variants: # pragma: no cover
print ("Option --no-variants is deprecated. "
"Pass --variants=default instead.")
assert not options.variants
options.variants = "default"
if options.exhaustive_variants: # pragma: no cover
# TODO(machenbach): Switch infra to --variants=exhaustive after M65.
print ("Option --exhaustive-variants is deprecated. "
"Pass --variants=exhaustive instead.")
# This is used on many bots. It includes a larger set of default
# variants.
# Other options for manipulating variants still apply afterwards.
assert not options.variants
options.variants = "exhaustive"
if options.quickcheck:
assert not options.variants
options.variants = "stress,default"
options.slow_tests = "skip"
options.pass_fail_tests = "skip"
if self.build_config.predictable:
options.variants = "default"
options.extra_flags.append("--predictable")
options.extra_flags.append("--verify_predictable")
options.extra_flags.append("--no-inline-new")
# Add predictable wrapper to command prefix.
options.command_prefix = (
[sys.executable, PREDICTABLE_WRAPPER] + options.command_prefix)
# TODO(machenbach): Figure out how to test a bigger subset of variants on
# msan.
if self.build_config.msan:
options.variants = "default"
if options.variants == "infra_staging":
options.variants = "exhaustive"
self._variants = self._parse_variants(options.variants)
def CheckTestMode(name, option): # pragma: no cover
if not option in ["run", "skip", "dontcare"]:
print "Unknown %s mode %s" % (name, option)
raise base_runner.TestRunnerError()
CheckTestMode("slow test", options.slow_tests)
CheckTestMode("pass|fail test", options.pass_fail_tests)
if self.build_config.no_i18n:
base_runner.TEST_MAP["bot_default"].remove("intl")
base_runner.TEST_MAP["default"].remove("intl")
# TODO(machenbach): uncomment after infra side lands.
# base_runner.TEST_MAP["d8_default"].remove("intl")
def _parse_variants(self, aliases_str):
# Use developer defaults if no variant was specified.
aliases_str = aliases_str or 'dev'
aliases = aliases_str.split(',')
user_variants = set(reduce(
list.__add__, [VARIANT_ALIASES.get(a, [a]) for a in aliases]))
result = [v for v in ALL_VARIANTS if v in user_variants]
if len(result) == len(user_variants):
return result
for v in user_variants:
if v not in ALL_VARIANTS:
print 'Unknown variant: %s' % v
raise base_runner.TestRunnerError()
assert False, 'Unreachable'
def _setup_env(self):
super(StandardTestRunner, self)._setup_env()
symbolizer_option = self._get_external_symbolizer_option()
if self.sancov_dir:
os.environ['ASAN_OPTIONS'] = ":".join([
'coverage=1',
'coverage_dir=%s' % self.sancov_dir,
symbolizer_option,
"allow_user_segv_handler=1",
])
def _get_statusfile_variables(self, options):
variables = (
super(StandardTestRunner, self)._get_statusfile_variables(options))
simulator_run = (
not options.dont_skip_simulator_slow_tests and
self.build_config.arch in [
'arm64', 'arm', 'mipsel', 'mips', 'mips64', 'mips64el', 'ppc',
'ppc64', 's390', 's390x'] and
bool(ARCH_GUESS) and
self.build_config.arch != ARCH_GUESS)
variables.update({
'gc_stress': options.gc_stress or options.random_gc_stress,
'gc_fuzzer': options.random_gc_stress,
'novfp3': options.novfp3,
'simulator_run': simulator_run,
})
return variables
def _do_execute(self, tests, args, options):
jobs = options.j
print '>>> Running with test processors'
loader = LoadProc()
tests_counter = TestsCounter()
results = ResultsTracker()
indicators = self._create_progress_indicators(options)
outproc_factory = None
if self.build_config.predictable:
outproc_factory = predictable.get_outproc
execproc = ExecutionProc(jobs, outproc_factory)
sigproc = self._create_signal_proc()
procs = [
loader,
NameFilterProc(args) if args else None,
StatusFileFilterProc(options.slow_tests, options.pass_fail_tests),
self._create_shard_proc(options),
tests_counter,
VariantProc(self._variants),
StatusFileFilterProc(options.slow_tests, options.pass_fail_tests),
self._create_predictable_filter(),
self._create_seed_proc(options),
sigproc,
] + indicators + [
results,
self._create_timeout_proc(options),
self._create_rerun_proc(options),
execproc,
]
self._prepare_procs(procs)
tests.sort(key=lambda t: t.is_slow, reverse=True)
loader.load_tests(tests)
print '>>> Running %d base tests' % tests_counter.total
tests_counter.remove_from_chain()
# This starts up worker processes and blocks until all tests are
# processed.
execproc.run()
for indicator in indicators:
indicator.finished()
print '>>> %d tests ran' % (results.total - results.remaining)
exit_code = utils.EXIT_CODE_PASS
if results.failed:
exit_code = utils.EXIT_CODE_FAILURES
if not results.total:
exit_code = utils.EXIT_CODE_NO_TESTS
# Indicate if a SIGINT or SIGTERM happened.
exit_code = max(exit_code, sigproc.exit_code)
if exit_code == utils.EXIT_CODE_FAILURES and options.json_test_results:
print("Force exit code 0 after failures. Json test results file "
"generated with failure information.")
exit_code = utils.EXIT_CODE_PASS
return exit_code
def _create_predictable_filter(self):
if not self.build_config.predictable:
return None
return predictable.PredictableFilterProc()
def _create_seed_proc(self, options):
if options.random_seed_stress_count == 1:
return None
return SeedProc(options.random_seed_stress_count, options.random_seed,
options.j * 4)
if __name__ == '__main__':
sys.exit(StandardTestRunner().execute())