blob: ce317d80517723b52b21768aca180361a08011c2 [file] [log] [blame]
#!/usr/bin/env 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 print_function, unicode_literals
import math, os, platform, posixpath, shlex, shutil, subprocess, sys, traceback
def add_libdir_to_path():
from os.path import dirname, exists, join, realpath
js_src_dir = dirname(dirname(realpath(sys.argv[0])))
assert exists(join(js_src_dir, 'jsapi.h'))
sys.path.insert(0, join(js_src_dir, 'lib'))
sys.path.insert(0, join(js_src_dir, 'tests', 'lib'))
add_libdir_to_path()
import jittests
from tests import get_jitflags, get_cpu_count, get_environment_overlay, \
change_env
# Python 3.3 added shutil.which, but we can't use that yet.
def which(name):
if name.find(os.path.sep) != -1:
return os.path.abspath(name)
for path in os.environ["PATH"].split(os.pathsep):
full = os.path.join(path, name)
if os.path.exists(full):
return os.path.abspath(full)
return name
def main(argv):
# The [TESTS] optional arguments are paths of test files relative
# to the jit-test/tests directory.
from optparse import OptionParser
op = OptionParser(usage='%prog [options] JS_SHELL [TESTS]')
op.add_option('-s', '--show-cmd', dest='show_cmd', action='store_true',
help='show js shell command run')
op.add_option('-f', '--show-failed-cmd', dest='show_failed',
action='store_true',
help='show command lines of failed tests')
op.add_option('-o', '--show-output', dest='show_output',
action='store_true',
help='show output from js shell')
op.add_option('-F', '--failed-only', dest='failed_only',
action='store_true',
help="if --show-output is given, only print output for"
" failed tests")
op.add_option('--no-show-failed', dest='no_show_failed',
action='store_true',
help="don't print output for failed tests"
" (no-op with --show-output)")
op.add_option('-x', '--exclude', dest='exclude', action='append',
help='exclude given test dir or path')
op.add_option('--slow', dest='run_slow', action='store_true',
help='also run tests marked as slow')
op.add_option('--no-slow', dest='run_slow', action='store_false',
help='do not run tests marked as slow (the default)')
op.add_option('-t', '--timeout', dest='timeout', type=float, default=150.0,
help='set test timeout in seconds')
op.add_option('--no-progress', dest='hide_progress', action='store_true',
help='hide progress bar')
op.add_option('--tinderbox', dest='format', action='store_const',
const='automation',
help='Use automation-parseable output format')
op.add_option('--format', dest='format', default='none', type='choice',
choices=['automation', 'none'],
help='Output format. Either automation or none'
' (default %default).')
op.add_option('--args', dest='shell_args', default='',
help='extra args to pass to the JS shell')
op.add_option('-w', '--write-failures', dest='write_failures',
metavar='FILE',
help='Write a list of failed tests to [FILE]')
op.add_option('-r', '--read-tests', dest='read_tests', metavar='FILE',
help='Run test files listed in [FILE]')
op.add_option('-R', '--retest', dest='retest', metavar='FILE',
help='Retest using test list file [FILE]')
op.add_option('-g', '--debug', action='store_const', const='gdb', dest='debugger',
help='Run a single test under the gdb debugger')
op.add_option('-G', '--debug-rr', action='store_const', const='rr', dest='debugger',
help='Run a single test under the rr debugger')
op.add_option('--debugger', type='string',
help='Run a single test under the specified debugger')
op.add_option('--valgrind', dest='valgrind', action='store_true',
help='Enable the |valgrind| flag, if valgrind is in $PATH.')
op.add_option('--valgrind-all', dest='valgrind_all', action='store_true',
help='Run all tests with valgrind, if valgrind is in $PATH.')
op.add_option('--jitflags', dest='jitflags', default='none', type='string',
help='IonMonkey option combinations. One of all, debug,'
' ion, and none (default %default).')
op.add_option('--avoid-stdio', dest='avoid_stdio', action='store_true',
help='Use js-shell file indirection instead of piping stdio.')
op.add_option('--write-failure-output', dest='write_failure_output',
action='store_true',
help='With --write-failures=FILE, additionally write the'
' output of failed tests to [FILE]')
op.add_option('--ion', dest='ion', action='store_true',
help='Run tests once with --ion-eager and once with'
' --baseline-eager (ignores --jitflags)')
op.add_option('--tbpl', dest='tbpl', action='store_true',
help='Run tests with all IonMonkey option combinations'
' (ignores --jitflags)')
op.add_option('-j', '--worker-count', dest='max_jobs', type=int,
default=max(1, get_cpu_count()),
help='Number of tests to run in parallel (default %default)')
op.add_option('--remote', action='store_true',
help='Run tests on a remote device')
op.add_option('--deviceIP', action='store',
type='string', dest='device_ip',
help='IP address of remote device to test')
op.add_option('--devicePort', action='store',
type=int, dest='device_port', default=20701,
help='port of remote device to test')
op.add_option('--deviceSerial', action='store',
type='string', dest='device_serial', default=None,
help='ADB device serial number of remote device to test')
op.add_option('--deviceTransport', action='store',
type='string', dest='device_transport', default='sut',
help='The transport to use to communicate with device:'
' [adb|sut]; default=sut')
op.add_option('--remoteTestRoot', dest='remote_test_root', action='store',
type='string', default='/data/local/tests',
help='The remote directory to use as test root'
' (eg. /data/local/tests)')
op.add_option('--localLib', dest='local_lib', action='store',
type='string',
help='The location of libraries to push -- preferably'
' stripped')
op.add_option('--repeat', type=int, default=1,
help='Repeat tests the given number of times.')
op.add_option('--this-chunk', type=int, default=1,
help='The test chunk to run.')
op.add_option('--total-chunks', type=int, default=1,
help='The total number of test chunks.')
op.add_option('--ignore-timeouts', dest='ignore_timeouts', metavar='FILE',
help='Ignore timeouts of tests listed in [FILE]')
op.add_option('--test-reflect-stringify', dest="test_reflect_stringify",
help="instead of running tests, use them to test the "
"Reflect.stringify code in specified file")
options, args = op.parse_args(argv)
if len(args) < 1:
op.error('missing JS_SHELL argument')
js_shell = which(args[0])
test_args = args[1:]
test_environment = get_environment_overlay(js_shell)
if not (os.path.isfile(js_shell) and os.access(js_shell, os.X_OK)):
if (platform.system() != 'Windows' or
os.path.isfile(js_shell) or not
os.path.isfile(js_shell + ".exe") or not
os.access(js_shell + ".exe", os.X_OK)):
op.error('shell is not executable: ' + js_shell)
if jittests.stdio_might_be_broken():
# Prefer erring on the side of caution and not using stdio if
# it might be broken on this platform. The file-redirect
# fallback should work on any platform, so at worst by
# guessing wrong we might have slowed down the tests a bit.
#
# XXX technically we could check for broken stdio, but it
# really seems like overkill.
options.avoid_stdio = True
if options.retest:
options.read_tests = options.retest
options.write_failures = options.retest
test_list = []
read_all = True
# Forbid running several variants of the same asmjs test, when debugging.
options.can_test_also_noasmjs = not options.debugger
if test_args:
read_all = False
for arg in test_args:
test_list += jittests.find_tests(arg)
if options.read_tests:
read_all = False
try:
f = open(options.read_tests)
for line in f:
test_list.append(os.path.join(jittests.TEST_DIR,
line.strip('\n')))
f.close()
except IOError:
if options.retest:
read_all = True
else:
sys.stderr.write("Exception thrown trying to read test file"
" '{}'\n".format(options.read_tests))
traceback.print_exc()
sys.stderr.write('---\n')
if read_all:
test_list = jittests.find_tests()
if options.exclude:
exclude_list = []
for exclude in options.exclude:
exclude_list += jittests.find_tests(exclude)
test_list = [test for test in test_list
if test not in set(exclude_list)]
if not test_list:
print("No tests found matching command line arguments.",
file=sys.stderr)
sys.exit(0)
test_list = [jittests.JitTest.from_file(_, options) for _ in test_list]
if not options.run_slow:
test_list = [_ for _ in test_list if not _.slow]
if options.test_reflect_stringify is not None:
for test in test_list:
test.test_reflect_stringify = options.test_reflect_stringify
# If chunking is enabled, determine which tests are part of this chunk.
# This code was adapted from testing/mochitest/runtestsremote.py.
if options.total_chunks > 1:
total_tests = len(test_list)
tests_per_chunk = math.ceil(total_tests / float(options.total_chunks))
start = int(round((options.this_chunk - 1) * tests_per_chunk))
end = int(round(options.this_chunk * tests_per_chunk))
test_list = test_list[start:end]
# The full test list is ready. Now create copies for each JIT configuration.
job_list = []
test_flags = []
if options.tbpl:
# Running all bits would take forever. Instead, we test a few
# interesting combinations.
test_flags = get_jitflags('all')
elif options.ion:
test_flags = get_jitflags('ion')
else:
test_flags = get_jitflags(options.jitflags)
job_list = [_ for test in test_list
for _ in test.copy_variants(test_flags)]
if options.ignore_timeouts:
read_all = False
try:
with open(options.ignore_timeouts) as f:
options.ignore_timeouts = set(
[line.strip('\n') for line in f.readlines()])
except IOError:
sys.exit("Error reading file: " + options.ignore_timeouts)
else:
options.ignore_timeouts = set()
prefix = [js_shell] + shlex.split(options.shell_args)
prologue = os.path.join(jittests.LIB_DIR, 'prologue.js')
if options.remote:
prologue = posixpath.join(options.remote_test_root,
'jit-tests', 'jit-tests', 'lib', 'prologue.js')
prefix += ['-f', prologue]
# Clean up any remnants from previous crashes etc
shutil.rmtree(jittests.JS_CACHE_DIR, ignore_errors=True)
os.mkdir(jittests.JS_CACHE_DIR)
if options.debugger:
if len(job_list) > 1:
print('Multiple tests match command line'
' arguments, debugger can only run one')
for tc in job_list:
print(' {}'.format(tc.path))
sys.exit(1)
tc = job_list[0]
if options.debugger == 'gdb':
debug_cmd = ['gdb', '--args']
elif options.debugger == 'lldb':
debug_cmd = ['lldb', '--']
elif options.debugger == 'rr':
debug_cmd = ['rr', 'record']
else:
debug_cmd = options.debugger.split()
with change_env(test_environment):
subprocess.call(debug_cmd + tc.command(prefix, jittests.LIB_DIR, jittests.MODULE_DIR))
if options.debugger == 'rr':
subprocess.call(['rr', 'replay'])
sys.exit()
try:
ok = None
if options.remote:
ok = jittests.run_tests_remote(job_list, prefix, options)
else:
with change_env(test_environment):
ok = jittests.run_tests(job_list, prefix, options)
if not ok:
sys.exit(2)
except OSError:
if not os.path.exists(prefix[0]):
print("JS shell argument: file does not exist:"
" '{}'".format(prefix[0]), file=sys.stderr)
sys.exit(1)
else:
raise
if __name__ == '__main__':
main(sys.argv[1:])