# 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/. """ Runs the Mochitest test harness. """ import os import sys SCRIPT_DIR = os.path.abspath(os.path.realpath(os.path.dirname(__file__))) sys.path.insert(0, SCRIPT_DIR) import ctypes import glob import json import numbers import platform import re import shlex import shutil import signal import socket import subprocess import sys import tempfile import time import traceback import uuid import zipfile from argparse import Namespace from collections import defaultdict from contextlib import closing from ctypes.util import find_library from datetime import datetime, timedelta from distutils import spawn import bisection import mozcrash import mozdebug import mozinfo import mozprocess import mozrunner from manifestparser import TestManifest from manifestparser.filters import ( chunk_by_dir, chunk_by_runtime, chunk_by_slice, failures, pathprefix, subsuite, tags, ) from manifestparser.util import normsep from mozgeckoprofiler import symbolicate_profile_json, view_gecko_profile from mozserve import DoHServer, Http2Server, Http3Server try: from marionette_driver.addons import Addons from marionette_driver.marionette import Marionette except ImportError as e: # noqa # Defer ImportError until attempt to use Marionette def reraise(*args, **kwargs): raise (e) # noqa Marionette = reraise import mozleak from leaks import LSANLeaks, ShutdownLeaks from mochitest_options import ( MochitestArgumentParser, build_obj, get_default_valgrind_suppression_files, ) from mozlog import commandline, get_proxy_logger from mozprofile import Profile from mozprofile.cli import KeyValueParseError, parse_key_value, parse_preferences from mozprofile.permissions import ServerLocations from mozrunner.utils import get_stack_fixer_function, test_environment from mozscreenshot import dump_screen HAVE_PSUTIL = False try: import psutil HAVE_PSUTIL = True except ImportError: pass import six from six.moves.urllib.parse import quote_plus as encodeURIComponent from six.moves.urllib_request import urlopen try: from mozbuild.base import MozbuildObject build = MozbuildObject.from_environment(cwd=SCRIPT_DIR) except ImportError: build = None here = os.path.abspath(os.path.dirname(__file__)) NO_TESTS_FOUND = """ No tests were found for flavor '{}' and the following manifest filters: {} Make sure the test paths (if any) are spelt correctly and the corresponding --flavor and --subsuite are being used. See `mach mochitest --help` for a list of valid flavors. """.lstrip() ######################################## # Option for MOZ (former NSPR) logging # ######################################## # Set the desired log modules you want a log be produced # by a try run for, or leave blank to disable the feature. # This will be passed to MOZ_LOG environment variable. # Try run will then put a download link for a zip archive # of all the log files on treeherder. MOZ_LOG = "" ######################################## # Option for web server log # ######################################## # If True, debug logging from the web server will be # written to mochitest-server-%d.txt artifacts on # treeherder. MOCHITEST_SERVER_LOGGING = False ##################### # Test log handling # ##################### # output processing TBPL_RETRY = 4 # Defined in mozharness class MessageLogger(object): """File-like object for logging messages (structured logs)""" BUFFERING_THRESHOLD = 100 # This is a delimiter used by the JS side to avoid logs interleaving DELIMITER = "\ue175\uee31\u2c32\uacbf" BUFFERED_ACTIONS = set(["test_status", "log"]) VALID_ACTIONS = set( [ "suite_start", "suite_end", "group_start", "group_end", "test_start", "test_end", "test_status", "log", "assertion_count", "buffering_on", "buffering_off", ] ) # Regexes that will be replaced with an empty string if found in a test # name. We do this to normalize test names which may contain URLs and test # package prefixes. TEST_PATH_PREFIXES = [ r"^/tests/", r"^\w+://[\w\.]+(:\d+)?(/\w+)?/(tests?|a11y|chrome)/", r"^\w+://[\w\.]+(:\d+)?(/\w+)?/(tests?|browser)/", ] def __init__(self, logger, buffering=True, structured=True): self.logger = logger self.structured = structured self.gecko_id = "GECKO" self.is_test_running = False self._manifest = None # Even if buffering is enabled, we only want to buffer messages between # TEST-START/TEST-END. So it is off to begin, but will be enabled after # a TEST-START comes in. self._buffering = False self.restore_buffering = buffering # Guard to ensure we never buffer if this value was initially `False` self._buffering_initially_enabled = buffering # Message buffering self.buffered_messages = [] def setManifest(self, name): self._manifest = name def validate(self, obj): """Tests whether the given object is a valid structured message (only does a superficial validation)""" if not ( isinstance(obj, dict) and "action" in obj and obj["action"] in MessageLogger.VALID_ACTIONS ): raise ValueError def _fix_subtest_name(self, message): """Make sure subtest name is a string""" if "subtest" in message and not isinstance( message["subtest"], six.string_types ): message["subtest"] = str(message["subtest"]) def _fix_test_name(self, message): """Normalize a logged test path to match the relative path from the sourcedir.""" if message.get("test") is not None: test = message["test"] for pattern in MessageLogger.TEST_PATH_PREFIXES: test = re.sub(pattern, "", test) if test != message["test"]: message["test"] = test break def _fix_message_format(self, message): if "message" in message: if isinstance(message["message"], bytes): message["message"] = message["message"].decode("utf-8", "replace") elif not isinstance(message["message"], six.text_type): message["message"] = six.text_type(message["message"]) def parse_line(self, line): """Takes a given line of input (structured or not) and returns a list of structured messages""" if isinstance(line, six.binary_type): # if line is a sequence of bytes, let's decode it line = line.rstrip().decode("UTF-8", "replace") else: # line is in unicode - so let's use it as it is line = line.rstrip() messages = [] for fragment in line.split(MessageLogger.DELIMITER): if not fragment: continue try: message = json.loads(fragment) self.validate(message) except ValueError: if self.structured: message = dict( action="process_output", process=self.gecko_id, data=fragment, ) else: message = dict( action="log", level="info", message=fragment, ) self._fix_subtest_name(message) self._fix_test_name(message) self._fix_message_format(message) message["group"] = self._manifest messages.append(message) return messages @property def buffering(self): if not self._buffering_initially_enabled: return False return self._buffering @buffering.setter def buffering(self, val): self._buffering = val def process_message(self, message): """Processes a structured message. Takes into account buffering, errors, ...""" # Activation/deactivating message buffering from the JS side if message["action"] == "buffering_on": if self.is_test_running: self.buffering = True return if message["action"] == "buffering_off": self.buffering = False return # Error detection also supports "raw" errors (in log messages) because some tests # manually dump 'TEST-UNEXPECTED-FAIL'. if "expected" in message or ( message["action"] == "log" and message.get("message", "").startswith("TEST-UNEXPECTED") ): self.restore_buffering = self.restore_buffering or self.buffering self.buffering = False if self.buffered_messages: snipped = len(self.buffered_messages) - self.BUFFERING_THRESHOLD if snipped > 0: self.logger.info( "".format(snipped) ) # Dumping previously buffered messages self.dump_buffered(limit=True) # Logging the error message self.logger.log_raw(message) # Determine if message should be buffered elif ( self.buffering and self.structured and message["action"] in self.BUFFERED_ACTIONS ): self.buffered_messages.append(message) # Otherwise log the message directly else: self.logger.log_raw(message) # If a test ended, we clean the buffer if message["action"] == "test_end": self.is_test_running = False self.buffered_messages = [] self.restore_buffering = self.restore_buffering or self.buffering self.buffering = False if message["action"] == "test_start": self.is_test_running = True if self.restore_buffering: self.restore_buffering = False self.buffering = True def write(self, line): messages = self.parse_line(line) for message in messages: self.process_message(message) return messages def flush(self): sys.stdout.flush() def dump_buffered(self, limit=False): if limit: dumped_messages = self.buffered_messages[-self.BUFFERING_THRESHOLD :] else: dumped_messages = self.buffered_messages last_timestamp = None for buf in dumped_messages: # pylint --py3k W1619 timestamp = datetime.fromtimestamp(buf["time"] / 1000).strftime("%H:%M:%S") if timestamp != last_timestamp: self.logger.info("Buffered messages logged at {}".format(timestamp)) last_timestamp = timestamp self.logger.log_raw(buf) self.logger.info("Buffered messages finished") # Cleaning the list of buffered messages self.buffered_messages = [] def finish(self): self.dump_buffered() self.buffering = False self.logger.suite_end() #################### # PROCESS HANDLING # #################### def call(*args, **kwargs): """wraps mozprocess.run_and_wait with process output logging""" log = get_proxy_logger("mochitest") def on_output(proc, line): cmdline = subprocess.list2cmdline(proc.args) log.process_output( process=proc.pid, data=line, command=cmdline, ) process = mozprocess.run_and_wait(*args, output_line_handler=on_output, **kwargs) return process.returncode def killPid(pid, log): # see also https://bugzilla.mozilla.org/show_bug.cgi?id=911249#c58 if HAVE_PSUTIL: # Kill a process tree (including grandchildren) with signal.SIGTERM if pid == os.getpid(): raise RuntimeError("Error: trying to kill ourselves, not another process") try: parent = psutil.Process(pid) children = parent.children(recursive=True) children.append(parent) for p in children: p.send_signal(signal.SIGTERM) gone, alive = psutil.wait_procs(children, timeout=30) for p in gone: log.info("psutil found pid %s dead" % p.pid) for p in alive: log.info("failed to kill pid %d after 30s" % p.pid) except Exception as e: log.info("Error: Failed to kill process %d: %s" % (pid, str(e))) else: try: os.kill(pid, getattr(signal, "SIGKILL", signal.SIGTERM)) except Exception as e: log.info("Failed to kill process %d: %s" % (pid, str(e))) if mozinfo.isWin: import ctypes.wintypes def isPidAlive(pid): STILL_ACTIVE = 259 PROCESS_QUERY_LIMITED_INFORMATION = 0x1000 pHandle = ctypes.windll.kernel32.OpenProcess( PROCESS_QUERY_LIMITED_INFORMATION, 0, pid ) if not pHandle: return False try: pExitCode = ctypes.wintypes.DWORD() ctypes.windll.kernel32.GetExitCodeProcess(pHandle, ctypes.byref(pExitCode)) if pExitCode.value != STILL_ACTIVE: return False # We have a live process handle. But Windows aggressively # re-uses pids, so let's attempt to verify that this is # actually Firefox. namesize = 1024 pName = ctypes.create_string_buffer(namesize) namelen = ctypes.windll.psapi.GetProcessImageFileNameA( pHandle, pName, namesize ) if namelen == 0: # Still an active process, so conservatively assume it's Firefox. return True return pName.value.endswith((b"firefox.exe", b"plugin-container.exe")) finally: ctypes.windll.kernel32.CloseHandle(pHandle) else: import errno def isPidAlive(pid): try: # kill(pid, 0) checks for a valid PID without actually sending a signal # The method throws OSError if the PID is invalid, which we catch # below. os.kill(pid, 0) # Wait on it to see if it's a zombie. This can throw OSError.ECHILD if # the process terminates before we get to this point. wpid, wstatus = os.waitpid(pid, os.WNOHANG) return wpid == 0 except OSError as err: # Catch the errors we might expect from os.kill/os.waitpid, # and re-raise any others if err.errno in (errno.ESRCH, errno.ECHILD, errno.EPERM): return False raise # TODO: ^ upstream isPidAlive to mozprocess ####################### # HTTP SERVER SUPPORT # ####################### class MochitestServer(object): "Web server used to serve Mochitests, for closer fidelity to the real web." instance_count = 0 def __init__(self, options, logger): if isinstance(options, Namespace): options = vars(options) self._log = logger self._keep_open = bool(options["keep_open"]) self._utilityPath = options["utilityPath"] self._xrePath = options["xrePath"] self._profileDir = options["profilePath"] self.webServer = options["webServer"] self.httpPort = options["httpPort"] if options.get("remoteWebServer") == "10.0.2.2": # probably running an Android emulator and 10.0.2.2 will # not be visible from host shutdownServer = "127.0.0.1" else: shutdownServer = self.webServer self.shutdownURL = "http://%(server)s:%(port)s/server/shutdown" % { "server": shutdownServer, "port": self.httpPort, } self.debugURL = "http://%(server)s:%(port)s/server/debug?2" % { "server": shutdownServer, "port": self.httpPort, } self.testPrefix = "undefined" if options.get("httpdPath"): self._httpdPath = options["httpdPath"] else: self._httpdPath = SCRIPT_DIR self._httpdPath = os.path.abspath(self._httpdPath) MochitestServer.instance_count += 1 def start(self): "Run the Mochitest server, returning the process ID of the server." # get testing environment env = test_environment(xrePath=self._xrePath, log=self._log) env["XPCOM_DEBUG_BREAK"] = "warn" if "LD_LIBRARY_PATH" not in env or env["LD_LIBRARY_PATH"] is None: env["LD_LIBRARY_PATH"] = self._xrePath else: env["LD_LIBRARY_PATH"] = ":".join([self._xrePath, env["LD_LIBRARY_PATH"]]) # When running with an ASan build, our xpcshell server will also be ASan-enabled, # thus consuming too much resources when running together with the browser on # the test machines. Try to limit the amount of resources by disabling certain # features. env["ASAN_OPTIONS"] = "quarantine_size=1:redzone=32:malloc_context_size=5" # Likewise, when running with a TSan build, our xpcshell server will # also be TSan-enabled. Except that in this case, we don't really # care about races in xpcshell. So disable TSan for the server. env["TSAN_OPTIONS"] = "report_bugs=0" # Don't use socket process for the xpcshell server. env["MOZ_DISABLE_SOCKET_PROCESS"] = "1" if mozinfo.isWin: env["PATH"] = env["PATH"] + ";" + str(self._xrePath) args = [ "-g", self._xrePath, "-e", "const _PROFILE_PATH = '%(profile)s'; const _SERVER_PORT = '%(port)s'; " "const _SERVER_ADDR = '%(server)s'; const _TEST_PREFIX = %(testPrefix)s; " "const _DISPLAY_RESULTS = %(displayResults)s; " "const _HTTPD_PATH = '%(httpdPath)s';" % { "httpdPath": self._httpdPath.replace("\\", "\\\\"), "profile": self._profileDir.replace("\\", "\\\\"), "port": self.httpPort, "server": self.webServer, "testPrefix": self.testPrefix, "displayResults": str(self._keep_open).lower(), }, "-f", os.path.join(SCRIPT_DIR, "server.js"), ] xpcshell = os.path.join( self._utilityPath, "xpcshell" + mozinfo.info["bin_suffix"] ) command = [xpcshell] + args if MOCHITEST_SERVER_LOGGING and "MOZ_UPLOAD_DIR" in os.environ: server_logfile_path = os.path.join( os.environ["MOZ_UPLOAD_DIR"], "mochitest-server-%d.txt" % MochitestServer.instance_count, ) self.server_logfile = open(server_logfile_path, "w") self._process = subprocess.Popen( command, cwd=SCRIPT_DIR, env=env, stdout=self.server_logfile, stderr=subprocess.STDOUT, ) else: self.server_logfile = None self._process = subprocess.Popen( command, cwd=SCRIPT_DIR, env=env, ) self._log.info("%s : launching %s" % (self.__class__.__name__, command)) pid = self._process.pid self._log.info("runtests.py | Server pid: %d" % pid) if MOCHITEST_SERVER_LOGGING and "MOZ_UPLOAD_DIR" in os.environ: self._log.info("runtests.py enabling server debugging...") i = 0 while i < 5: try: with closing(urlopen(self.debugURL)) as c: self._log.info(six.ensure_text(c.read())) break except Exception as e: self._log.info("exception when enabling debugging: %s" % str(e)) time.sleep(1) i += 1 def ensureReady(self, timeout): assert timeout >= 0 aliveFile = os.path.join(self._profileDir, "server_alive.txt") i = 0 while i < timeout: if os.path.exists(aliveFile): break time.sleep(0.05) i += 0.05 else: self._log.error( "TEST-UNEXPECTED-FAIL | runtests.py | Timed out while waiting for server startup." ) self.stop() sys.exit(1) def stop(self): try: with closing(urlopen(self.shutdownURL)) as c: self._log.info(six.ensure_text(c.read())) except Exception: self._log.info("Failed to stop web server on %s" % self.shutdownURL) traceback.print_exc() finally: if self.server_logfile is not None: self.server_logfile.close() if self._process is not None: # Kill the server immediately to avoid logging intermittent # shutdown crashes, sometimes observed on Windows 10. self._process.kill() self._log.info("Web server killed.") class WebSocketServer(object): "Class which encapsulates the mod_pywebsocket server" def __init__(self, options, scriptdir, logger, debuggerInfo=None): self.port = options.webSocketPort self.debuggerInfo = debuggerInfo self._log = logger self._scriptdir = scriptdir def start(self): # Invoke pywebsocket through a wrapper which adds special SIGINT handling. # # If we're in an interactive debugger, the wrapper causes the server to # ignore SIGINT so the server doesn't capture a ctrl+c meant for the # debugger. # # If we're not in an interactive debugger, the wrapper causes the server to # die silently upon receiving a SIGINT. scriptPath = "pywebsocket_wrapper.py" script = os.path.join(self._scriptdir, scriptPath) cmd = [sys.executable, script] if self.debuggerInfo and self.debuggerInfo.interactive: cmd += ["--interactive"] # We need to use 0.0.0.0 to listen on all interfaces because # Android tests connect from a different hosts cmd += [ "-H", "0.0.0.0", "-p", str(self.port), "-w", self._scriptdir, "-l", os.path.join(self._scriptdir, "websock.log"), "--log-level=debug", "--allow-handlers-outside-root-dir", ] env = dict(os.environ) env["PYTHONPATH"] = os.pathsep.join(sys.path) # Start the process. Ignore stderr so that exceptions from the server # are not treated as failures when parsing the test log. self._process = subprocess.Popen( cmd, cwd=SCRIPT_DIR, env=env, stderr=subprocess.DEVNULL ) pid = self._process.pid self._log.info("runtests.py | Websocket server pid: %d" % pid) def stop(self): if self._process is not None: self._process.kill() class SSLTunnel: def __init__(self, options, logger): self.log = logger self.process = None self.utilityPath = options.utilityPath self.xrePath = options.xrePath self.certPath = options.certPath self.sslPort = options.sslPort self.httpPort = options.httpPort self.webServer = options.webServer self.webSocketPort = options.webSocketPort self.customCertRE = re.compile("^cert=(?P[0-9a-zA-Z_ ]+)") self.clientAuthRE = re.compile("^clientauth=(?P[a-z]+)") self.redirRE = re.compile("^redir=(?P[0-9a-zA-Z_ .]+)") def writeLocation(self, config, loc): for option in loc.options: match = self.customCertRE.match(option) if match: customcert = match.group("nickname") config.write( "listen:%s:%s:%s:%s\n" % (loc.host, loc.port, self.sslPort, customcert) ) match = self.clientAuthRE.match(option) if match: clientauth = match.group("clientauth") config.write( "clientauth:%s:%s:%s:%s\n" % (loc.host, loc.port, self.sslPort, clientauth) ) match = self.redirRE.match(option) if match: redirhost = match.group("redirhost") config.write( "redirhost:%s:%s:%s:%s\n" % (loc.host, loc.port, self.sslPort, redirhost) ) if option in ( "tls1", "tls1_1", "tls1_2", "tls1_3", "ssl3", "3des", "failHandshake", ): config.write( "%s:%s:%s:%s\n" % (option, loc.host, loc.port, self.sslPort) ) def buildConfig(self, locations, public=None): """Create the ssltunnel configuration file""" configFd, self.configFile = tempfile.mkstemp(prefix="ssltunnel", suffix=".cfg") with os.fdopen(configFd, "w") as config: config.write("httpproxy:1\n") config.write("certdbdir:%s\n" % self.certPath) config.write("forward:127.0.0.1:%s\n" % self.httpPort) wsserver = self.webServer if self.webServer == "10.0.2.2": wsserver = "127.0.0.1" config.write("websocketserver:%s:%s\n" % (wsserver, self.webSocketPort)) # Use "*" to tell ssltunnel to listen on the public ip # address instead of the loopback address 127.0.0.1. This # may have the side-effect of causing firewall warnings on # macOS and Windows. Use "127.0.0.1" to listen on the # loopback address. Remote tests using physical or # emulated Android devices must use the public ip address # in order for the sslproxy to work but Desktop tests # which run on the same host as ssltunnel may use the # loopback address. listen_address = "*" if public else "127.0.0.1" config.write("listen:%s:%s:pgoserver\n" % (listen_address, self.sslPort)) for loc in locations: if loc.scheme == "https" and "nocert" not in loc.options: self.writeLocation(config, loc) def start(self): """Starts the SSL Tunnel""" # start ssltunnel to provide https:// URLs capability ssltunnel = os.path.join(self.utilityPath, "ssltunnel") if os.name == "nt": ssltunnel += ".exe" if not os.path.exists(ssltunnel): self.log.error( "INFO | runtests.py | expected to find ssltunnel at %s" % ssltunnel ) sys.exit(1) env = test_environment(xrePath=self.xrePath, log=self.log) env["LD_LIBRARY_PATH"] = self.xrePath self.process = subprocess.Popen([ssltunnel, self.configFile], env=env) self.log.info("runtests.py | SSL tunnel pid: %d" % self.process.pid) def stop(self): """Stops the SSL Tunnel and cleans up""" if self.process is not None: self.process.kill() if os.path.exists(self.configFile): os.remove(self.configFile) def checkAndConfigureV4l2loopback(device): """ Determine if a given device path is a v4l2loopback device, and if so toggle a few settings on it via fcntl. Very linux-specific. Returns (status, device name) where status is a boolean. """ if not mozinfo.isLinux: return False, "" libc = ctypes.cdll.LoadLibrary(find_library("c")) O_RDWR = 2 # These are from linux/videodev2.h class v4l2_capability(ctypes.Structure): _fields_ = [ ("driver", ctypes.c_char * 16), ("card", ctypes.c_char * 32), ("bus_info", ctypes.c_char * 32), ("version", ctypes.c_uint32), ("capabilities", ctypes.c_uint32), ("device_caps", ctypes.c_uint32), ("reserved", ctypes.c_uint32 * 3), ] VIDIOC_QUERYCAP = 0x80685600 fd = libc.open(six.ensure_binary(device), O_RDWR) if fd < 0: return False, "" vcap = v4l2_capability() if libc.ioctl(fd, VIDIOC_QUERYCAP, ctypes.byref(vcap)) != 0: return False, "" if six.ensure_text(vcap.driver) != "v4l2 loopback": return False, "" class v4l2_control(ctypes.Structure): _fields_ = [("id", ctypes.c_uint32), ("value", ctypes.c_int32)] # These are private v4l2 control IDs, see: # https://github.com/umlaeute/v4l2loopback/blob/fd822cf0faaccdf5f548cddd9a5a3dcebb6d584d/v4l2loopback.c#L131 KEEP_FORMAT = 0x8000000 SUSTAIN_FRAMERATE = 0x8000001 VIDIOC_S_CTRL = 0xC008561C control = v4l2_control() control.id = KEEP_FORMAT control.value = 1 libc.ioctl(fd, VIDIOC_S_CTRL, ctypes.byref(control)) control.id = SUSTAIN_FRAMERATE control.value = 1 libc.ioctl(fd, VIDIOC_S_CTRL, ctypes.byref(control)) libc.close(fd) return True, six.ensure_text(vcap.card) def findTestMediaDevices(log): """ Find the test media devices configured on this system, and return a dict containing information about them. The dict will have keys for 'audio' and 'video', each containing the name of the media device to use. If audio and video devices could not be found, return None. This method is only currently implemented for Linux. """ if not mozinfo.isLinux: return None info = {} # Look for a v4l2loopback device. name = None device = None for dev in sorted(glob.glob("/dev/video*")): result, name_ = checkAndConfigureV4l2loopback(dev) if result: name = name_ device = dev break if not (name and device): log.error("Couldn't find a v4l2loopback video device") return None # Feed it a frame of output so it has something to display gst01 = spawn.find_executable("gst-launch-0.1") gst010 = spawn.find_executable("gst-launch-0.10") gst10 = spawn.find_executable("gst-launch-1.0") if gst01: gst = gst01 if gst010: gst = gst010 else: gst = gst10 process = subprocess.Popen( [ gst, "--no-fault", "videotestsrc", "pattern=green", "num-buffers=1", "!", "v4l2sink", "device=%s" % device, ] ) info["video"] = {"name": name, "process": process} # Hardcode the PulseAudio module-null-sink name since it's always the same. info["audio"] = {"name": "Monitor of Null Output"} return info def create_zip(path): """ Takes a `path` on disk and creates a zipfile with its contents. Returns a path to the location of the temporary zip file. """ with tempfile.NamedTemporaryFile() as f: # `shutil.make_archive` writes to "{f.name}.zip", so we're really just # using `NamedTemporaryFile` as a way to get a random path. return shutil.make_archive(f.name, "zip", path) def update_mozinfo(): """walk up directories to find mozinfo.json update the info""" # TODO: This should go in a more generic place, e.g. mozinfo path = SCRIPT_DIR dirs = set() while path != os.path.expanduser("~"): if path in dirs: break dirs.add(path) path = os.path.split(path)[0] mozinfo.find_and_update_from_json(*dirs) class MochitestDesktop(object): """ Mochitest class for desktop firefox. """ oldcwd = os.getcwd() # Path to the test script on the server TEST_PATH = "tests" CHROME_PATH = "redirect.html" certdbNew = False sslTunnel = None DEFAULT_TIMEOUT = 60.0 mediaDevices = None mozinfo_variables_shown = False patternFiles = {} # XXX use automation.py for test name to avoid breaking legacy # TODO: replace this with 'runtests.py' or 'mochitest' or the like test_name = "automation.py" def __init__(self, flavor, logger_options, staged_addons=None, quiet=False): update_mozinfo() self.flavor = flavor self.staged_addons = staged_addons self.server = None self.wsserver = None self.websocketProcessBridge = None self.sslTunnel = None self.manifest = None self.tests_by_manifest = defaultdict(list) self.args_by_manifest = defaultdict(set) self.prefs_by_manifest = defaultdict(set) self.env_vars_by_manifest = defaultdict(set) self.tests_dirs_by_manifest = defaultdict(set) self._active_tests = None self.currentTests = None self._locations = None self.browserEnv = None self.marionette = None self.start_script = None self.mozLogs = None self.start_script_kwargs = {} self.extraArgs = [] self.extraPrefs = {} self.extraEnv = {} self.extraTestsDirs = [] self.conditioned_profile_dir = None if logger_options.get("log"): self.log = logger_options["log"] else: self.log = commandline.setup_logging( "mochitest", logger_options, {"tbpl": sys.stdout} ) self.message_logger = MessageLogger( logger=self.log, buffering=quiet, structured=True ) # Max time in seconds to wait for server startup before tests will fail -- if # this seems big, it's mostly for debug machines where cold startup # (particularly after a build) takes forever. self.SERVER_STARTUP_TIMEOUT = 180 if mozinfo.info.get("debug") else 90 # metro browser sub process id self.browserProcessId = None self.haveDumpedScreen = False # Create variables to count the number of passes, fails, todos. self.countpass = 0 self.countfail = 0 self.counttodo = 0 self.expectedError = {} self.result = {} self.start_script = os.path.join(here, "start_desktop.js") # Used to temporarily serve a performance profile self.profiler_tempdir = None def environment(self, **kwargs): kwargs["log"] = self.log return test_environment(**kwargs) def getFullPath(self, path): "Get an absolute path relative to self.oldcwd." return os.path.normpath(os.path.join(self.oldcwd, os.path.expanduser(path))) def getLogFilePath(self, logFile): """return the log file path relative to the device we are testing on, in most cases it will be the full path on the local system """ return self.getFullPath(logFile) @property def locations(self): if self._locations is not None: return self._locations locations_file = os.path.join(SCRIPT_DIR, "server-locations.txt") self._locations = ServerLocations(locations_file) return self._locations def buildURLOptions(self, options, env): """Add test control options from the command line to the url URL parameters to test URL: autorun -- kick off tests automatically closeWhenDone -- closes the browser after the tests hideResultsTable -- hides the table of individual test results logFile -- logs test run to an absolute path startAt -- name of test to start at endAt -- name of test to end at timeout -- per-test timeout in seconds repeat -- How many times to repeat the test, ie: repeat=1 will run the test twice. """ self.urlOpts = [] if not hasattr(options, "logFile"): options.logFile = "" if not hasattr(options, "fileLevel"): options.fileLevel = "INFO" # allow relative paths for logFile if options.logFile: options.logFile = self.getLogFilePath(options.logFile) if options.flavor in ("a11y", "browser", "chrome"): self.makeTestConfig(options) else: if options.autorun: self.urlOpts.append("autorun=1") if options.timeout: self.urlOpts.append("timeout=%d" % options.timeout) if options.maxTimeouts: self.urlOpts.append("maxTimeouts=%d" % options.maxTimeouts) if not options.keep_open: self.urlOpts.append("closeWhenDone=1") if options.logFile: self.urlOpts.append("logFile=" + encodeURIComponent(options.logFile)) self.urlOpts.append( "fileLevel=" + encodeURIComponent(options.fileLevel) ) if options.consoleLevel: self.urlOpts.append( "consoleLevel=" + encodeURIComponent(options.consoleLevel) ) if options.startAt: self.urlOpts.append("startAt=%s" % options.startAt) if options.endAt: self.urlOpts.append("endAt=%s" % options.endAt) if options.shuffle: self.urlOpts.append("shuffle=1") if "MOZ_HIDE_RESULTS_TABLE" in env and env["MOZ_HIDE_RESULTS_TABLE"] == "1": self.urlOpts.append("hideResultsTable=1") if options.runUntilFailure: self.urlOpts.append("runUntilFailure=1") if options.repeat: self.urlOpts.append("repeat=%d" % options.repeat) if len(options.test_paths) == 1 and os.path.isfile( os.path.join( self.oldcwd, os.path.dirname(__file__), self.TEST_PATH, options.test_paths[0], ) ): self.urlOpts.append( "testname=%s" % "/".join([self.TEST_PATH, options.test_paths[0]]) ) if options.manifestFile: self.urlOpts.append("manifestFile=%s" % options.manifestFile) if options.failureFile: self.urlOpts.append( "failureFile=%s" % self.getFullPath(options.failureFile) ) if options.runSlower: self.urlOpts.append("runSlower=true") if options.debugOnFailure: self.urlOpts.append("debugOnFailure=true") if options.dumpOutputDirectory: self.urlOpts.append( "dumpOutputDirectory=%s" % encodeURIComponent(options.dumpOutputDirectory) ) if options.dumpAboutMemoryAfterTest: self.urlOpts.append("dumpAboutMemoryAfterTest=true") if options.dumpDMDAfterTest: self.urlOpts.append("dumpDMDAfterTest=true") if options.debugger or options.jsdebugger: self.urlOpts.append("interactiveDebugger=true") if options.jscov_dir_prefix: self.urlOpts.append("jscovDirPrefix=%s" % options.jscov_dir_prefix) if options.cleanupCrashes: self.urlOpts.append("cleanupCrashes=true") if "MOZ_XORIGIN_MOCHITEST" in env and env["MOZ_XORIGIN_MOCHITEST"] == "1": options.xOriginTests = True if options.xOriginTests: self.urlOpts.append("xOriginTests=true") if options.comparePrefs: self.urlOpts.append("comparePrefs=true") self.urlOpts.append("ignorePrefsFile=ignorePrefs.json") def normflavor(self, flavor): """ In some places the string 'browser-chrome' is expected instead of 'browser' and 'mochitest' instead of 'plain'. Normalize the flavor strings for those instances. """ # TODO Use consistent flavor strings everywhere and remove this if flavor == "browser": return "browser-chrome" elif flavor == "plain": return "mochitest" return flavor # This check can be removed when bug 983867 is fixed. def isTest(self, options, filename): allow_js_css = False if options.flavor == "browser": allow_js_css = True testPattern = re.compile(r"browser_.+\.js") elif options.flavor in ("a11y", "chrome"): testPattern = re.compile(r"(browser|test)_.+\.(xul|html|js|xhtml)") else: testPattern = re.compile(r"test_") if not allow_js_css and (".js" in filename or ".css" in filename): return False pathPieces = filename.split("/") return testPattern.match(pathPieces[-1]) and not re.search( r"\^headers\^$", filename ) def setTestRoot(self, options): if options.flavor != "plain": self.testRoot = options.flavor else: self.testRoot = self.TEST_PATH self.testRootAbs = os.path.join(SCRIPT_DIR, self.testRoot) def buildTestURL(self, options, scheme="http"): if scheme == "https": testHost = "https://example.com:443" elif options.xOriginTests: testHost = "http://mochi.xorigin-test:8888" else: testHost = "http://mochi.test:8888" testURL = "/".join([testHost, self.TEST_PATH]) if len(options.test_paths) == 1: if os.path.isfile( os.path.join( self.oldcwd, os.path.dirname(__file__), self.TEST_PATH, options.test_paths[0], ) ): testURL = "/".join([testURL, os.path.dirname(options.test_paths[0])]) else: testURL = "/".join([testURL, options.test_paths[0]]) if options.flavor in ("a11y", "chrome"): testURL = "/".join([testHost, self.CHROME_PATH]) elif options.flavor == "browser": testURL = "about:blank" return testURL def parseAndCreateTestsDirs(self, m): testsDirs = list(self.tests_dirs_by_manifest[m])[0] self.extraTestsDirs = [] if testsDirs: self.extraTestsDirs = testsDirs.strip().split() self.log.info( "The following extra test directories will be created:\n {}".format( "\n ".join(self.extraTestsDirs) ) ) self.createExtraTestsDirs(self.extraTestsDirs, m) def createExtraTestsDirs(self, extraTestsDirs=None, manifest=None): """Take a list of directories that might be needed to exist by the test prior to even the main process be executed, and: - verify it does not already exists - create it if it does Removal of those directories is handled in cleanup() """ if type(extraTestsDirs) != list: return for d in extraTestsDirs: if os.path.exists(d): raise FileExistsError( "Directory '{}' already exists. This is a member of " "test-directories in manifest {}.".format(d, manifest) ) created = [] for d in extraTestsDirs: os.makedirs(d) created += [d] if created != extraTestsDirs: raise EnvironmentError( "Not all directories were created: extraTestsDirs={} -- created={}".format( extraTestsDirs, created ) ) def getTestsByScheme( self, options, testsToFilter=None, disabled=True, manifestToFilter=None ): """Build the url path to the specific test harness and test file or directory Build a manifest of tests to run and write out a json file for the harness to read testsToFilter option is used to filter/keep the tests provided in the list disabled -- This allows to add all disabled tests on the build side and then on the run side to only run the enabled ones """ tests = self.getActiveTests(options, disabled) paths = [] for test in tests: if testsToFilter and (test["path"] not in testsToFilter): continue # If we are running a specific manifest, the previously computed set of active # tests should be filtered out based on the manifest that contains that entry. # # This is especially important when a test file is listed in multiple # manifests (e.g. because the same test runs under a different configuration, # and so it is being included in multiple manifests), without filtering the # active tests based on the current manifest (configuration) that we are # running for each of the N manifests we would be executing the active tests # exactly N times (and so NxN runs instead of the expected N runs, one for each # manifest). if manifestToFilter and (test["manifest"] not in manifestToFilter): continue paths.append(test) # Generate test by schemes for scheme, grouped_tests in self.groupTestsByScheme(paths).items(): # Bug 883865 - add this functionality into manifestparser with open( os.path.join(SCRIPT_DIR, options.testRunManifestFile), "w" ) as manifestFile: manifestFile.write(json.dumps({"tests": grouped_tests})) options.manifestFile = options.testRunManifestFile yield (scheme, grouped_tests) def startWebSocketServer(self, options, debuggerInfo): """Launch the websocket server""" self.wsserver = WebSocketServer(options, SCRIPT_DIR, self.log, debuggerInfo) self.wsserver.start() def startWebServer(self, options): """Create the webserver and start it up""" self.server = MochitestServer(options, self.log) self.server.start() if options.pidFile != "": with open(options.pidFile + ".xpcshell.pid", "w") as f: f.write("%s" % self.server._process.pid) def startWebsocketProcessBridge(self, options): """Create a websocket server that can launch various processes that JS needs (eg; ICE server for webrtc testing) """ command = [ sys.executable, os.path.join("websocketprocessbridge", "websocketprocessbridge.py"), "--port", options.websocket_process_bridge_port, ] self.websocketProcessBridge = subprocess.Popen(command, cwd=SCRIPT_DIR) self.log.info( "runtests.py | websocket/process bridge pid: %d" % self.websocketProcessBridge.pid ) # ensure the server is up, wait for at most ten seconds for i in range(1, 100): if self.websocketProcessBridge.poll() is not None: self.log.error( "runtests.py | websocket/process bridge failed " "to launch. Are all the dependencies installed?" ) return try: sock = socket.create_connection(("127.0.0.1", 8191)) sock.close() break except Exception: time.sleep(0.1) else: self.log.error( "runtests.py | Timed out while waiting for " "websocket/process bridge startup." ) def needsWebsocketProcessBridge(self, options): """ Returns a bool indicating if the current test configuration needs to start the websocket process bridge or not. The boils down to if WebRTC tests that need the bridge are present. """ tests = self.getActiveTests(options) is_webrtc_tag_present = False for test in tests: if "webrtc" in test.get("tags", ""): is_webrtc_tag_present = True break return is_webrtc_tag_present and options.subsuite in ["media"] def startHttp3Server(self, options): """ Start a Http3 test server. """ http3ServerPath = os.path.join( options.utilityPath, "http3server" + mozinfo.info["bin_suffix"] ) serverOptions = {} serverOptions["http3ServerPath"] = http3ServerPath serverOptions["profilePath"] = options.profilePath serverOptions["isMochitest"] = True serverOptions["isWin"] = mozinfo.isWin serverOptions["proxyPort"] = options.http3ServerPort env = test_environment(xrePath=options.xrePath, log=self.log) serverEnv = env.copy() serverLog = env.get("MOZHTTP3_SERVER_LOG") if serverLog is not None: serverEnv["RUST_LOG"] = serverLog self.http3Server = Http3Server(serverOptions, serverEnv, self.log) self.http3Server.start() port = self.http3Server.ports().get("MOZHTTP3_PORT_PROXY") if int(port) != options.http3ServerPort: self.http3Server = None raise RuntimeError("Error: Unable to start Http/3 server") def findNodeBin(self): # We try to find the node executable in the path given to us by the user in # the MOZ_NODE_PATH environment variable nodeBin = os.getenv("MOZ_NODE_PATH", None) self.log.info("Use MOZ_NODE_PATH at %s" % (nodeBin)) if not nodeBin and build: nodeBin = build.substs.get("NODEJS") self.log.info("Use build node at %s" % (nodeBin)) return nodeBin def startHttp2Server(self, options): """ Start a Http2 test server. """ serverOptions = {} serverOptions["serverPath"] = os.path.join( SCRIPT_DIR, "Http2Server", "http2_server.js" ) serverOptions["nodeBin"] = self.findNodeBin() serverOptions["isWin"] = mozinfo.isWin serverOptions["port"] = options.http2ServerPort env = test_environment(xrePath=options.xrePath, log=self.log) self.http2Server = Http2Server(serverOptions, env, self.log) self.http2Server.start() port = self.http2Server.port() if port != options.http2ServerPort: raise RuntimeError("Error: Unable to start Http2 server") def startDoHServer(self, options, dstServerPort, alpn): serverOptions = {} serverOptions["serverPath"] = os.path.join( SCRIPT_DIR, "DoHServer", "doh_server.js" ) serverOptions["nodeBin"] = self.findNodeBin() serverOptions["dstServerPort"] = dstServerPort serverOptions["isWin"] = mozinfo.isWin serverOptions["port"] = options.dohServerPort serverOptions["alpn"] = alpn env = test_environment(xrePath=options.xrePath, log=self.log) self.dohServer = DoHServer(serverOptions, env, self.log) self.dohServer.start() port = self.dohServer.port() if port != options.dohServerPort: raise RuntimeError("Error: Unable to start DoH server") def startServers(self, options, debuggerInfo, public=None): # start servers and set ports # TODO: pass these values, don't set on `self` self.webServer = options.webServer self.httpPort = options.httpPort self.sslPort = options.sslPort self.webSocketPort = options.webSocketPort # httpd-path is specified by standard makefile targets and may be specified # on the command line to select a particular version of httpd.js. If not # specified, try to select the one from hostutils.zip, as required in # bug 882932. if not options.httpdPath: options.httpdPath = os.path.join(options.utilityPath, "components") self.startWebServer(options) self.startWebSocketServer(options, debuggerInfo) # Only webrtc mochitests in the media suite need the websocketprocessbridge. if self.needsWebsocketProcessBridge(options): self.startWebsocketProcessBridge(options) # start SSL pipe self.sslTunnel = SSLTunnel(options, logger=self.log) self.sslTunnel.buildConfig(self.locations, public=public) self.sslTunnel.start() # If we're lucky, the server has fully started by now, and all paths are # ready, etc. However, xpcshell cold start times suck, at least for debug # builds. We'll try to connect to the server for awhile, and if we fail, # we'll try to kill the server and exit with an error. if self.server is not None: self.server.ensureReady(self.SERVER_STARTUP_TIMEOUT) self.log.info("use http3 server: %d" % options.useHttp3Server) self.http3Server = None self.http2Server = None self.dohServer = None if options.useHttp3Server: self.startHttp3Server(options) self.startDoHServer(options, options.http3ServerPort, "h3") elif options.useHttp2Server: self.startHttp2Server(options) self.startDoHServer(options, options.http2ServerPort, "h2") def stopServers(self): """Servers are no longer needed, and perhaps more importantly, anything they might spew to console might confuse things.""" if self.server is not None: try: self.log.info("Stopping web server") self.server.stop() except Exception: self.log.critical("Exception when stopping web server") if self.wsserver is not None: try: self.log.info("Stopping web socket server") self.wsserver.stop() except Exception: self.log.critical("Exception when stopping web socket server") if self.sslTunnel is not None: try: self.log.info("Stopping ssltunnel") self.sslTunnel.stop() except Exception: self.log.critical("Exception stopping ssltunnel") if self.websocketProcessBridge is not None: try: self.websocketProcessBridge.kill() self.websocketProcessBridge.wait() self.log.info("Stopping websocket/process bridge") except Exception: self.log.critical("Exception stopping websocket/process bridge") if self.http3Server is not None: try: self.http3Server.stop() except Exception: self.log.critical("Exception stopping http3 server") if self.http2Server is not None: try: self.http2Server.stop() except Exception: self.log.critical("Exception stopping http2 server") if self.dohServer is not None: try: self.dohServer.stop() except Exception: self.log.critical("Exception stopping doh server") if hasattr(self, "gstForV4l2loopbackProcess"): try: self.gstForV4l2loopbackProcess.kill() self.gstForV4l2loopbackProcess.wait() self.log.info("Stopping gst for v4l2loopback") except Exception: self.log.critical("Exception stopping gst for v4l2loopback") def copyExtraFilesToProfile(self, options): "Copy extra files or dirs specified on the command line to the testing profile." for f in options.extraProfileFiles: abspath = self.getFullPath(f) if os.path.isfile(abspath): shutil.copy2(abspath, options.profilePath) elif os.path.isdir(abspath): dest = os.path.join(options.profilePath, os.path.basename(abspath)) shutil.copytree(abspath, dest) else: self.log.warning("runtests.py | Failed to copy %s to profile" % abspath) def getChromeTestDir(self, options): dir = os.path.join(os.path.abspath("."), SCRIPT_DIR) + "/" if mozinfo.isWin: dir = "file:///" + dir.replace("\\", "/") return dir def writeChromeManifest(self, options): manifest = os.path.join(options.profilePath, "tests.manifest") with open(manifest, "w") as manifestFile: # Register chrome directory. chrometestDir = self.getChromeTestDir(options) manifestFile.write( "content mochitests %s contentaccessible=yes\n" % chrometestDir ) manifestFile.write( "content mochitests-any %s contentaccessible=yes remoteenabled=yes\n" % chrometestDir ) manifestFile.write( "content mochitests-content %s contentaccessible=yes remoterequired=yes\n" % chrometestDir ) if options.testingModulesDir is not None: manifestFile.write( "resource testing-common file:///%s\n" % options.testingModulesDir ) if options.store_chrome_manifest: shutil.copyfile(manifest, options.store_chrome_manifest) return manifest def addChromeToProfile(self, options): "Adds MochiKit chrome tests to the profile." # Create (empty) chrome directory. chromedir = os.path.join(options.profilePath, "chrome") os.mkdir(chromedir) # Write userChrome.css. chrome = """ /* set default namespace to XUL */ @namespace url("http://www.mozilla.org/keymaster/gatekeeper/there.is.only.xul"); toolbar, toolbarpalette { background-color: rgb(235, 235, 235) !important; } toolbar#nav-bar { background-image: none !important; } """ with open( os.path.join(options.profilePath, "userChrome.css"), "a" ) as chromeFile: chromeFile.write(chrome) manifest = self.writeChromeManifest(options) return manifest def getExtensionsToInstall(self, options): "Return a list of extensions to install in the profile" extensions = [] appDir = ( options.app[: options.app.rfind(os.sep)] if options.app else options.utilityPath ) extensionDirs = [ # Extensions distributed with the test harness. os.path.normpath(os.path.join(SCRIPT_DIR, "extensions")), ] if appDir: # Extensions distributed with the application. extensionDirs.append(os.path.join(appDir, "distribution", "extensions")) for extensionDir in extensionDirs: if os.path.isdir(extensionDir): for dirEntry in os.listdir(extensionDir): if dirEntry not in options.extensionsToExclude: path = os.path.join(extensionDir, dirEntry) if os.path.isdir(path) or ( os.path.isfile(path) and path.endswith(".xpi") ): extensions.append(path) extensions.extend(options.extensionsToInstall) return extensions def logPreamble(self, tests): """Logs a suite_start message and test_start/test_end at the beginning of a run.""" self.log.suite_start( self.tests_by_manifest, name="mochitest-{}".format(self.flavor) ) for test in tests: if "disabled" in test: self.log.test_start(test["path"]) self.log.test_end(test["path"], "SKIP", message=test["disabled"]) def loadFailurePatternFile(self, pat_file): if pat_file in self.patternFiles: return self.patternFiles[pat_file] if not os.path.isfile(pat_file): self.log.warning( "runtests.py | Cannot find failure pattern file " + pat_file ) return None # Using ":error" to ensure it shows up in the failure summary. self.log.warning( "[runtests.py:error] Using {} to filter failures. If there " "is any number mismatch below, you could have fixed " "something documented in that file. Please reduce the " "failure count appropriately.".format(pat_file) ) patternRE = re.compile( r""" ^\s*\*\s* # list bullet (test_\S+|\.{3}) # test name (?:\s*(`.+?`|asserts))? # failure pattern (?::.+)? # optional description \s*\[(\d+|\*)\] # expected count \s*$ """, re.X, ) patterns = {} with open(pat_file) as f: last_name = None for line in f: match = patternRE.match(line) if not match: continue name = match.group(1) name = last_name if name == "..." else name last_name = name pat = match.group(2) if pat is not None: pat = "ASSERTION" if pat == "asserts" else pat[1:-1] count = match.group(3) count = None if count == "*" else int(count) if name not in patterns: patterns[name] = [] patterns[name].append((pat, count)) self.patternFiles[pat_file] = patterns return patterns def getFailurePatterns(self, pat_file, test_name): patterns = self.loadFailurePatternFile(pat_file) if patterns: return patterns.get(test_name, None) def getActiveTests(self, options, disabled=True): """ This method is used to parse the manifest and return active filtered tests. """ if self._active_tests: return self._active_tests tests = [] manifest = self.getTestManifest(options) if manifest: if options.extra_mozinfo_json: mozinfo.update(options.extra_mozinfo_json) info = mozinfo.info filters = [ subsuite(options.subsuite), ] # Allow for only running tests/manifests which match this tag if options.conditionedProfile: if not options.test_tags: options.test_tags = [] options.test_tags.append("condprof") if options.test_tags: filters.append(tags(options.test_tags)) if options.test_paths: options.test_paths = self.normalize_paths(options.test_paths) filters.append(pathprefix(options.test_paths)) # Add chunking filters if specified if options.totalChunks: if options.chunkByDir: filters.append( chunk_by_dir( options.thisChunk, options.totalChunks, options.chunkByDir ) ) elif options.chunkByRuntime: if mozinfo.info["os"] == "android": platkey = "android" elif mozinfo.isWin: platkey = "windows" else: platkey = "unix" runtime_file = os.path.join( SCRIPT_DIR, "runtimes", "manifest-runtimes-{}.json".format(platkey), ) if not os.path.exists(runtime_file): self.log.error("runtime file %s not found!" % runtime_file) sys.exit(1) # Given the mochitest flavor, load the runtimes information # for only that flavor due to manifest runtime format change in Bug 1637463. with open(runtime_file, "r") as f: if "suite_name" in options: runtimes = json.load(f).get(options.suite_name, {}) else: runtimes = {} filters.append( chunk_by_runtime( options.thisChunk, options.totalChunks, runtimes ) ) else: filters.append( chunk_by_slice(options.thisChunk, options.totalChunks) ) noDefaultFilters = False if options.runFailures: filters.append(failures(options.runFailures)) noDefaultFilters = True tests = manifest.active_tests( exists=False, disabled=disabled, filters=filters, noDefaultFilters=noDefaultFilters, **info ) if len(tests) == 0: self.log.error( NO_TESTS_FOUND.format(options.flavor, manifest.fmt_filters()) ) paths = [] for test in tests: if len(tests) == 1 and "disabled" in test: del test["disabled"] pathAbs = os.path.abspath(test["path"]) assert os.path.normcase(pathAbs).startswith( os.path.normcase(self.testRootAbs) ) tp = pathAbs[len(self.testRootAbs) :].replace("\\", "/").strip("/") if not self.isTest(options, tp): self.log.warning( "Warning: %s from manifest %s is not a valid test" % (test["name"], test["manifest"]) ) continue manifest_key = test["manifest_relpath"] # Ignore ancestor_manifests that live at the root (e.g, don't have a # path separator). if "ancestor_manifest" in test and "/" in normsep( test["ancestor_manifest"] ): manifest_key = "{}:{}".format(test["ancestor_manifest"], manifest_key) manifest_key = manifest_key.replace("\\", "/") self.tests_by_manifest[manifest_key].append(tp) self.args_by_manifest[manifest_key].add(test.get("args")) self.prefs_by_manifest[manifest_key].add(test.get("prefs")) self.env_vars_by_manifest[manifest_key].add(test.get("environment")) self.tests_dirs_by_manifest[manifest_key].add(test.get("test-directories")) for key in ["args", "prefs", "environment", "test-directories"]: if key in test and not options.runByManifest and "disabled" not in test: self.log.error( "parsing {}: runByManifest mode must be enabled to " "set the `{}` key".format(test["manifest_relpath"], key) ) sys.exit(1) testob = {"path": tp, "manifest": manifest_key} if "disabled" in test: testob["disabled"] = test["disabled"] if "expected" in test: testob["expected"] = test["expected"] if "https_first_disabled" in test: testob["https_first_disabled"] = test["https_first_disabled"] == "true" if "allow_xul_xbl" in test: testob["allow_xul_xbl"] = test["allow_xul_xbl"] == "true" if "scheme" in test: testob["scheme"] = test["scheme"] if "tags" in test: testob["tags"] = test["tags"] if options.failure_pattern_file: pat_file = os.path.join( os.path.dirname(test["manifest"]), options.failure_pattern_file ) patterns = self.getFailurePatterns(pat_file, test["name"]) if patterns: testob["expected"] = patterns paths.append(testob) # The 'args' key needs to be set in the DEFAULT section, unfortunately # we can't tell what comes from DEFAULT or not. So to validate this, we # stash all args from tests in the same manifest into a set. If the # length of the set > 1, then we know 'args' didn't come from DEFAULT. args_not_default = [ m for m, p in six.iteritems(self.args_by_manifest) if len(p) > 1 ] if args_not_default: self.log.error( "The 'args' key must be set in the DEFAULT section of a " "manifest. Fix the following manifests: {}".format( "\n".join(args_not_default) ) ) sys.exit(1) # The 'prefs' key needs to be set in the DEFAULT section too. pref_not_default = [ m for m, p in six.iteritems(self.prefs_by_manifest) if len(p) > 1 ] if pref_not_default: self.log.error( "The 'prefs' key must be set in the DEFAULT section of a " "manifest. Fix the following manifests: {}".format( "\n".join(pref_not_default) ) ) sys.exit(1) # The 'environment' key needs to be set in the DEFAULT section too. env_not_default = [ m for m, p in six.iteritems(self.env_vars_by_manifest) if len(p) > 1 ] if env_not_default: self.log.error( "The 'environment' key must be set in the DEFAULT section of a " "manifest. Fix the following manifests: {}".format( "\n".join(env_not_default) ) ) sys.exit(1) paths.sort(key=lambda p: p["path"].split("/")) if options.dump_tests: options.dump_tests = os.path.expanduser(options.dump_tests) assert os.path.exists(os.path.dirname(options.dump_tests)) with open(options.dump_tests, "w") as dumpFile: dumpFile.write(json.dumps({"active_tests": paths})) self.log.info("Dumping active_tests to %s file." % options.dump_tests) sys.exit() # Upload a list of test manifests that were executed in this run. if "MOZ_UPLOAD_DIR" in os.environ: artifact = os.path.join(os.environ["MOZ_UPLOAD_DIR"], "manifests.list") with open(artifact, "a") as fh: fh.write("\n".join(sorted(self.tests_by_manifest.keys()))) self._active_tests = paths return self._active_tests def getTestManifest(self, options): if isinstance(options.manifestFile, TestManifest): manifest = options.manifestFile elif options.manifestFile and os.path.isfile(options.manifestFile): manifestFileAbs = os.path.abspath(options.manifestFile) assert manifestFileAbs.startswith(SCRIPT_DIR) manifest = TestManifest([options.manifestFile], strict=False) elif options.manifestFile and os.path.isfile( os.path.join(SCRIPT_DIR, options.manifestFile) ): manifestFileAbs = os.path.abspath( os.path.join(SCRIPT_DIR, options.manifestFile) ) assert manifestFileAbs.startswith(SCRIPT_DIR) manifest = TestManifest([manifestFileAbs], strict=False) else: masterName = self.normflavor(options.flavor) + ".toml" masterPath = os.path.join(SCRIPT_DIR, self.testRoot, masterName) if not os.path.exists(masterPath): masterName = self.normflavor(options.flavor) + ".ini" masterPath = os.path.join(SCRIPT_DIR, self.testRoot, masterName) if os.path.exists(masterPath): manifest = TestManifest([masterPath], strict=False) else: manifest = None self.log.warning( "TestManifest masterPath %s does not exist" % masterPath ) return manifest def makeTestConfig(self, options): "Creates a test configuration file for customizing test execution." options.logFile = options.logFile.replace("\\", "\\\\") if ( "MOZ_HIDE_RESULTS_TABLE" in os.environ and os.environ["MOZ_HIDE_RESULTS_TABLE"] == "1" ): options.hideResultsTable = True # strip certain unnecessary items to avoid serialization errors in json.dumps() d = dict( (k, v) for k, v in options.__dict__.items() if (v is None) or isinstance(v, (six.string_types, numbers.Number)) ) d["testRoot"] = self.testRoot if options.jscov_dir_prefix: d["jscovDirPrefix"] = options.jscov_dir_prefix if not options.keep_open: d["closeWhenDone"] = "1" d["runFailures"] = False if options.runFailures: d["runFailures"] = True shutil.copy( os.path.join(SCRIPT_DIR, "ignorePrefs.json"), os.path.join(options.profilePath, "ignorePrefs.json"), ) d["ignorePrefsFile"] = "ignorePrefs.json" content = json.dumps(d) with open(os.path.join(options.profilePath, "testConfig.js"), "w") as config: config.write(content) def buildBrowserEnv(self, options, debugger=False, env=None): """build the environment variables for the specific test and operating system""" if mozinfo.info["asan"] and mozinfo.isLinux and mozinfo.bits == 64: useLSan = True else: useLSan = False browserEnv = self.environment( xrePath=options.xrePath, env=env, debugger=debugger, useLSan=useLSan ) if options.headless: browserEnv["MOZ_HEADLESS"] = "1" if not options.e10s: browserEnv["MOZ_FORCE_DISABLE_E10S"] = "1" if options.dmd: browserEnv["DMD"] = os.environ.get("DMD", "1") # bug 1443327: do not set MOZ_CRASHREPORTER_SHUTDOWN during browser-chrome # tests, since some browser-chrome tests test content process crashes; # also exclude non-e10s since at least one non-e10s mochitest is problematic if ( options.flavor == "browser" or not options.e10s ) and "MOZ_CRASHREPORTER_SHUTDOWN" in browserEnv: del browserEnv["MOZ_CRASHREPORTER_SHUTDOWN"] try: browserEnv.update( dict( parse_key_value( self.extraEnv, context="environment variable in manifest" ) ) ) except KeyValueParseError as e: self.log.error(str(e)) return None # These variables are necessary for correct application startup; change # via the commandline at your own risk. browserEnv["XPCOM_DEBUG_BREAK"] = "stack" # interpolate environment passed with options try: browserEnv.update( dict(parse_key_value(options.environment, context="--setenv")) ) except KeyValueParseError as e: self.log.error(str(e)) return None if ( "MOZ_PROFILER_STARTUP_FEATURES" not in browserEnv or "nativeallocations" not in browserEnv["MOZ_PROFILER_STARTUP_FEATURES"].split(",") ): # Only turn on the bloat log if the profiler's native allocation feature is # not enabled. The two are not compatible. browserEnv["XPCOM_MEM_BLOAT_LOG"] = self.leak_report_file # If profiling options are enabled, turn on the gecko profiler by using the # profiler environmental variables. if options.profiler: # The user wants to capture a profile, and automatically view it. The # profile will be saved to a temporary folder, then deleted after # opening in profiler.firefox.com. self.profiler_tempdir = tempfile.mkdtemp() browserEnv["MOZ_PROFILER_SHUTDOWN"] = os.path.join( self.profiler_tempdir, "mochitest-profile.json" ) browserEnv["MOZ_PROFILER_STARTUP"] = "1" if options.profilerSaveOnly: # The user wants to capture a profile, but only to save it. This defaults # to the MOZ_UPLOAD_DIR. browserEnv["MOZ_PROFILER_STARTUP"] = "1" if "MOZ_UPLOAD_DIR" in browserEnv: browserEnv["MOZ_PROFILER_SHUTDOWN"] = os.path.join( browserEnv["MOZ_UPLOAD_DIR"], "mochitest-profile.json" ) else: self.log.error( "--profiler-save-only was specified, but no MOZ_UPLOAD_DIR " "environment variable was provided. Please set this " "environment variable to a directory path in order to save " "a performance profile." ) return None try: gmp_path = self.getGMPPluginPath(options) if gmp_path is not None: browserEnv["MOZ_GMP_PATH"] = gmp_path except EnvironmentError: self.log.error("Could not find path to gmp-fake plugin!") return None if options.fatalAssertions: browserEnv["XPCOM_DEBUG_BREAK"] = "stack-and-abort" # Produce a mozlog, if setup (see MOZ_LOG global at the top of # this script). self.mozLogs = MOZ_LOG and "MOZ_UPLOAD_DIR" in os.environ if self.mozLogs: browserEnv["MOZ_LOG"] = MOZ_LOG return browserEnv def killNamedProc(self, pname, orphans=True): """Kill processes matching the given command name""" self.log.info("Checking for %s processes..." % pname) if HAVE_PSUTIL: for proc in psutil.process_iter(): try: if proc.name() == pname: procd = proc.as_dict(attrs=["pid", "ppid", "name", "username"]) if proc.ppid() == 1 or not orphans: self.log.info("killing %s" % procd) killPid(proc.pid, self.log) else: self.log.info("NOT killing %s (not an orphan?)" % procd) except Exception as e: self.log.info( "Warning: Unable to kill process %s: %s" % (pname, str(e)) ) # may not be able to access process info for all processes continue else: def _psInfo(_, line): if pname in line: self.log.info(line) mozprocess.run_and_wait( ["ps", "-f"], output_line_handler=_psInfo, ) def _psKill(_, line): parts = line.split() if len(parts) == 3 and parts[0].isdigit(): pid = int(parts[0]) ppid = int(parts[1]) if parts[2] == pname: if ppid == 1 or not orphans: self.log.info("killing %s (pid %d)" % (pname, pid)) killPid(pid, self.log) else: self.log.info( "NOT killing %s (pid %d) (not an orphan?)" % (pname, pid) ) mozprocess.run_and_wait( ["ps", "-o", "pid,ppid,comm"], output_line_handler=_psKill, ) def execute_start_script(self): if not self.start_script or not self.marionette: return if os.path.isfile(self.start_script): with open(self.start_script, "r") as fh: script = fh.read() else: script = self.start_script with self.marionette.using_context("chrome"): return self.marionette.execute_script( script, script_args=(self.start_script_kwargs,) ) def fillCertificateDB(self, options): # TODO: move -> mozprofile: # https://bugzilla.mozilla.org/show_bug.cgi?id=746243#c35 pwfilePath = os.path.join(options.profilePath, ".crtdbpw") with open(pwfilePath, "w") as pwfile: pwfile.write("\n") # Pre-create the certification database for the profile env = self.environment(xrePath=options.xrePath) env["LD_LIBRARY_PATH"] = options.xrePath bin_suffix = mozinfo.info.get("bin_suffix", "") certutil = os.path.join(options.utilityPath, "certutil" + bin_suffix) pk12util = os.path.join(options.utilityPath, "pk12util" + bin_suffix) toolsEnv = env if mozinfo.info["asan"]: # Disable leak checking when running these tools toolsEnv["ASAN_OPTIONS"] = "detect_leaks=0" if mozinfo.info["tsan"]: # Disable race checking when running these tools toolsEnv["TSAN_OPTIONS"] = "report_bugs=0" if self.certdbNew: # android uses the new DB formats exclusively certdbPath = "sql:" + options.profilePath else: # desktop seems to use the old certdbPath = options.profilePath # certutil.exe depends on some DLLs in the app directory # When running tests against an MSIX-installed Firefox, these DLLs # cannot be used out of the install directory, they must be copied # elsewhere first. if "WindowsApps" in options.app: install_dir = os.path.dirname(options.app) for f in os.listdir(install_dir): if f.endswith(".dll"): shutil.copy(os.path.join(install_dir, f), options.utilityPath) status = call( [certutil, "-N", "-d", certdbPath, "-f", pwfilePath], env=toolsEnv ) if status: return status # Walk the cert directory and add custom CAs and client certs files = os.listdir(options.certPath) for item in files: root, ext = os.path.splitext(item) if ext == ".ca": trustBits = "CT,," if root.endswith("-object"): trustBits = "CT,,CT" call( [ certutil, "-A", "-i", os.path.join(options.certPath, item), "-d", certdbPath, "-f", pwfilePath, "-n", root, "-t", trustBits, ], env=toolsEnv, ) elif ext == ".client": call( [ pk12util, "-i", os.path.join(options.certPath, item), "-w", pwfilePath, "-d", certdbPath, ], env=toolsEnv, ) os.unlink(pwfilePath) return 0 def findFreePort(self, type): with closing(socket.socket(socket.AF_INET, type)) as s: s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) s.bind(("127.0.0.1", 0)) return s.getsockname()[1] def proxy(self, options): # proxy # use SSL port for legacy compatibility; see # - https://bugzilla.mozilla.org/show_bug.cgi?id=688667#c66 # - https://bugzilla.mozilla.org/show_bug.cgi?id=899221 # - https://github.com/mozilla/mozbase/commit/43f9510e3d58bfed32790c82a57edac5f928474d # 'ws': str(self.webSocketPort) proxyOptions = { "remote": options.webServer, "http": options.httpPort, "https": options.sslPort, "ws": options.sslPort, } if options.useHttp3Server: options.dohServerPort = self.findFreePort(socket.SOCK_STREAM) options.http3ServerPort = self.findFreePort(socket.SOCK_DGRAM) proxyOptions["dohServerPort"] = options.dohServerPort self.log.info("use doh server at port: %d" % options.dohServerPort) self.log.info("use http3 server at port: %d" % options.http3ServerPort) elif options.useHttp2Server: options.dohServerPort = self.findFreePort(socket.SOCK_STREAM) options.http2ServerPort = self.findFreePort(socket.SOCK_STREAM) proxyOptions["dohServerPort"] = options.dohServerPort self.log.info("use doh server at port: %d" % options.dohServerPort) self.log.info("use http2 server at port: %d" % options.http2ServerPort) return proxyOptions def merge_base_profiles(self, options, category): """Merge extra profile data from testing/profiles.""" # In test packages used in CI, the profile_data directory is installed # in the SCRIPT_DIR. profile_data_dir = os.path.join(SCRIPT_DIR, "profile_data") # If possible, read profile data from topsrcdir. This prevents us from # requiring a re-build to pick up newly added extensions in the # /extensions directory. if build_obj: path = os.path.join(build_obj.topsrcdir, "testing", "profiles") if os.path.isdir(path): profile_data_dir = path # Still not found? Look for testing/profiles relative to testing/mochitest. if not os.path.isdir(profile_data_dir): path = os.path.abspath(os.path.join(SCRIPT_DIR, "..", "profiles")) if os.path.isdir(path): profile_data_dir = path with open(os.path.join(profile_data_dir, "profiles.json"), "r") as fh: base_profiles = json.load(fh)[category] # values to use when interpolating preferences interpolation = { "server": "%s:%s" % (options.webServer, options.httpPort), } for profile in base_profiles: path = os.path.join(profile_data_dir, profile) self.profile.merge(path, interpolation=interpolation) @property def conditioned_profile_copy(self): """Returns a copy of the original conditioned profile that was created.""" condprof_copy = os.path.join(tempfile.mkdtemp(), "profile") shutil.copytree( self.conditioned_profile_dir, condprof_copy, ignore=shutil.ignore_patterns("lock"), ) self.log.info("Created a conditioned-profile copy: %s" % condprof_copy) return condprof_copy def downloadConditionedProfile(self, profile_scenario, app): from condprof.client import get_profile from condprof.util import get_current_platform, get_version if self.conditioned_profile_dir: # We already have a directory, so provide a copy that # will get deleted after it's done with return self.conditioned_profile_copy temp_download_dir = tempfile.mkdtemp() # Call condprof's client API to yield our platform-specific # conditioned-profile binary platform = get_current_platform() if not profile_scenario: profile_scenario = "settled" version = get_version(app) try: cond_prof_target_dir = get_profile( temp_download_dir, platform, profile_scenario, repo="mozilla-central", version=version, retries=2, # quicker failure ) except Exception: if version is None: # any other error is a showstopper self.log.critical("Could not get the conditioned profile") traceback.print_exc() raise version = None try: self.log.info("retrying a profile with no version specified") cond_prof_target_dir = get_profile( temp_download_dir, platform, profile_scenario, repo="mozilla-central", version=version, ) except Exception: self.log.critical("Could not get the conditioned profile") traceback.print_exc() raise # Now get the full directory path to our fetched conditioned profile self.conditioned_profile_dir = os.path.join( temp_download_dir, cond_prof_target_dir ) if not os.path.exists(cond_prof_target_dir): self.log.critical( "Can't find target_dir {}, from get_profile()" "temp_download_dir {}, platform {}, scenario {}".format( cond_prof_target_dir, temp_download_dir, platform, profile_scenario ) ) raise OSError self.log.info( "Original self.conditioned_profile_dir is now set: {}".format( self.conditioned_profile_dir ) ) return self.conditioned_profile_copy def buildProfile(self, options): """create the profile and add optional chrome bits and files if requested""" # get extensions to install extensions = self.getExtensionsToInstall(options) # Whitelist the _tests directory (../..) so that TESTING_JS_MODULES work tests_dir = os.path.dirname(os.path.dirname(SCRIPT_DIR)) sandbox_whitelist_paths = [tests_dir] + options.sandboxReadWhitelist if platform.system() == "Linux" or platform.system() in ( "Windows", "Microsoft", ): # Trailing slashes are needed to indicate directories on Linux and Windows sandbox_whitelist_paths = [ os.path.join(p, "") for p in sandbox_whitelist_paths ] if options.conditionedProfile: if options.profilePath and os.path.exists(options.profilePath): shutil.rmtree(options.profilePath, ignore_errors=True) options.profilePath = self.downloadConditionedProfile("full", options.app) # This is causing `certutil -N -d -f`` to not use -f (pwd file) try: os.remove(os.path.join(options.profilePath, "key4.db")) except Exception as e: self.log.info( "Caught exception while removing key4.db" "during setup of conditioned profile: %s" % e ) # Create the profile self.profile = Profile( profile=options.profilePath, addons=extensions, locations=self.locations, proxy=self.proxy(options), whitelistpaths=sandbox_whitelist_paths, ) # Fix options.profilePath for legacy consumers. options.profilePath = self.profile.profile manifest = self.addChromeToProfile(options) self.copyExtraFilesToProfile(options) # create certificate database for the profile # TODO: this should really be upstreamed somewhere, maybe mozprofile certificateStatus = self.fillCertificateDB(options) if certificateStatus: self.log.error( "TEST-UNEXPECTED-FAIL | runtests.py | Certificate integration failed" ) return None # Set preferences in the following order (latter overrides former): # 1) Preferences from base profile (e.g from testing/profiles) # 2) Prefs hardcoded in this function # 3) Prefs from --setpref # Prefs from base profiles self.merge_base_profiles(options, "mochitest") # Hardcoded prefs (TODO move these into a base profile) prefs = { # Enable tracing output for detailed failures in case of # failing connection attempts, and hangs (bug 1397201) "remote.log.level": "Trace", # Disable async font fallback, because the unpredictable # extra reflow it can trigger (potentially affecting a later # test) results in spurious intermittent failures. "gfx.font_rendering.fallback.async": False, } test_timeout = None if options.flavor == "browser" and options.timeout: test_timeout = options.timeout # browser-chrome tests use a fairly short default timeout of 45 seconds; # this is sometimes too short on asan and debug, where we expect reduced # performance. if ( (mozinfo.info["asan"] or mozinfo.info["debug"]) and options.flavor == "browser" and options.timeout is None ): self.log.info("Increasing default timeout to 90 seconds (asan or debug)") test_timeout = 90 # tsan builds need even more time if ( mozinfo.info["tsan"] and options.flavor == "browser" and options.timeout is None ): self.log.info("Increasing default timeout to 120 seconds (tsan)") test_timeout = 120 if mozinfo.info["os"] == "win" and mozinfo.info["processor"] == "aarch64": test_timeout = self.DEFAULT_TIMEOUT * 4 self.log.info( "Increasing default timeout to {} seconds (win aarch64)".format( test_timeout ) ) if "MOZ_CHAOSMODE=0xfb" in options.environment and test_timeout: test_timeout *= 2 self.log.info( "Increasing default timeout to {} seconds (MOZ_CHAOSMODE)".format( test_timeout ) ) if test_timeout: prefs["testing.browserTestHarness.timeout"] = test_timeout if getattr(self, "testRootAbs", None): prefs["mochitest.testRoot"] = self.testRootAbs # See if we should use fake media devices. if options.useTestMediaDevices: prefs["media.audio_loopback_dev"] = self.mediaDevices["audio"]["name"] prefs["media.video_loopback_dev"] = self.mediaDevices["video"]["name"] prefs["media.cubeb.output_device"] = "Null Output" prefs["media.volume_scale"] = "1.0" self.gstForV4l2loopbackProcess = self.mediaDevices["video"]["process"] self.profile.set_preferences(prefs) # Extra prefs from --setpref self.profile.set_preferences(self.extraPrefs) return manifest def getGMPPluginPath(self, options): if options.gmp_path: return options.gmp_path gmp_parentdirs = [ # For local builds, GMP plugins will be under dist/bin. options.xrePath, # For packaged builds, GMP plugins will get copied under # $profile/plugins. os.path.join(self.profile.profile, "plugins"), ] gmp_subdirs = [ os.path.join("gmp-fake", "1.0"), os.path.join("gmp-fakeopenh264", "1.0"), os.path.join("gmp-clearkey", "0.1"), ] gmp_paths = [ os.path.join(parent, sub) for parent in gmp_parentdirs for sub in gmp_subdirs if os.path.isdir(os.path.join(parent, sub)) ] if not gmp_paths: # This is fatal for desktop environments. raise EnvironmentError("Could not find test gmp plugins") return os.pathsep.join(gmp_paths) def cleanup(self, options, final=False): """remove temporary files, profile and virtual audio input device""" if hasattr(self, "manifest") and self.manifest is not None: if os.path.exists(self.manifest): os.remove(self.manifest) if hasattr(self, "profile"): del self.profile if hasattr(self, "extraTestsDirs"): for d in self.extraTestsDirs: if os.path.exists(d): shutil.rmtree(d) if options.pidFile != "" and os.path.exists(options.pidFile): try: os.remove(options.pidFile) if os.path.exists(options.pidFile + ".xpcshell.pid"): os.remove(options.pidFile + ".xpcshell.pid") except Exception: self.log.warning( "cleaning up pidfile '%s' was unsuccessful from the test harness" % options.pidFile ) options.manifestFile = None if hasattr(self, "virtualDeviceIdList"): pactl = spawn.find_executable("pactl") if not pactl: self.log.error("Could not find pactl on system") return None for id in self.virtualDeviceIdList: try: subprocess.check_call([pactl, "unload-module", str(id)]) except subprocess.CalledProcessError: self.log.error( "Could not remove pulse module with id {}".format(id) ) return None self.virtualDeviceIdList = [] def dumpScreen(self, utilityPath): if self.haveDumpedScreen: self.log.info( "Not taking screenshot here: see the one that was previously logged" ) return self.haveDumpedScreen = True dump_screen(utilityPath, self.log) def killAndGetStack(self, processPID, utilityPath, debuggerInfo, dump_screen=False): """ Kill the process, preferrably in a way that gets us a stack trace. Also attempts to obtain a screenshot before killing the process if specified. """ self.log.info("Killing process: %s" % processPID) if dump_screen: self.dumpScreen(utilityPath) if mozinfo.info.get("crashreporter", True) and not debuggerInfo: try: minidump_path = os.path.join(self.profile.profile, "minidumps") mozcrash.kill_and_get_minidump(processPID, minidump_path, utilityPath) except OSError: # https://bugzilla.mozilla.org/show_bug.cgi?id=921509 self.log.info("Can't trigger Breakpad, process no longer exists") return self.log.info("Can't trigger Breakpad, just killing process") killPid(processPID, self.log) def extract_child_pids(self, process_log, parent_pid=None): """Parses the given log file for the pids of any processes launched by the main process and returns them as a list. If parent_pid is provided, and psutil is available, returns children of parent_pid according to psutil. """ rv = [] if parent_pid and HAVE_PSUTIL: self.log.info("Determining child pids from psutil...") try: rv = [p.pid for p in psutil.Process(parent_pid).children()] self.log.info(str(rv)) except psutil.NoSuchProcess: self.log.warning("Failed to lookup children of pid %d" % parent_pid) rv = set(rv) pid_re = re.compile(r"==> process \d+ launched child process (\d+)") with open(process_log) as fd: for line in fd: self.log.info(line.rstrip()) m = pid_re.search(line) if m: rv.add(int(m.group(1))) return rv def checkForZombies(self, processLog, utilityPath, debuggerInfo): """Look for hung processes""" if not os.path.exists(processLog): self.log.info("Automation Error: PID log not found: %s" % processLog) # Whilst no hung process was found, the run should still display as # a failure return True # scan processLog for zombies self.log.info("zombiecheck | Reading PID log: %s" % processLog) processList = self.extract_child_pids(processLog) # kill zombies foundZombie = False for processPID in processList: self.log.info( "zombiecheck | Checking for orphan process with PID: %d" % processPID ) if isPidAlive(processPID): foundZombie = True self.log.error( "TEST-UNEXPECTED-FAIL | zombiecheck | child process " "%d still alive after shutdown" % processPID ) self.killAndGetStack( processPID, utilityPath, debuggerInfo, dump_screen=not debuggerInfo ) return foundZombie def checkForRunningBrowsers(self): firefoxes = "" if HAVE_PSUTIL: attrs = ["pid", "ppid", "name", "cmdline", "username"] for proc in psutil.process_iter(): try: if "firefox" in proc.name(): firefoxes = "%s%s\n" % (firefoxes, proc.as_dict(attrs=attrs)) except Exception: # may not be able to access process info for all processes continue if len(firefoxes) > 0: # In automation, this warning is unexpected and should be investigated. # In local testing, this is probably okay, as long as the browser is not # running a marionette server. self.log.warning("Found 'firefox' running before starting test browser!") self.log.warning(firefoxes) def runApp( self, testUrl, env, app, profile, extraArgs, utilityPath, debuggerInfo=None, valgrindPath=None, valgrindArgs=None, valgrindSuppFiles=None, symbolsPath=None, timeout=-1, detectShutdownLeaks=False, screenshotOnFail=False, bisectChunk=None, restartAfterFailure=False, marionette_args=None, e10s=True, runFailures=False, crashAsPass=False, currentManifest=None, ): """ Run the app, log the duration it took to execute, return the status code. Kills the app if it runs for longer than |maxTime| seconds, or outputs nothing for |timeout| seconds. """ # It can't be the case that both a with-debugger and an # on-Valgrind run have been requested. doTests() should have # already excluded this possibility. assert not (valgrindPath and debuggerInfo) # debugger information interactive = False debug_args = None if debuggerInfo: interactive = debuggerInfo.interactive debug_args = [debuggerInfo.path] + debuggerInfo.args # Set up Valgrind arguments. if valgrindPath: interactive = False valgrindArgs_split = ( [] if valgrindArgs is None else shlex.split(valgrindArgs) ) valgrindSuppFiles_final = [] if valgrindSuppFiles is not None: valgrindSuppFiles_final = [ "--suppressions=" + path for path in valgrindSuppFiles.split(",") ] debug_args = ( [valgrindPath] + mozdebug.get_default_valgrind_args() + valgrindArgs_split + valgrindSuppFiles_final ) # fix default timeout if timeout == -1: timeout = self.DEFAULT_TIMEOUT # Note in the log if running on Valgrind if valgrindPath: self.log.info( "runtests.py | Running on Valgrind. " + "Using timeout of %d seconds." % timeout ) # copy env so we don't munge the caller's environment env = env.copy() # Used to defer a possible IOError exception from Marionette marionette_exception = None temp_file_paths = [] # make sure we clean up after ourselves. try: # set process log environment variable tmpfd, processLog = tempfile.mkstemp(suffix="pidlog") os.close(tmpfd) env["MOZ_PROCESS_LOG"] = processLog if debuggerInfo: # If a debugger is attached, don't use timeouts, and don't # capture ctrl-c. timeout = None signal.signal(signal.SIGINT, lambda sigid, frame: None) # build command line cmd = os.path.abspath(app) args = list(extraArgs) args.append("-marionette") # TODO: mozrunner should use -foreground at least for mac # https://bugzilla.mozilla.org/show_bug.cgi?id=916512 args.append("-foreground") self.start_script_kwargs["testUrl"] = testUrl or "about:blank" if detectShutdownLeaks: env["MOZ_LOG"] = ( env["MOZ_LOG"] + "," if env["MOZ_LOG"] else "" ) + "DocShellAndDOMWindowLeak:3" shutdownLeaks = ShutdownLeaks(self.log) else: shutdownLeaks = None if mozinfo.info["asan"] and mozinfo.isLinux and mozinfo.bits == 64: lsanLeaks = LSANLeaks(self.log) else: lsanLeaks = None # create an instance to process the output outputHandler = self.OutputHandler( harness=self, utilityPath=utilityPath, symbolsPath=symbolsPath, dump_screen_on_timeout=not debuggerInfo, dump_screen_on_fail=screenshotOnFail, shutdownLeaks=shutdownLeaks, lsanLeaks=lsanLeaks, bisectChunk=bisectChunk, restartAfterFailure=restartAfterFailure, ) def timeoutHandler(): browserProcessId = outputHandler.browserProcessId self.handleTimeout( timeout, proc, utilityPath, debuggerInfo, browserProcessId, processLog, ) kp_kwargs = { "kill_on_timeout": False, "cwd": SCRIPT_DIR, "onTimeout": [timeoutHandler], } kp_kwargs["processOutputLine"] = [outputHandler] self.checkForRunningBrowsers() # create mozrunner instance and start the system under test process self.lastTestSeen = self.test_name self.lastManifest = currentManifest startTime = datetime.now() runner_cls = mozrunner.runners.get( mozinfo.info.get("appname", "firefox"), mozrunner.Runner ) runner = runner_cls( profile=self.profile, binary=cmd, cmdargs=args, env=env, process_class=mozprocess.ProcessHandlerMixin, process_args=kp_kwargs, ) # start the runner try: runner.start( debug_args=debug_args, interactive=interactive, outputTimeout=timeout, ) proc = runner.process_handler self.log.info("runtests.py | Application pid: %d" % proc.pid) gecko_id = "GECKO(%d)" % proc.pid self.log.process_start(gecko_id) self.message_logger.gecko_id = gecko_id except PermissionError: # treat machine as bad, return return TBPL_RETRY, "Failure to launch browser" except Exception as e: raise e # unknown error try: # start marionette and kick off the tests marionette_args = marionette_args or {} self.marionette = Marionette(**marionette_args) self.marionette.start_session() # install specialpowers and mochikit addons addons = Addons(self.marionette) if self.staged_addons: for addon_path in self.staged_addons: if not os.path.isdir(addon_path): self.log.error( "TEST-UNEXPECTED-FAIL | invalid setup: missing extension at %s" % addon_path ) return 1, self.lastTestSeen temp_addon_path = create_zip(addon_path) temp_file_paths.append(temp_addon_path) addons.install(temp_addon_path) self.execute_start_script() # an open marionette session interacts badly with mochitest, # delete it until we figure out why. self.marionette.delete_session() del self.marionette except IOError: # Any IOError as thrown by Marionette means that something is # wrong with the process, like a crash or the socket is no # longer open. We defer raising this specific error so that # post-test checks for leaks and crashes are performed and # reported first. marionette_exception = sys.exc_info() # wait until app is finished # XXX copy functionality from # https://github.com/mozilla/mozbase/blob/master/mozrunner/mozrunner/runner.py#L61 # until bug 913970 is fixed regarding mozrunner `wait` not returning status # see https://bugzilla.mozilla.org/show_bug.cgi?id=913970 self.log.info("runtests.py | Waiting for browser...") status = proc.wait() if status is None: self.log.warning( "runtests.py | Failed to get app exit code - running/crashed?" ) # must report an integer to process_exit() status = 0 self.log.process_exit("Main app process", status) runner.process_handler = None # finalize output handler outputHandler.finish() # record post-test information if status: # no need to keep return code 137, 245, etc. status = 1 self.message_logger.dump_buffered() msg = "application terminated with exit code %s" % status # self.message_logger.is_test_running indicates we need to send a test_end if crashAsPass and self.message_logger.is_test_running: # this works for browser-chrome, mochitest-plain has status=0 message = { "action": "test_end", "status": "CRASH", "expected": "CRASH", "thread": None, "pid": None, "source": "mochitest", "time": int(time.time()) * 1000, "test": self.lastTestSeen, "message": msg, } # need to send a test_end in order to have mozharness process messages properly # this requires a custom message vs log.error/log.warning/etc. self.message_logger.process_message(message) else: self.lastTestSeen = ( currentManifest or "Main app process exited normally" ) self.log.info( "runtests.py | Application ran for: %s" % str(datetime.now() - startTime) ) # Do a final check for zombie child processes. zombieProcesses = self.checkForZombies( processLog, utilityPath, debuggerInfo ) # check for crashes quiet = False if crashAsPass: quiet = True minidump_path = os.path.join(self.profile.profile, "minidumps") crash_count = mozcrash.log_crashes( self.log, minidump_path, symbolsPath, test=self.lastTestSeen, quiet=quiet, ) expected = None if crashAsPass or crash_count > 0: # self.message_logger.is_test_running indicates we need a test_end message if self.message_logger.is_test_running: # this works for browser-chrome, mochitest-plain has status=0 expected = "CRASH" if crashAsPass: status = 0 elif crash_count or zombieProcesses: if self.message_logger.is_test_running: expected = "PASS" status = 1 if expected: # send this out so we always wrap up the test-end message message = { "action": "test_end", "status": "CRASH", "expected": expected, "thread": None, "pid": None, "source": "mochitest", "time": int(time.time()) * 1000, "test": self.lastTestSeen, "message": "application terminated with exit code %s" % status, } # need to send a test_end in order to have mozharness process messages properly # this requires a custom message vs log.error/log.warning/etc. self.message_logger.process_message(message) finally: # cleanup if os.path.exists(processLog): os.remove(processLog) for p in temp_file_paths: os.remove(p) if marionette_exception is not None: exc, value, tb = marionette_exception six.reraise(exc, value, tb) return status, self.lastTestSeen def initializeLooping(self, options): """ This method is used to clear the contents before each run of for loop. This method is used for --run-by-dir and --bisect-chunk. """ if options.conditionedProfile: if options.profilePath and os.path.exists(options.profilePath): shutil.rmtree(options.profilePath, ignore_errors=True) if options.manifestFile and os.path.exists(options.manifestFile): os.remove(options.manifestFile) self.expectedError.clear() self.result.clear() options.manifestFile = None options.profilePath = None def initializeVirtualAudioDevices(self): """ Configure the system to have a number of virtual audio devices: 2 output devices, and 4 input devices that each produce a tone at a particular frequency. This method is only currently implemented for Linux. """ if not mozinfo.isLinux: return pactl = spawn.find_executable("pactl") if not pactl: self.log.error("Could not find pactl on system") return def getModuleIds(moduleName): o = subprocess.check_output([pactl, "list", "modules", "short"]) list = [] for input in o.splitlines(): device = input.decode().split("\t") if device[1] == moduleName: list.append(int(device[0])) return list OUTPUT_DEVICES_COUNT = 2 INPUT_DEVICES_COUNT = 4 DEVICES_BASE_FREQUENCY = 110 # Hz # If the device are already present, find their id and return early outputDeviceIdList = getModuleIds("module-null-sink") inputDeviceIdList = getModuleIds("module-sine-source") if ( len(outputDeviceIdList) == OUTPUT_DEVICES_COUNT and len(inputDeviceIdList) == INPUT_DEVICES_COUNT ): self.virtualDeviceIdList = outputDeviceIdList + inputDeviceIdList return else: # Remove any existing devices and reinitialize properly for id in outputDeviceIdList + inputDeviceIdList: try: subprocess.check_call([pactl, "unload-module", str(id)]) except subprocess.CalledProcessError: log.error("Could not remove pulse module with id {}".format(id)) return None idList = [] command = [pactl, "load-module", "module-null-sink"] try: # device for "media.audio_loopback_dev" pref o = subprocess.check_output(command + ["rate=44100"]) idList.append(int(o)) except subprocess.CalledProcessError: self.log.error("Could not load module-null-sink") try: o = subprocess.check_output( command + [ "rate=48000", "sink_properties='device.description=\"48000 Hz Null Output\"'", ] ) idList.append(int(o)) except subprocess.CalledProcessError: self.log.error("Could not load module-null-sink at rate=48000") # We want quite a number of input devices, each with a different tone # frequency and device name so that we can recognize them easily during # testing. command = [pactl, "load-module", "module-sine-source", "rate=44100"] for i in range(1, INPUT_DEVICES_COUNT + 1): freq = i * DEVICES_BASE_FREQUENCY complete_command = command + [ "source_name=sine-{}".format(freq), "frequency={}".format(freq), ] try: o = subprocess.check_output(complete_command) idList.append(int(o)) except subprocess.CalledProcessError: self.log.error( "Could not create device with module-sine-source" " (freq={})".format(freq) ) self.virtualDeviceIdList = idList def normalize_paths(self, paths): # Normalize test paths so they are relative to test root norm_paths = [] for p in paths: abspath = os.path.abspath(os.path.join(self.oldcwd, p)) if abspath.startswith(self.testRootAbs): norm_paths.append(os.path.relpath(abspath, self.testRootAbs)) else: norm_paths.append(p) return norm_paths def runMochitests(self, options, testsToRun, manifestToFilter=None): "This is a base method for calling other methods in this class for --bisect-chunk." # Making an instance of bisect class for --bisect-chunk option. bisect = bisection.Bisect(self) finished = False status = 0 bisection_log = 0 while not finished: if options.bisectChunk: testsToRun = bisect.pre_test(options, testsToRun, status) # To inform that we are in the process of bisection, and to # look for bleedthrough if options.bisectChunk != "default" and not bisection_log: self.log.error( "TEST-UNEXPECTED-FAIL | Bisection | Please ignore repeats " "and look for 'Bleedthrough' (if any) at the end of " "the failure list" ) bisection_log = 1 result = self.doTests(options, testsToRun, manifestToFilter) if result == TBPL_RETRY: # terminate task return result if options.bisectChunk: status = bisect.post_test(options, self.expectedError, self.result) elif options.restartAfterFailure: # NOTE: ideally browser will halt on first failure, then this will always be the last test if not self.expectedError: status = -1 else: firstFail = len(testsToRun) for key in self.expectedError: full_key = [x for x in testsToRun if key in x] if full_key: if testsToRun.index(full_key[0]) < firstFail: firstFail = testsToRun.index(full_key[0]) testsToRun = testsToRun[firstFail + 1 :] if testsToRun == []: status = -1 else: status = -1 if status == -1: finished = True # We need to print the summary only if options.bisectChunk has a value. # Also we need to make sure that we do not print the summary in between # running tests via --run-by-dir. if options.bisectChunk and options.bisectChunk in self.result: bisect.print_summary() return result def groupTestsByScheme(self, tests): """ split tests into groups by schemes. test is classified as http if no scheme specified """ httpTests = [] httpsTests = [] for test in tests: if not test.get("scheme") or test.get("scheme") == "http": httpTests.append(test) elif test.get("scheme") == "https": httpsTests.append(test) return {"http": httpTests, "https": httpsTests} def verifyTests(self, options): """ Support --verify mode: Run test(s) many times in a variety of configurations/environments in an effort to find intermittent failures. """ # Number of times to repeat test(s) when running with --repeat VERIFY_REPEAT = 10 # Number of times to repeat test(s) when running test in VERIFY_REPEAT_SINGLE_BROWSER = 5 def step1(): options.repeat = VERIFY_REPEAT options.keep_open = False options.runUntilFailure = True options.profilePath = None options.comparePrefs = True result = self.runTests(options) result = result or (-2 if self.countfail > 0 else 0) self.message_logger.finish() return result def step2(): options.repeat = 0 options.keep_open = False options.runUntilFailure = False for i in range(VERIFY_REPEAT_SINGLE_BROWSER): options.profilePath = None result = self.runTests(options) result = result or (-2 if self.countfail > 0 else 0) self.message_logger.finish() if result != 0: break return result def step3(): options.repeat = VERIFY_REPEAT options.keep_open = False options.runUntilFailure = True options.environment.append("MOZ_CHAOSMODE=0xfb") options.profilePath = None result = self.runTests(options) options.environment.remove("MOZ_CHAOSMODE=0xfb") result = result or (-2 if self.countfail > 0 else 0) self.message_logger.finish() return result def step4(): options.repeat = 0 options.keep_open = False options.runUntilFailure = False options.environment.append("MOZ_CHAOSMODE=0xfb") for i in range(VERIFY_REPEAT_SINGLE_BROWSER): options.profilePath = None result = self.runTests(options) result = result or (-2 if self.countfail > 0 else 0) self.message_logger.finish() if result != 0: break options.environment.remove("MOZ_CHAOSMODE=0xfb") return result def fission_step(fission_pref): if fission_pref not in options.extraPrefs: options.extraPrefs.append(fission_pref) options.keep_open = False options.runUntilFailure = True options.profilePath = None result = self.runTests(options) result = result or (-2 if self.countfail > 0 else 0) self.message_logger.finish() return result def fission_step1(): return fission_step("fission.autostart=false") def fission_step2(): return fission_step("fission.autostart=true") if options.verify_fission: steps = [ ("1. Run each test without fission.", fission_step1), ("2. Run each test with fission.", fission_step2), ] else: steps = [ ("1. Run each test %d times in one browser." % VERIFY_REPEAT, step1), ( "2. Run each test %d times in a new browser each time." % VERIFY_REPEAT_SINGLE_BROWSER, step2, ), ( "3. Run each test %d times in one browser, in chaos mode." % VERIFY_REPEAT, step3, ), ( "4. Run each test %d times in a new browser each time, " "in chaos mode." % VERIFY_REPEAT_SINGLE_BROWSER, step4, ), ] stepResults = {} for descr, step in steps: stepResults[descr] = "not run / incomplete" startTime = datetime.now() maxTime = timedelta(seconds=options.verify_max_time) finalResult = "PASSED" for descr, step in steps: if (datetime.now() - startTime) > maxTime: self.log.info("::: Test verification is taking too long: Giving up!") self.log.info( "::: So far, all checks passed, but not all checks were run." ) break self.log.info(":::") self.log.info('::: Running test verification step "%s"...' % descr) self.log.info(":::") result = step() if result != 0: stepResults[descr] = "FAIL" finalResult = "FAILED!" break stepResults[descr] = "Pass" self.logPreamble([]) self.log.info(":::") self.log.info("::: Test verification summary for:") self.log.info(":::") tests = self.getActiveTests(options) for test in tests: self.log.info("::: " + test["path"]) self.log.info(":::") for descr in sorted(stepResults.keys()): self.log.info("::: %s : %s" % (descr, stepResults[descr])) self.log.info(":::") self.log.info("::: Test verification %s" % finalResult) self.log.info(":::") return 0 def runTests(self, options): """Prepare, configure, run tests and cleanup""" self.extraPrefs = parse_preferences(options.extraPrefs) self.extraPrefs["fission.autostart"] = not options.disable_fission # for test manifest parsing. mozinfo.update( { "a11y_checks": options.a11y_checks, "e10s": options.e10s, "fission": not options.disable_fission, "headless": options.headless, "http3": options.useHttp3Server, "http2": options.useHttp2Server, # Until the test harness can understand default pref values, # (https://bugzilla.mozilla.org/show_bug.cgi?id=1577912) this value # should by synchronized with the default pref value indicated in # StaticPrefList.yaml. # # Currently for automation, the pref defaults to true (but can be # overridden with --setpref). "serviceworker_e10s": True, "sessionHistoryInParent": not options.disable_fission or not self.extraPrefs.get( "fission.disableSessionHistoryInParent", mozinfo.info["os"] == "android", ), "socketprocess_e10s": self.extraPrefs.get( "network.process.enabled", False ), "socketprocess_networking": self.extraPrefs.get( "network.http.network_access_on_socket_process.enabled", False ), "swgl": self.extraPrefs.get("gfx.webrender.software", False), "verify": options.verify, "verify_fission": options.verify_fission, "webgl_ipc": self.extraPrefs.get("webgl.out-of-process", False), "wmfme": ( self.extraPrefs.get("media.wmf.media-engine.enabled", 0) and self.extraPrefs.get( "media.wmf.media-engine.channel-decoder.enabled", False ) ), "mda_gpu": self.extraPrefs.get( "media.hardware-video-decoding.force-enabled", False ), "xorigin": options.xOriginTests, "condprof": options.conditionedProfile, "msix": "WindowsApps" in options.app, } ) if not self.mozinfo_variables_shown: self.mozinfo_variables_shown = True self.log.info( "These variables are available in the mozinfo environment and " "can be used to skip tests conditionally:" ) for info in sorted(mozinfo.info.items(), key=lambda item: item[0]): self.log.info(" {key}: {value}".format(key=info[0], value=info[1])) self.setTestRoot(options) # Despite our efforts to clean up servers started by this script, in practice # we still see infrequent cases where a process is orphaned and interferes # with future tests, typically because the old server is keeping the port in use. # Try to avoid those failures by checking for and killing servers before # trying to start new ones. self.killNamedProc("ssltunnel") self.killNamedProc("xpcshell") if options.cleanupCrashes: mozcrash.cleanup_pending_crash_reports() tests = self.getActiveTests(options) self.logPreamble(tests) if mozinfo.info["fission"] and not mozinfo.info["e10s"]: # Make sure this is logged *after* suite_start so it gets associated with the # current suite in the summary formatters. self.log.error("Fission is not supported without e10s.") return 1 tests = [t for t in tests if "disabled" not in t] # Until we have all green, this does not run on a11y (for perf reasons) if not options.runByManifest: result = self.runMochitests(options, [t["path"] for t in tests]) self.handleShutdownProfile(options) return result # code for --run-by-manifest manifests = set(t["manifest"].replace("\\", "/") for t in tests) result = 0 origPrefs = self.extraPrefs.copy() for m in sorted(manifests): self.log.group_start(name=m) self.log.info("Running manifest: {}".format(m)) self.message_logger.setManifest(m) args = list(self.args_by_manifest[m])[0] self.extraArgs = [] if args: for arg in args.strip().split(): # Split off the argument value if available so that both # name and value will be set individually self.extraArgs.extend(arg.split("=")) self.log.info( "The following arguments will be set:\n {}".format( "\n ".join(self.extraArgs) ) ) prefs = list(self.prefs_by_manifest[m])[0] self.extraPrefs = origPrefs.copy() if prefs: prefs = prefs.strip().split() self.log.info( "The following extra prefs will be set:\n {}".format( "\n ".join(prefs) ) ) self.extraPrefs.update(parse_preferences(prefs)) envVars = list(self.env_vars_by_manifest[m])[0] self.extraEnv = {} if envVars: self.extraEnv = envVars.strip().split() self.log.info( "The following extra environment variables will be set:\n {}".format( "\n ".join(self.extraEnv) ) ) self.parseAndCreateTestsDirs(m) # If we are using --run-by-manifest, we should not use the profile path (if) provided # by the user, since we need to create a new directory for each run. We would face # problems if we use the directory provided by the user. tests_in_manifest = [t["path"] for t in tests if t["manifest"] == m] res = self.runMochitests(options, tests_in_manifest, manifestToFilter=m) if res == TBPL_RETRY: # terminate task return res result = result or res # Dump the logging buffer self.message_logger.dump_buffered() self.log.group_end(name=m) if res == -1: break if self.manifest is not None: self.cleanup(options, True) e10s_mode = "e10s" if options.e10s else "non-e10s" # for failure mode: where browser window has crashed and we have no reported results if ( self.countpass == self.countfail == self.counttodo == 0 and options.crashAsPass ): self.countpass = 1 self.result = 0 # printing total number of tests if options.flavor == "browser": print("TEST-INFO | checking window state") print("Browser Chrome Test Summary") print("\tPassed: %s" % self.countpass) print("\tFailed: %s" % self.countfail) print("\tTodo: %s" % self.counttodo) print("\tMode: %s" % e10s_mode) print("*** End BrowserChrome Test Results ***") else: print("0 INFO TEST-START | Shutdown") print("1 INFO Passed: %s" % self.countpass) print("2 INFO Failed: %s" % self.countfail) print("3 INFO Todo: %s" % self.counttodo) print("4 INFO Mode: %s" % e10s_mode) print("5 INFO SimpleTest FINISHED") self.handleShutdownProfile(options) if not result: if self.countfail or not (self.countpass or self.counttodo): # at least one test failed, or # no tests passed, and no tests failed (possibly a crash) result = 1 return result def handleShutdownProfile(self, options): # If shutdown profiling was enabled, then the user will want to access the # performance profile. The following code will display helpful log messages # and automatically open the profile if it is requested. if self.browserEnv and "MOZ_PROFILER_SHUTDOWN" in self.browserEnv: profile_path = self.browserEnv["MOZ_PROFILER_SHUTDOWN"] profiler_logger = get_proxy_logger("profiler") profiler_logger.info("Shutdown performance profiling was enabled") profiler_logger.info("Profile saved locally to: %s" % profile_path) if options.profilerSaveOnly or options.profiler: # Only do the extra work of symbolicating and viewing the profile if # officially requested through a command line flag. The MOZ_PROFILER_* # flags can be set by a user. symbolicate_profile_json(profile_path, options.topobjdir) view_gecko_profile_from_mochitest( profile_path, options, profiler_logger ) else: profiler_logger.info( "The profiler was enabled outside of the mochitests. " "Use --profiler instead of MOZ_PROFILER_SHUTDOWN to " "symbolicate and open the profile automatically." ) # Clean up the temporary file if it exists. if self.profiler_tempdir: shutil.rmtree(self.profiler_tempdir) def doTests(self, options, testsToFilter=None, manifestToFilter=None): # A call to initializeLooping method is required in case of --run-by-dir or --bisect-chunk # since we need to initialize variables for each loop. if options.bisectChunk or options.runByManifest: self.initializeLooping(options) # get debugger info, a dict of: # {'path': path to the debugger (string), # 'interactive': whether the debugger is interactive or not (bool) # 'args': arguments to the debugger (list) # TODO: use mozrunner.local.debugger_arguments: # https://github.com/mozilla/mozbase/blob/master/mozrunner/mozrunner/local.py#L42 debuggerInfo = None if options.debugger: debuggerInfo = mozdebug.get_debugger_info( options.debugger, options.debuggerArgs, options.debuggerInteractive ) if options.useTestMediaDevices: self.initializeVirtualAudioDevices() devices = findTestMediaDevices(self.log) if not devices: self.log.error("Could not find test media devices to use") return 1 self.mediaDevices = devices # See if we were asked to run on Valgrind valgrindPath = None valgrindArgs = None valgrindSuppFiles = None if options.valgrind: valgrindPath = options.valgrind if options.valgrindArgs: valgrindArgs = options.valgrindArgs if options.valgrindSuppFiles: valgrindSuppFiles = options.valgrindSuppFiles if (valgrindArgs or valgrindSuppFiles) and not valgrindPath: self.log.error( "Specified --valgrind-args or --valgrind-supp-files," " but not --valgrind" ) return 1 if valgrindPath and debuggerInfo: self.log.error("Can't use both --debugger and --valgrind together") return 1 if valgrindPath and not valgrindSuppFiles: valgrindSuppFiles = ",".join(get_default_valgrind_suppression_files()) # buildProfile sets self.profile . # This relies on sideeffects and isn't very stateful: # https://bugzilla.mozilla.org/show_bug.cgi?id=919300 self.manifest = self.buildProfile(options) if self.manifest is None: return 1 self.leak_report_file = os.path.join(options.profilePath, "runtests_leaks.log") self.browserEnv = self.buildBrowserEnv(options, debuggerInfo is not None) if self.browserEnv is None: return 1 if self.mozLogs: self.browserEnv["MOZ_LOG_FILE"] = "{}/moz-pid=%PID-uid={}.log".format( self.browserEnv["MOZ_UPLOAD_DIR"], str(uuid.uuid4()) ) status = 0 try: self.startServers(options, debuggerInfo) if options.jsconsole: options.browserArgs.extend(["--jsconsole"]) if options.jsdebugger: options.browserArgs.extend(["-wait-for-jsdebugger", "-jsdebugger"]) # -jsdebugger takes a binary path as an optional argument. # Append jsdebuggerPath right after `-jsdebugger`. if options.jsdebuggerPath: options.browserArgs.extend([options.jsdebuggerPath]) # Remove the leak detection file so it can't "leak" to the tests run. # The file is not there if leak logging was not enabled in the # application build. if os.path.exists(self.leak_report_file): os.remove(self.leak_report_file) # then again to actually run mochitest if options.timeout: timeout = options.timeout + 30 elif options.debugger or options.jsdebugger or not options.autorun: timeout = None else: # We generally want the JS harness or marionette to handle # timeouts if they can. # The default JS harness timeout is currently 300 seconds. # The default Marionette socket timeout is currently 360 seconds. # Wait a little (10 seconds) more before timing out here. # See bug 479518 and bug 1414063. timeout = 370.0 if "MOZ_CHAOSMODE=0xfb" in options.environment and timeout: timeout *= 2 # Detect shutdown leaks for m-bc runs if # code coverage is not enabled. detectShutdownLeaks = False if options.jscov_dir_prefix is None: detectShutdownLeaks = ( mozinfo.info["debug"] and options.flavor == "browser" and options.subsuite != "thunderbird" and not options.crashAsPass ) self.start_script_kwargs["flavor"] = self.normflavor(options.flavor) marionette_args = { "symbols_path": options.symbolsPath, "socket_timeout": options.marionette_socket_timeout, "startup_timeout": options.marionette_startup_timeout, } if options.marionette: host, port = options.marionette.split(":") marionette_args["host"] = host marionette_args["port"] = int(port) # testsToFilter parameter is used to filter out the test list that # is sent to getTestsByScheme for scheme, tests in self.getTestsByScheme( options, testsToFilter, True, manifestToFilter ): # read the number of tests here, if we are not going to run any, # terminate early if not tests: continue self.currentTests = [t["path"] for t in tests] testURL = self.buildTestURL(options, scheme=scheme) self.buildURLOptions(options, self.browserEnv) if self.urlOpts: testURL += "?" + "&".join(self.urlOpts) if options.runFailures: testURL += "&runFailures=true" if options.timeoutAsPass: testURL += "&timeoutAsPass=true" if options.conditionedProfile: testURL += "&conditionedProfile=true" self.log.info("runtests.py | Running with scheme: {}".format(scheme)) self.log.info( "runtests.py | Running with e10s: {}".format(options.e10s) ) self.log.info( "runtests.py | Running with fission: {}".format( mozinfo.info.get("fission", True) ) ) self.log.info( "runtests.py | Running with cross-origin iframes: {}".format( mozinfo.info.get("xorigin", False) ) ) self.log.info( "runtests.py | Running with serviceworker_e10s: {}".format( mozinfo.info.get("serviceworker_e10s", False) ) ) self.log.info( "runtests.py | Running with socketprocess_e10s: {}".format( mozinfo.info.get("socketprocess_e10s", False) ) ) self.log.info("runtests.py | Running tests: start.\n") ret, _ = self.runApp( testURL, self.browserEnv, options.app, profile=self.profile, extraArgs=options.browserArgs + self.extraArgs, utilityPath=options.utilityPath, debuggerInfo=debuggerInfo, valgrindPath=valgrindPath, valgrindArgs=valgrindArgs, valgrindSuppFiles=valgrindSuppFiles, symbolsPath=options.symbolsPath, timeout=timeout, detectShutdownLeaks=detectShutdownLeaks, screenshotOnFail=options.screenshotOnFail, bisectChunk=options.bisectChunk, restartAfterFailure=options.restartAfterFailure, marionette_args=marionette_args, e10s=options.e10s, runFailures=options.runFailures, crashAsPass=options.crashAsPass, currentManifest=manifestToFilter, ) status = ret or status except KeyboardInterrupt: self.log.info("runtests.py | Received keyboard interrupt.\n") status = -1 except Exception as e: traceback.print_exc() self.log.error( "Automation Error: Received unexpected exception while running application\n" ) if "ADBTimeoutError" in repr(e): self.log.info("runtests.py | Device disconnected. Aborting test.\n") raise status = 1 finally: self.stopServers() ignoreMissingLeaks = options.ignoreMissingLeaks leakThresholds = options.leakThresholds if options.crashAsPass: ignoreMissingLeaks.append("tab") ignoreMissingLeaks.append("socket") # Provide a floor for Windows chrome leak detection, because we know # we have some Windows-specific shutdown hangs that we avoid by timing # out and leaking memory. if options.flavor == "chrome" and mozinfo.isWin: leakThresholds["default"] += 1296 # Stop leak detection if m-bc code coverage is enabled # by maxing out the leak threshold for all processes. if options.jscov_dir_prefix: for processType in leakThresholds: ignoreMissingLeaks.append(processType) leakThresholds[processType] = sys.maxsize utilityPath = options.utilityPath or options.xrePath if status == 0: # ignore leak checks for crashes mozleak.process_leak_log( self.leak_report_file, leak_thresholds=leakThresholds, ignore_missing_leaks=ignoreMissingLeaks, log=self.log, stack_fixer=get_stack_fixer_function(utilityPath, options.symbolsPath), scope=manifestToFilter, ) self.log.info("runtests.py | Running tests: end.") if self.manifest is not None: self.cleanup(options, False) return status def handleTimeout( self, timeout, proc, utilityPath, debuggerInfo, browser_pid, processLog ): """handle process output timeout""" # TODO: bug 913975 : _processOutput should call self.processOutputLine # one more time one timeout (I think) message = { "action": "test_end", "status": "TIMEOUT", "expected": "PASS", "thread": None, "pid": None, "source": "mochitest", "time": int(time.time()) * 1000, "test": self.lastTestSeen, "message": "application timed out after %d seconds with no output" % int(timeout), } # need to send a test_end in order to have mozharness process messages properly # this requires a custom message vs log.error/log.warning/etc. self.message_logger.process_message(message) self.message_logger.dump_buffered() self.message_logger.buffering = False self.log.warning("Force-terminating active process(es).") browser_pid = browser_pid or proc.pid child_pids = self.extract_child_pids(processLog, browser_pid) self.log.info("Found child pids: %s" % child_pids) if HAVE_PSUTIL: try: browser_proc = [psutil.Process(browser_pid)] except Exception: self.log.info("Failed to get proc for pid %d" % browser_pid) browser_proc = [] try: child_procs = [psutil.Process(pid) for pid in child_pids] except Exception: self.log.info("Failed to get child procs") child_procs = [] for pid in child_pids: self.killAndGetStack( pid, utilityPath, debuggerInfo, dump_screen=not debuggerInfo ) gone, alive = psutil.wait_procs(child_procs, timeout=30) for p in gone: self.log.info("psutil found pid %s dead" % p.pid) for p in alive: self.log.warning("failed to kill pid %d after 30s" % p.pid) self.killAndGetStack( browser_pid, utilityPath, debuggerInfo, dump_screen=not debuggerInfo ) gone, alive = psutil.wait_procs(browser_proc, timeout=30) for p in gone: self.log.info("psutil found pid %s dead" % p.pid) for p in alive: self.log.warning("failed to kill pid %d after 30s" % p.pid) else: self.log.error( "psutil not available! Will wait 30s before " "attempting to kill parent process. This should " "not occur in mozilla automation. See bug 1143547." ) for pid in child_pids: self.killAndGetStack( pid, utilityPath, debuggerInfo, dump_screen=not debuggerInfo ) if child_pids: time.sleep(30) self.killAndGetStack( browser_pid, utilityPath, debuggerInfo, dump_screen=not debuggerInfo ) def archiveMozLogs(self): if self.mozLogs: with zipfile.ZipFile( "{}/mozLogs.zip".format(os.environ["MOZ_UPLOAD_DIR"]), "w", zipfile.ZIP_DEFLATED, ) as logzip: for logfile in glob.glob( "{}/moz*.log*".format(os.environ["MOZ_UPLOAD_DIR"]) ): logzip.write(logfile, os.path.basename(logfile)) os.remove(logfile) logzip.close() class OutputHandler(object): """line output handler for mozrunner""" def __init__( self, harness, utilityPath, symbolsPath=None, dump_screen_on_timeout=True, dump_screen_on_fail=False, shutdownLeaks=None, lsanLeaks=None, bisectChunk=None, restartAfterFailure=None, ): """ harness -- harness instance dump_screen_on_timeout -- whether to dump the screen on timeout """ self.harness = harness self.utilityPath = utilityPath self.symbolsPath = symbolsPath self.dump_screen_on_timeout = dump_screen_on_timeout self.dump_screen_on_fail = dump_screen_on_fail self.shutdownLeaks = shutdownLeaks self.lsanLeaks = lsanLeaks self.bisectChunk = bisectChunk self.restartAfterFailure = restartAfterFailure self.browserProcessId = None self.stackFixerFunction = self.stackFixer() def processOutputLine(self, line): """per line handler of output for mozprocess""" # Parsing the line (by the structured messages logger). messages = self.harness.message_logger.parse_line(line) for message in messages: # Passing the message to the handlers msg = message for handler in self.outputHandlers(): msg = handler(msg) # Processing the message by the logger self.harness.message_logger.process_message(msg) __call__ = processOutputLine def outputHandlers(self): """returns ordered list of output handlers""" handlers = [ self.fix_stack, self.record_last_test, self.dumpScreenOnTimeout, self.dumpScreenOnFail, self.trackShutdownLeaks, self.trackLSANLeaks, self.countline, ] if self.bisectChunk or self.restartAfterFailure: handlers.append(self.record_result) handlers.append(self.first_error) return handlers def stackFixer(self): """ return get_stack_fixer_function, if any, to use on the output lines """ return get_stack_fixer_function(self.utilityPath, self.symbolsPath) def finish(self): if self.shutdownLeaks: numFailures, errorMessages = self.shutdownLeaks.process() self.harness.countfail += numFailures for message in errorMessages: msg = { "action": "test_end", "status": "FAIL", "expected": "PASS", "thread": None, "pid": None, "source": "mochitest", "time": int(time.time()) * 1000, "test": message["test"], "message": message["msg"], } self.harness.message_logger.process_message(msg) if self.lsanLeaks: self.harness.countfail += self.lsanLeaks.process() # output message handlers: # these take a message and return a message def record_result(self, message): # by default make the result key equal to pass. if message["action"] == "test_start": key = message["test"].split("/")[-1].strip() self.harness.result[key] = "PASS" elif message["action"] == "test_status": if "expected" in message: key = message["test"].split("/")[-1].strip() self.harness.result[key] = "FAIL" elif message["status"] == "FAIL": key = message["test"].split("/")[-1].strip() self.harness.result[key] = "TODO" return message def first_error(self, message): if ( message["action"] == "test_status" and "expected" in message and message["status"] == "FAIL" ): key = message["test"].split("/")[-1].strip() if key not in self.harness.expectedError: self.harness.expectedError[key] = message.get( "message", message["subtest"] ).strip() return message def countline(self, message): if message["action"] == "log": line = message.get("message", "") elif message["action"] == "process_output": line = message.get("data", "") else: return message val = 0 try: val = int(line.split(":")[-1].strip()) except (AttributeError, ValueError): return message if "Passed:" in line: self.harness.countpass += val elif "Failed:" in line: self.harness.countfail += val elif "Todo:" in line: self.harness.counttodo += val return message def fix_stack(self, message): if self.stackFixerFunction: if message["action"] == "log": message["message"] = self.stackFixerFunction(message["message"]) elif message["action"] == "process_output": message["data"] = self.stackFixerFunction(message["data"]) return message def record_last_test(self, message): """record last test on harness""" if message["action"] == "test_start": self.harness.lastTestSeen = message["test"] elif message["action"] == "test_end": self.harness.lastTestSeen = "{} (finished)".format(message["test"]) return message def dumpScreenOnTimeout(self, message): if ( not self.dump_screen_on_fail and self.dump_screen_on_timeout and message["action"] == "test_status" and "expected" in message and "Test timed out" in message["subtest"] ): self.harness.dumpScreen(self.utilityPath) return message def dumpScreenOnFail(self, message): if ( self.dump_screen_on_fail and "expected" in message and message["status"] == "FAIL" ): self.harness.dumpScreen(self.utilityPath) return message def trackLSANLeaks(self, message): if self.lsanLeaks and message["action"] in ("log", "process_output"): line = ( message.get("message", "") if message["action"] == "log" else message["data"] ) if "(finished)" in self.harness.lastTestSeen: self.lsanLeaks.log(line, self.harness.lastManifest) else: self.lsanLeaks.log(line, self.harness.lastTestSeen) return message def trackShutdownLeaks(self, message): if self.shutdownLeaks: self.shutdownLeaks.log(message) return message def view_gecko_profile_from_mochitest(profile_path, options, profiler_logger): """Getting shutdown performance profiles from just the command line arguments is difficult. This function makes the developer ergonomics a bit easier by taking the generated Gecko profile, and automatically serving it to profiler.firefox.com. The Gecko profile during shutdown is dumped to disk at: {objdir}/_tests/testing/mochitest/{profilename} This function takes that file, and launches a local webserver, and then points a browser to profiler.firefox.com to view it. From there it's easy to publish or save the profile. """ if options.profilerSaveOnly: # The user did not want this to automatically open, only share the location. return if not os.path.exists(profile_path): profiler_logger.error( "No profile was found at the profile path, cannot " "launch profiler.firefox.com." ) return profiler_logger.info("Loading this profile in the Firefox Profiler") view_gecko_profile(profile_path) def run_test_harness(parser, options): parser.validate(options) logger_options = { key: value for key, value in six.iteritems(vars(options)) if key.startswith("log") or key == "valgrind" } runner = MochitestDesktop( options.flavor, logger_options, options.stagedAddons, quiet=options.quiet ) options.runByManifest = False if options.flavor in ("plain", "a11y", "browser", "chrome"): options.runByManifest = True # run until failure, then loop until all tests have ran # using looping similar to bisection code if options.restartAfterFailure: options.runUntilFailure = True if options.verify or options.verify_fission: result = runner.verifyTests(options) else: result = runner.runTests(options) runner.archiveMozLogs() runner.message_logger.finish() return result def cli(args=sys.argv[1:]): # parse command line options parser = MochitestArgumentParser(app="generic") options = parser.parse_args(args) if options is None: # parsing error sys.exit(1) return run_test_harness(parser, options) if __name__ == "__main__": sys.exit(cli())