meson/mesonbuild/mtest.py

# Copyright 2016-2017 The Meson development team

# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at

#     http://www.apache.org/licenses/LICENSE-2.0

# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# A tool to run tests in many different ways.

from pathlib import Path
from collections import deque
from copy import deepcopy
from itertools import islice
import argparse
import asyncio
import datetime
import enum
import json
import multiprocessing
import os
import pickle
import platform
import random
import re
import signal
import subprocess
import shlex
import sys
import textwrap
import time
import typing as T
import unicodedata
import xml.etree.ElementTree as et

from . import build
from . import environment
from . import mlog
from .coredata import major_versions_differ, MesonVersionMismatchException
from .coredata import version as coredata_version
from .mesonlib import (MesonException, OrderedSet, RealPathAction,
                       get_wine_shortpath, join_args, split_args)
from .mintro import get_infodir, load_info_file
from .programs import ExternalProgram
from .backend.backends import TestProtocol, TestSerialisation

# GNU autotools interprets a return code of 77 from tests it executes to
# mean that the test should be skipped.
GNU_SKIP_RETURNCODE = 77

# GNU autotools interprets a return code of 99 from tests it executes to
# mean that the test failed even before testing what it is supposed to test.
GNU_ERROR_RETURNCODE = 99

# Exit if 3 Ctrl-C's are received within one second
MAX_CTRLC = 3

def is_windows() -> bool:
    platname = platform.system().lower()
    return platname == 'windows'

def is_cygwin() -> bool:
    return sys.platform == 'cygwin'

UNIWIDTH_MAPPING = {'F': 2, 'H': 1, 'W': 2, 'Na': 1, 'N': 1, 'A': 1}
def uniwidth(s: str) -> int:
    result = 0
    for c in s:
        w = unicodedata.east_asian_width(c)
        result += UNIWIDTH_MAPPING[w]
    return result

def determine_worker_count() -> int:
    varname = 'MESON_TESTTHREADS'
    if varname in os.environ:
        try:
            num_workers = int(os.environ[varname])
        except ValueError:
            print(f'Invalid value in {varname}, using 1 thread.')
            num_workers = 1
    else:
        try:
            # Fails in some weird environments such as Debian
            # reproducible build.
            num_workers = multiprocessing.cpu_count()
        except Exception:
            num_workers = 1
    return num_workers

def add_arguments(parser: argparse.ArgumentParser) -> None:
    parser.add_argument('--repeat', default=1, dest='repeat', type=int,
                        help='Number of times to run the tests.')
    parser.add_argument('--no-rebuild', default=False, action='store_true',
                        help='Do not rebuild before running tests.')
    parser.add_argument('--gdb', default=False, dest='gdb', action='store_true',
                        help='Run test under gdb.')
    parser.add_argument('--gdb-path', default='gdb', dest='gdb_path',
                        help='Path to the gdb binary (default: gdb).')
    parser.add_argument('--list', default=False, dest='list', action='store_true',
                        help='List available tests.')
    parser.add_argument('--wrapper', default=None, dest='wrapper', type=split_args,
                        help='wrapper to run tests with (e.g. Valgrind)')
    parser.add_argument('-C', dest='wd', action=RealPathAction,
                        # https://github.com/python/typeshed/issues/3107
                        # https://github.com/python/mypy/issues/7177
                        type=os.path.abspath,  # type: ignore
                        help='directory to cd into before running')
    parser.add_argument('--suite', default=[], dest='include_suites', action='append', metavar='SUITE',
                        help='Only run tests belonging to the given suite.')
    parser.add_argument('--no-suite', default=[], dest='exclude_suites', action='append', metavar='SUITE',
                        help='Do not run tests belonging to the given suite.')
    parser.add_argument('--no-stdsplit', default=True, dest='split', action='store_false',
                        help='Do not split stderr and stdout in test logs.')
    parser.add_argument('--print-errorlogs', default=False, action='store_true',
                        help="Whether to print failing tests' logs.")
    parser.add_argument('--benchmark', default=False, action='store_true',
                        help="Run benchmarks instead of tests.")
    parser.add_argument('--logbase', default='testlog',
                        help="Base name for log file.")
    parser.add_argument('--num-processes', default=determine_worker_count(), type=int,
                        help='How many parallel processes to use.')
    parser.add_argument('-v', '--verbose', default=False, action='store_true',
                        help='Do not redirect stdout and stderr')
    parser.add_argument('-q', '--quiet', default=False, action='store_true',
                        help='Produce less output to the terminal.')
    parser.add_argument('-t', '--timeout-multiplier', type=float, default=None,
                        help='Define a multiplier for test timeout, for example '
                        ' when running tests in particular conditions they might take'
                        ' more time to execute. (<= 0 to disable timeout)')
    parser.add_argument('--setup', default=None, dest='setup',
                        help='Which test setup to use.')
    parser.add_argument('--test-args', default=[], type=split_args,
                        help='Arguments to pass to the specified test(s) or all tests')
    parser.add_argument('args', nargs='*',
                        help='Optional list of test names to run. "testname" to run all tests with that name, '
                        '"subprojname:testname" to specifically run "testname" from "subprojname", '
                        '"subprojname:" to run all tests defined by "subprojname".')


def print_safe(s: str) -> None:
    end = '' if s[-1] == '\n' else '\n'
    try:
        print(s, end=end)
    except UnicodeEncodeError:
        s = s.encode('ascii', errors='backslashreplace').decode('ascii')
        print(s, end=end)

def join_lines(a: str, b: str) -> str:
    if not a:
        return b
    if not b:
        return a
    return a + '\n' + b

def dashes(s: str, dash: str, cols: int) -> str:
    if not s:
        return dash * cols
    s = ' ' + s + ' '
    width = uniwidth(s)
    first = (cols - width) // 2
    s = dash * first + s
    return s + dash * (cols - first - width)

def returncode_to_status(retcode: int) -> str:
    # Note: We can't use `os.WIFSIGNALED(result.returncode)` and the related
    # functions here because the status returned by subprocess is munged. It
    # returns a negative value if the process was killed by a signal rather than
    # the raw status returned by `wait()`. Also, If a shell sits between Meson
    # the the actual unit test that shell is likely to convert a termination due
    # to a signal into an exit status of 128 plus the signal number.
    if retcode < 0:
        signum = -retcode
        try:
            signame = signal.Signals(signum).name
        except ValueError:
            signame = 'SIGinvalid'
        return f'killed by signal {signum} {signame}'

    if retcode <= 128:
        return f'exit status {retcode}'

    signum = retcode - 128
    try:
        signame = signal.Signals(signum).name
    except ValueError:
        signame = 'SIGinvalid'
    return f'(exit status {retcode} or signal {signum} {signame})'

# TODO for Windows
sh_quote: T.Callable[[str], str] = lambda x: x
if not is_windows():
    sh_quote = shlex.quote

def env_tuple_to_str(env: T.Iterable[T.Tuple[str, str]]) -> str:
    return ''.join(["{}={} ".format(k, sh_quote(v)) for k, v in env])


class TestException(MesonException):
    pass


@enum.unique
class ConsoleUser(enum.Enum):

    # the logger can use the console
    LOGGER = 0

    # the console is used by gdb
    GDB = 1

    # the console is used to write stdout/stderr
    STDOUT = 2


@enum.unique
class TestResult(enum.Enum):

    PENDING = 'PENDING'
    RUNNING = 'RUNNING'
    OK = 'OK'
    TIMEOUT = 'TIMEOUT'
    INTERRUPT = 'INTERRUPT'
    SKIP = 'SKIP'
    FAIL = 'FAIL'
    EXPECTEDFAIL = 'EXPECTEDFAIL'
    UNEXPECTEDPASS = 'UNEXPECTEDPASS'
    ERROR = 'ERROR'

    @staticmethod
    def maxlen() -> int:
        return 14 # len(UNEXPECTEDPASS)

    def is_ok(self) -> bool:
        return self in {TestResult.OK, TestResult.EXPECTEDFAIL}

    def is_bad(self) -> bool:
        return self in {TestResult.FAIL, TestResult.TIMEOUT, TestResult.INTERRUPT,
                        TestResult.UNEXPECTEDPASS, TestResult.ERROR}

    def is_finished(self) -> bool:
        return self not in {TestResult.PENDING, TestResult.RUNNING}

    def was_killed(self) -> bool:
        return self in (TestResult.TIMEOUT, TestResult.INTERRUPT)

    def colorize(self, s: str) -> mlog.AnsiDecorator:
        if self.is_bad():
            decorator = mlog.red
        elif self in (TestResult.SKIP, TestResult.EXPECTEDFAIL):
            decorator = mlog.yellow
        elif self.is_finished():
            decorator = mlog.green
        else:
            decorator = mlog.blue
        return decorator(s)

    def get_text(self, colorize: bool) -> str:
        result_str = '{res:{reslen}}'.format(res=self.value, reslen=self.maxlen())
        return self.colorize(result_str).get_text(colorize)


TYPE_TAPResult = T.Union['TAPParser.Test', 'TAPParser.Error', 'TAPParser.Version', 'TAPParser.Plan', 'TAPParser.Bailout']

class TAPParser:
    class Plan(T.NamedTuple):
        num_tests: int
        late: bool
        skipped: bool
        explanation: T.Optional[str]

    class Bailout(T.NamedTuple):
        message: str

    class Test(T.NamedTuple):
        number: int
        name: str
        result: TestResult
        explanation: T.Optional[str]

        def __str__(self) -> str:
            return f'{self.number} {self.name}'.strip()

    class Error(T.NamedTuple):
        message: str

    class Version(T.NamedTuple):
        version: int

    _MAIN = 1
    _AFTER_TEST = 2
    _YAML = 3

    _RE_BAILOUT = re.compile(r'Bail out!\s*(.*)')
    _RE_DIRECTIVE = re.compile(r'(?:\s*\#\s*([Ss][Kk][Ii][Pp]\S*|[Tt][Oo][Dd][Oo])\b\s*(.*))?')
    _RE_PLAN = re.compile(r'1\.\.([0-9]+)' + _RE_DIRECTIVE.pattern)
    _RE_TEST = re.compile(r'((?:not )?ok)\s*(?:([0-9]+)\s*)?([^#]*)' + _RE_DIRECTIVE.pattern)
    _RE_VERSION = re.compile(r'TAP version ([0-9]+)')
    _RE_YAML_START = re.compile(r'(\s+)---.*')
    _RE_YAML_END = re.compile(r'\s+\.\.\.\s*')

    found_late_test = False
    bailed_out = False
    plan: T.Optional[Plan] = None
    lineno = 0
    num_tests = 0
    yaml_lineno: T.Optional[int] = None
    yaml_indent = ''
    state = _MAIN
    version = 12

    def parse_test(self, ok: bool, num: int, name: str, directive: T.Optional[str], explanation: T.Optional[str]) -> \
            T.Generator[T.Union['TAPParser.Test', 'TAPParser.Error'], None, None]:
        name = name.strip()
        if name[0:2] == '- ':
            name = name[2:]
        explanation = explanation.strip() if explanation else None
        if directive is not None:
            directive = directive.upper()
            if directive.startswith('SKIP'):
                if ok:
                    yield self.Test(num, name, TestResult.SKIP, explanation)
                    return
            elif directive == 'TODO':
                yield self.Test(num, name, TestResult.UNEXPECTEDPASS if ok else TestResult.EXPECTEDFAIL, explanation)
                return
            else:
                yield self.Error(f'invalid directive "{directive}"')

        yield self.Test(num, name, TestResult.OK if ok else TestResult.FAIL, explanation)

    async def parse_async(self, lines: T.AsyncIterator[str]) -> T.AsyncIterator[TYPE_TAPResult]:
        async for line in lines:
            for event in self.parse_line(line):
                yield event
        for event in self.parse_line(None):
            yield event

    def parse(self, io: T.Iterator[str]) -> T.Iterator[TYPE_TAPResult]:
        for line in io:
            yield from self.parse_line(line)
        yield from self.parse_line(None)

    def parse_line(self, line: T.Optional[str]) -> T.Iterator[TYPE_TAPResult]:
        if line is not None:
            self.lineno += 1
            line = line.rstrip()

            # YAML blocks are only accepted after a test
            if self.state == self._AFTER_TEST:
                if self.version >= 13:
                    m = self._RE_YAML_START.match(line)
                    if m:
                        self.state = self._YAML
                        self.yaml_lineno = self.lineno
                        self.yaml_indent = m.group(1)
                        return
                self.state = self._MAIN

            elif self.state == self._YAML:
                if self._RE_YAML_END.match(line):
                    self.state = self._MAIN
                    return
                if line.startswith(self.yaml_indent):
                    return
                yield self.Error(f'YAML block not terminated (started on line {self.yaml_lineno})')
                self.state = self._MAIN

            assert self.state == self._MAIN
            if line.startswith('#'):
                return

            m = self._RE_TEST.match(line)
            if m:
                if self.plan and self.plan.late and not self.found_late_test:
                    yield self.Error('unexpected test after late plan')
                    self.found_late_test = True
                self.num_tests += 1
                num = self.num_tests if m.group(2) is None else int(m.group(2))
                if num != self.num_tests:
                    yield self.Error('out of order test numbers')
                yield from self.parse_test(m.group(1) == 'ok', num,
                                           m.group(3), m.group(4), m.group(5))
                self.state = self._AFTER_TEST
                return

            m = self._RE_PLAN.match(line)
            if m:
                if self.plan:
                    yield self.Error('more than one plan found')
                else:
                    num_tests = int(m.group(1))
                    skipped = (num_tests == 0)
                    if m.group(2):
                        if m.group(2).upper().startswith('SKIP'):
                            if num_tests > 0:
                                yield self.Error('invalid SKIP directive for plan')
                            skipped = True
                        else:
                            yield self.Error('invalid directive for plan')
                    self.plan = self.Plan(num_tests=num_tests, late=(self.num_tests > 0),
                                          skipped=skipped, explanation=m.group(3))
                    yield self.plan
                return

            m = self._RE_BAILOUT.match(line)
            if m:
                yield self.Bailout(m.group(1))
                self.bailed_out = True
                return

            m = self._RE_VERSION.match(line)
            if m:
                # The TAP version is only accepted as the first line
                if self.lineno != 1:
                    yield self.Error('version number must be on the first line')
                    return
                self.version = int(m.group(1))
                if self.version < 13:
                    yield self.Error('version number should be at least 13')
                else:
                    yield self.Version(version=self.version)
                return

            if not line:
                return

            yield self.Error('unexpected input at line {}'.format((self.lineno,)))
        else:
            # end of file
            if self.state == self._YAML:
                yield self.Error(f'YAML block not terminated (started on line {self.yaml_lineno})')

            if not self.bailed_out and self.plan and self.num_tests != self.plan.num_tests:
                if self.num_tests < self.plan.num_tests:
                    yield self.Error(f'Too few tests run (expected {self.plan.num_tests}, got {self.num_tests})')
                else:
                    yield self.Error(f'Too many tests run (expected {self.plan.num_tests}, got {self.num_tests})')

class TestLogger:
    def flush(self) -> None:
        pass

    def start(self, harness: 'TestHarness') -> None:
        pass

    def start_test(self, harness: 'TestHarness', test: 'TestRun') -> None:
        pass

    def log_subtest(self, harness: 'TestHarness', test: 'TestRun', s: str, res: TestResult) -> str:
        return ''

    def log(self, harness: 'TestHarness', result: 'TestRun') -> None:
        pass

    async def finish(self, harness: 'TestHarness') -> None:
        pass

    def close(self) -> None:
        pass


class TestFileLogger(TestLogger):
    def __init__(self, filename: str, errors: str = 'replace') -> None:
        self.filename = filename
        self.file = open(filename, 'w', encoding='utf-8', errors=errors)

    def close(self) -> None:
        if self.file:
            self.file.close()
            self.file = None


class ConsoleLogger(TestLogger):
    def __init__(self) -> None:
        self.update = asyncio.Event()
        self.running_tests = OrderedSet()  # type: OrderedSet['TestRun']
        self.progress_task = None          # type: T.Optional[asyncio.Future]
        self.max_left_width = 0            # type: int
        self.stop = False
        self.should_erase_line = ''
        self.test_count = 0
        self.started_tests = 0
        try:
            self.cols, _ = os.get_terminal_size(1)
            self.is_tty = True
        except OSError:
            self.cols = 80
            self.is_tty = False

    def flush(self) -> None:
        if self.should_erase_line:
            print(self.should_erase_line, end='')
            self.should_erase_line = ''

    def print_progress(self, lines: T.List[str]) -> None:
        line_count = len(lines)
        if line_count > 0:
            self.flush()
            for line in lines:
                print(line)
            print(f'\x1b[{line_count}A', end='')
            self.should_erase_line = '\x1b[K' + '\x1b[1B\x1b[K' * (line_count - 1)
            if line_count > 1:
                self.should_erase_line += f'\x1b[{line_count - 1}A'

    def request_update(self) -> None:
        self.update.set()

    def emit_progress(self, harness: 'TestHarness') -> None:
        lines: T.List[str] = []
        for test in islice(reversed(self.running_tests), 10):
            left = ' ' * (len(str(self.test_count)) * 2 + 2)
            right = '{spaces} {dur:{durlen}}'.format(
                spaces=' ' * TestResult.maxlen(),
                dur=int(time.time() - test.starttime),
                durlen=harness.duration_max_len)
            if test.timeout:
                right += '/{timeout:{durlen}}'.format(
                    timeout=test.timeout,
                    durlen=harness.duration_max_len)
            right += 's'
            lines = [harness.format(test, colorize=True,
                                     max_left_width=self.max_left_width,
                                     left=left,
                                     right=right)] + lines
        if len(self.running_tests) > 10:
            lines += [' ' * len(harness.get_test_num_prefix(0))
                      + f'[{len(self.running_tests) - 10} more tests running]']
        self.print_progress(lines)

    def start(self, harness: 'TestHarness') -> None:
        async def report_progress() -> None:
            loop = asyncio.get_event_loop()
            next_update = 0.0
            self.request_update()
            while not self.stop:
                await self.update.wait()
                self.update.clear()
                # We may get here simply because the progress line has been
                # overwritten, so do not always switch.  Only do so every
                # second, or if the printed test has finished
                if loop.time() >= next_update:
                    next_update = loop.time() + 1
                    loop.call_at(next_update, self.request_update)
                self.emit_progress(harness)
            self.flush()

        self.test_count = harness.test_count
        self.cols = max(self.cols, harness.max_left_width + 30)

        if self.is_tty and not harness.need_console:
            # Account for "[aa-bb/cc] OO " in the progress report
            self.max_left_width = 3 * len(str(self.test_count)) + 8
            self.progress_task = asyncio.ensure_future(report_progress())

    def start_test(self, harness: 'TestHarness', test: 'TestRun') -> None:
        if harness.options.verbose and test.cmdline:
            self.flush()
            print(harness.format(test, mlog.colorize_console(),
                                 max_left_width=self.max_left_width,
                                 right=test.res.get_text(mlog.colorize_console())))
        self.started_tests += 1
        self.running_tests.add(test)
        self.running_tests.move_to_end(test, last=False)
        self.request_update()

    @staticmethod
    def print_test_details_header(prefix: str, header: str) -> None:
        header += ':'
        print(prefix + mlog.italic(f'{header:<9}').get_text(mlog.colorize_console()))

    @staticmethod
    def print_test_details_line(prefix: str,
                                line: str,
                                end: str = '\n',
                                flush: bool = False) -> None:
        print(prefix + '  ' + line, flush=flush, end=end)

    @staticmethod
    def print_test_details(prefix: str,
                           header: str,
                           lines: T.Union[T.List[str], str],
                           clip: T.Optional[bool] = False) -> None:
        offset = 0
        if not isinstance(lines, list):
            lines = [lines]
        if clip and len(lines) > 100:
            offset = -100
            header += ' (only the last 100 lines from a long output included)'
        ConsoleLogger.print_test_details_header(prefix, header)
        for line in lines[offset:]:
            ConsoleLogger.print_test_details_line(prefix, line)

    def print_log(self,
                  harness: 'TestHarness',
                  result: 'TestRun',
                  no_output: bool = False) -> None:
        assert result.cmdline
        prefix = harness.get_test_num_prefix(result.num)
        self.print_test_details(prefix, "command", result.cmdline)
        self.print_test_details(prefix,
                                "exit details",
                                returncode_to_status(result.returncode))
        if not no_output:
            if result.stdo:
                if harness.options.split or result.stde:
                    name = 'stdout'
                else:
                    name = 'output'
                self.print_test_details(prefix,
                                        name,
                                        result.stdo.splitlines(),
                                        not harness.options.verbose)
            if result.stde:
                self.print_test_details(prefix,
                                        "stderr",
                                        result.stde.splitlines(),
                                        not harness.options.verbose)
            if result.additional_out:
                self.print_test_details(prefix,
                                        "additional output",
                                        result.additional_out.splitlines(),
                                        not harness.options.verbose)
            if result.additional_err:
                self.print_test_details(prefix,
                                        "additional error",
                                        result.additional_err.splitlines(),
                                        not harness.options.verbose)

    def log_subtest(self, harness: 'TestHarness', test: 'TestRun', s: str, result: TestResult) -> str:
        return 'subtest %s %s' % (s, result.get_text(mlog.colorize_console()))

    def log(self, harness: 'TestHarness', result: 'TestRun') -> None:
        self.running_tests.remove(result)
        if result.res is TestResult.TIMEOUT and (harness.options.verbose or
                                                 harness.options.print_errorlogs):
            result.additional_err += f'timed out (after {result.timeout} seconds)\n'

        if not harness.options.quiet or not result.res.is_ok():
            self.flush()
            print(harness.format(result, mlog.colorize_console(), max_left_width=self.max_left_width))
            if harness.options.verbose and not result.is_parallel and result.cmdline and not result.needs_parsing:
                # output already printed during execution
                self.print_log(harness, result, no_output=True)
            elif harness.options.verbose or (result.res.is_bad() and harness.options.print_errorlogs):
                # verbose or fail + print_errorlogs -> print
                self.print_log(harness, result)

        self.request_update()

    async def finish(self, harness: 'TestHarness') -> None:
        self.stop = True
        self.request_update()
        if self.progress_task:
            await self.progress_task

        if harness.collected_failures and \
                (harness.options.print_errorlogs or harness.options.verbose):
            print("\nSummary of Failures:\n")
            for i, result in enumerate(harness.collected_failures, 1):
                print(harness.format(result, mlog.colorize_console()))

        print(harness.summary())


class TextLogfileBuilder(TestFileLogger):
    def start(self, harness: 'TestHarness') -> None:
        self.file.write(f'Log of Meson test suite run on {datetime.datetime.now().isoformat()}\n\n')
        inherit_env = env_tuple_to_str(os.environ.items())
        self.file.write(f'Inherited environment: {inherit_env}\n\n')

    def log(self, harness: 'TestHarness', result: 'TestRun') -> None:
        self.file.write(harness.format(result, False) + '\n')
        cmdline = result.cmdline
        if cmdline:
            starttime_str = time.strftime("%H:%M:%S", time.gmtime(result.starttime))
            self.file.write(starttime_str + ' ' + cmdline + '\n')
            if result.stdo:
                self.file.write(dashes('stdout', '-', 78) + '\n')
                self.file.write(result.stdo + '\n')
                self.file.write(dashes('', '-', 78) + '\n\n')
            if result.stde:
                self.file.write(dashes('stderr', '-', 78) + '\n')
                self.file.write(result.stde + '\n')
                self.file.write(dashes('', '-', 78) + '\n\n')

    async def finish(self, harness: 'TestHarness') -> None:
        if harness.collected_failures:
            self.file.write("\nSummary of Failures:\n\n")
            for i, result in enumerate(harness.collected_failures, 1):
                self.file.write(harness.format(result, False) + '\n')
        self.file.write(harness.summary())

        print(f'Full log written to {self.filename}')


class JsonLogfileBuilder(TestFileLogger):
    def log(self, harness: 'TestHarness', result: 'TestRun') -> None:
        jresult = {'name': result.name,
                   'stdout': result.stdo,
                   'result': result.res.value,
                   'starttime': result.starttime,
                   'duration': result.duration,
                   'returncode': result.returncode,
                   'env': result.env,
                   'command': result.cmd}  # type: T.Dict[str, T.Any]
        if result.stde:
            jresult['stderr'] = result.stde
        self.file.write(json.dumps(jresult) + '\n')


class JunitBuilder(TestLogger):

    """Builder for Junit test results.

    Junit is impossible to stream out, it requires attributes counting the
    total number of tests, failures, skips, and errors in the root element
    and in each test suite. As such, we use a builder class to track each
    test case, and calculate all metadata before writing it out.

    For tests with multiple results (like from a TAP test), we record the
    test as a suite with the project_name.test_name. This allows us to track
    each result separately. For tests with only one result (such as exit-code
    tests) we record each one into a suite with the name project_name. The use
    of the project_name allows us to sort subproject tests separately from
    the root project.
    """

    def __init__(self, filename: str) -> None:
        self.filename = filename
        self.root = et.Element(
            'testsuites', tests='0', errors='0', failures='0')
        self.suites = {}  # type: T.Dict[str, et.Element]

    def log(self, harness: 'TestHarness', test: 'TestRun') -> None:
        """Log a single test case."""
        if test.junit is not None:
            for suite in test.junit.findall('.//testsuite'):
                # Assume that we don't need to merge anything here...
                suite.attrib['name'] = '{}.{}.{}'.format(test.project, test.name, suite.attrib['name'])

                # GTest can inject invalid attributes
                for case in suite.findall('.//testcase[@result]'):
                    del case.attrib['result']
                for case in suite.findall('.//testcase[@timestamp]'):
                    del case.attrib['timestamp']
                self.root.append(suite)
            return

        # In this case we have a test binary with multiple results.
        # We want to record this so that each result is recorded
        # separately
        if test.results:
            suitename = f'{test.project}.{test.name}'
            assert suitename not in self.suites or harness.options.repeat > 1, 'duplicate suite'

            suite = self.suites[suitename] = et.Element(
                'testsuite',
                name=suitename,
                tests=str(len(test.results)),
                errors=str(sum(1 for r in test.results if r.result in
                               {TestResult.INTERRUPT, TestResult.ERROR})),
                failures=str(sum(1 for r in test.results if r.result in
                                 {TestResult.FAIL, TestResult.UNEXPECTEDPASS, TestResult.TIMEOUT})),
                skipped=str(sum(1 for r in test.results if r.result is TestResult.SKIP)),
                time=str(test.duration),
            )

            for subtest in test.results:
                # Both name and classname are required. Use the suite name as
                # the class name, so that e.g. GitLab groups testcases correctly.
                testcase = et.SubElement(suite, 'testcase', name=str(subtest), classname=suitename)
                if subtest.result is TestResult.SKIP:
                    et.SubElement(testcase, 'skipped')
                elif subtest.result is TestResult.ERROR:
                    et.SubElement(testcase, 'error')
                elif subtest.result is TestResult.FAIL:
                    et.SubElement(testcase, 'failure')
                elif subtest.result is TestResult.UNEXPECTEDPASS:
                    fail = et.SubElement(testcase, 'failure')
                    fail.text = 'Test unexpected passed.'
                elif subtest.result is TestResult.INTERRUPT:
                    fail = et.SubElement(testcase, 'error')
                    fail.text = 'Test was interrupted by user.'
                elif subtest.result is TestResult.TIMEOUT:
                    fail = et.SubElement(testcase, 'error')
                    fail.text = 'Test did not finish before configured timeout.'
                if subtest.explanation:
                    et.SubElement(testcase, 'system-out').text = subtest.explanation
            if test.stdo:
                out = et.SubElement(suite, 'system-out')
                out.text = test.stdo.rstrip()
            if test.stde:
                err = et.SubElement(suite, 'system-err')
                err.text = test.stde.rstrip()
        else:
            if test.project not in self.suites:
                suite = self.suites[test.project] = et.Element(
                    'testsuite', name=test.project, tests='1', errors='0',
                    failures='0', skipped='0', time=str(test.duration))
            else:
                suite = self.suites[test.project]
                suite.attrib['tests'] = str(int(suite.attrib['tests']) + 1)

            testcase = et.SubElement(suite, 'testcase', name=test.name,
                                     classname=test.project, time=str(test.duration))
            if test.res is TestResult.SKIP:
                et.SubElement(testcase, 'skipped')
                suite.attrib['skipped'] = str(int(suite.attrib['skipped']) + 1)
            elif test.res is TestResult.ERROR:
                et.SubElement(testcase, 'error')
                suite.attrib['errors'] = str(int(suite.attrib['errors']) + 1)
            elif test.res is TestResult.FAIL:
                et.SubElement(testcase, 'failure')
                suite.attrib['failures'] = str(int(suite.attrib['failures']) + 1)
            if test.stdo:
                out = et.SubElement(testcase, 'system-out')
                out.text = test.stdo.rstrip()
            if test.stde:
                err = et.SubElement(testcase, 'system-err')
                err.text = test.stde.rstrip()

    async def finish(self, harness: 'TestHarness') -> None:
        """Calculate total test counts and write out the xml result."""
        for suite in self.suites.values():
            self.root.append(suite)
            # Skipped is really not allowed in the "testsuits" element
            for attr in ['tests', 'errors', 'failures']:
                self.root.attrib[attr] = str(int(self.root.attrib[attr]) + int(suite.attrib[attr]))

        tree = et.ElementTree(self.root)
        with open(self.filename, 'wb') as f:
            tree.write(f, encoding='utf-8', xml_declaration=True)


class TestRun:
    TEST_NUM = 0
    PROTOCOL_TO_CLASS: T.Dict[TestProtocol, T.Type['TestRun']] = {}

    def __new__(cls, test: TestSerialisation, *args: T.Any, **kwargs: T.Any) -> T.Any:
        return super().__new__(TestRun.PROTOCOL_TO_CLASS[test.protocol])

    def __init__(self, test: TestSerialisation, test_env: T.Dict[str, str],
                 name: str, timeout: T.Optional[int], is_parallel: bool):
        self.res = TestResult.PENDING
        self.test = test
        self._num = None       # type: T.Optional[int]
        self.name = name
        self.timeout = timeout
        self.results = list()  # type: T.List[TAPParser.Test]
        self.returncode = 0
        self.starttime = None  # type: T.Optional[float]
        self.duration = None   # type: T.Optional[float]
        self.stdo = None       # type: T.Optional[str]
        self.stde = None       # type: T.Optional[str]
        self.cmd = None        # type: T.Optional[T.List[str]]
        self.env = test_env    # type: T.Dict[str, str]
        self.should_fail = test.should_fail
        self.project = test.project_name
        self.junit = None      # type: T.Optional[et.ElementTree]
        self.is_parallel = is_parallel

    def start(self, cmd: T.List[str]) -> None:
        self.res = TestResult.RUNNING
        self.starttime = time.time()
        self.cmd = cmd

    @property
    def num(self) -> int:
        if self._num is None:
            TestRun.TEST_NUM += 1
            self._num = TestRun.TEST_NUM
        return self._num

    def detail(self) -> str:
        if self.res is TestResult.PENDING:
            return ''
        if self.returncode:
            return returncode_to_status(self.returncode)
        if self.results:
            # running or succeeded
            passed = sum(x.result.is_ok() for x in self.results)
            ran = sum(x.result is not TestResult.SKIP for x in self.results)
            if passed == ran:
                return f'{passed} subtests passed'
            else:
                return f'{passed}/{ran} subtests passed'
        return ''

    def _complete(self, returncode: int, res: TestResult,
                  stdo: T.Optional[str], stde: T.Optional[str],
                  additional_out: T.Optional[str], additional_err: T.Optional[str]) -> None:
        assert isinstance(res, TestResult)
        if self.should_fail and res in (TestResult.OK, TestResult.FAIL):
            res = TestResult.UNEXPECTEDPASS if res.is_ok() else TestResult.EXPECTEDFAIL

        self.res = res
        self.returncode = returncode
        self.duration = time.time() - self.starttime
        self.stdo = stdo
        self.stde = stde
        self.additional_out = additional_out
        self.additional_err = additional_err

    @property
    def cmdline(self) -> T.Optional[str]:
        if not self.cmd:
            return None
        test_only_env = set(self.env.items()) - set(os.environ.items())
        return env_tuple_to_str(test_only_env) + \
            ' '.join(sh_quote(x) for x in self.cmd)

    def complete_skip(self, message: str) -> None:
        self.starttime = time.time()
        self._complete(GNU_SKIP_RETURNCODE, TestResult.SKIP, message, None, None, None)

    def complete(self, returncode: int, res: TestResult,
                 stdo: T.Optional[str], stde: T.Optional[str],
                 additional_out: T.Optional[str], additional_err: T.Optional[str]) -> None:
        self._complete(returncode, res, stdo, stde, additional_out, additional_err)

    @property
    def needs_parsing(self) -> bool:
        return False

    async def parse(self, harness: 'TestHarness', lines: T.AsyncIterator[str]) -> T.Tuple[TestResult, str, str]:
        async for l in lines:
            pass
        return TestResult.OK, '', ''


class TestRunExitCode(TestRun):

    def complete(self, returncode: int, res: TestResult,
                 stdo: T.Optional[str], stde: T.Optional[str],
                 additional_out: T.Optional[str], additional_err: T.Optional[str]) -> None:
        if res:
            pass
        elif returncode == GNU_SKIP_RETURNCODE:
            res = TestResult.SKIP
        elif returncode == GNU_ERROR_RETURNCODE:
            res = TestResult.ERROR
        else:
            res = TestResult.FAIL if bool(returncode) else TestResult.OK
        super().complete(returncode, res, stdo, stde, additional_out, additional_err)

TestRun.PROTOCOL_TO_CLASS[TestProtocol.EXITCODE] = TestRunExitCode


class TestRunGTest(TestRunExitCode):
    def complete(self, returncode: int, res: TestResult,
                 stdo: T.Optional[str], stde: T.Optional[str],
                 additional_out: T.Optional[str], additional_err: T.Optional[str]) -> None:
        filename = f'{self.test.name}.xml'
        if self.test.workdir:
            filename = os.path.join(self.test.workdir, filename)

        try:
            self.junit = et.parse(filename)
        except FileNotFoundError:
            # This can happen if the test fails to run or complete for some
            # reason, like the rpath for libgtest isn't properly set. ExitCode
            # will handle the failure, don't generate a stacktrace.
            pass

        super().complete(returncode, res, stdo, stde, additional_out, additional_err)

TestRun.PROTOCOL_TO_CLASS[TestProtocol.GTEST] = TestRunGTest


class TestRunTAP(TestRun):
    @property
    def needs_parsing(self) -> bool:
        return True

    def complete(self, returncode: int, res: TestResult,
                 stdo: T.Optional[str], stde: T.Optional[str],
                 additional_out: T.Optional[str], additional_err: T.Optional[str]) -> None:
        if returncode != 0 and not res.was_killed():
            res = TestResult.ERROR
            stde = stde or ''
            stde += f'\n(test program exited with status code {returncode})'

        super().complete(returncode, res, stdo, stde, additional_out, additional_err)

    async def parse(self,
                    harness: 'TestHarness',
                    lines: T.AsyncIterator[str]) -> T.Tuple[TestResult, str, str]:
        res = TestResult.OK
        output = ''
        error = ''

        async for i in TAPParser().parse_async(lines):
            if isinstance(i, TAPParser.Bailout):
                res = TestResult.ERROR
                output += '\n' + harness.log_subtest(self, i.message, res)
            elif isinstance(i, TAPParser.Test):
                self.results.append(i)
                if i.result.is_bad():
                    res = TestResult.FAIL
                output += '\n' + harness.log_subtest(self, i.name or f'subtest {i.number}', i.result)
            elif isinstance(i, TAPParser.Error):
                error += '\nTAP parsing error: ' + i.message
                res = TestResult.ERROR

        if all(t.result is TestResult.SKIP for t in self.results):
            # This includes the case where self.results is empty
            res = TestResult.SKIP
        return res, output, error

TestRun.PROTOCOL_TO_CLASS[TestProtocol.TAP] = TestRunTAP


class TestRunRust(TestRun):
    @property
    def needs_parsing(self) -> bool:
        return True

    async def parse(self,
                    harness: 'TestHarness',
                    lines: T.AsyncIterator[str]) -> T.Tuple[TestResult, str, str]:
        def parse_res(n: int, name: str, result: str) -> TAPParser.Test:
            if result == 'ok':
                return TAPParser.Test(n, name, TestResult.OK, None)
            elif result == 'ignored':
                return TAPParser.Test(n, name, TestResult.SKIP, None)
            elif result == 'FAILED':
                return TAPParser.Test(n, name, TestResult.FAIL, None)
            return TAPParser.Test(n, name, TestResult.ERROR,
                                  f'Unsupported output from rust test: {result}')

        output = ''
        n = 1
        async for line in lines:
            if line.startswith('test ') and not line.startswith('test result'):
                _, name, _, result = line.rstrip().split(' ')
                name = name.replace('::', '.')
                t = parse_res(n, name, result)
                self.results.append(t)
                output += '\n' + harness.log_subtest(self, name, t.result)
                n += 1

        if all(t.result is TestResult.SKIP for t in self.results):
            # This includes the case where self.results is empty
            return TestResult.SKIP, output, ''
        elif any(t.result is TestResult.ERROR for t in self.results):
            return TestResult.ERROR, output, ''
        elif any(t.result is TestResult.FAIL for t in self.results):
            return TestResult.FAIL, output, ''
        return TestResult.OK, output, ''

TestRun.PROTOCOL_TO_CLASS[TestProtocol.RUST] = TestRunRust


def decode(stream: T.Union[None, bytes]) -> str:
    if stream is None:
        return ''
    try:
        return stream.decode('utf-8')
    except UnicodeDecodeError:
        return stream.decode('iso-8859-1', errors='ignore')

async def read_decode(reader: asyncio.StreamReader,
                      line_handler: T.Callable[[str], None]) -> str:
    stdo_lines = []
    try:
        while not reader.at_eof():
            line = decode(await reader.readline())
            if len(line) == 0:
                continue
            stdo_lines.append(line)
            if line_handler:
                line_handler(line)
        return ''.join(stdo_lines)
    except asyncio.CancelledError:
        return ''.join(stdo_lines)

# Extract lines out of the StreamReader.  Print them
# along the way if requested, and at the end collect
# them all into a future.
async def read_decode_lines(reader: asyncio.StreamReader, q: 'asyncio.Queue[T.Optional[str]]',
                            console_mode: ConsoleUser) -> str:
    stdo_lines = []
    try:
        while not reader.at_eof():
            line = decode(await reader.readline())
            stdo_lines.append(line)
            if console_mode is ConsoleUser.STDOUT:
                print(line, end='', flush=True)
            await q.put(line)
        return ''.join(stdo_lines)
    except asyncio.CancelledError:
        return ''.join(stdo_lines)
    finally:
        await q.put(None)

def run_with_mono(fname: str) -> bool:
    return fname.endswith('.exe') and not (is_windows() or is_cygwin())

def check_testdata(objs: T.List[TestSerialisation]) -> T.List[TestSerialisation]:
    if not isinstance(objs, list):
        raise MesonVersionMismatchException('<unknown>', coredata_version)
    for obj in objs:
        if not isinstance(obj, TestSerialisation):
            raise MesonVersionMismatchException('<unknown>', coredata_version)
        if not hasattr(obj, 'version'):
            raise MesonVersionMismatchException('<unknown>', coredata_version)
        if major_versions_differ(obj.version, coredata_version):
            raise MesonVersionMismatchException(obj.version, coredata_version)
    return objs

# Custom waiting primitives for asyncio

async def try_wait_one(*awaitables: T.Any, timeout: T.Optional[T.Union[int, float]]) -> None:
    """Wait for completion of one of the given futures, ignoring timeouts."""
    await asyncio.wait(awaitables,
                       timeout=timeout, return_when=asyncio.FIRST_COMPLETED)

async def queue_iter(q: 'asyncio.Queue[T.Optional[str]]') -> T.AsyncIterator[str]:
    while True:
        item = await q.get()
        q.task_done()
        if item is None:
            break
        yield item

async def complete(future: asyncio.Future) -> None:
    """Wait for completion of the given future, ignoring cancellation."""
    try:
        await future
    except asyncio.CancelledError:
        pass

async def complete_all(futures: T.Iterable[asyncio.Future],
                       timeout: T.Optional[T.Union[int, float]] = None) -> None:
    """Wait for completion of all the given futures, ignoring cancellation.
       If timeout is not None, raise an asyncio.TimeoutError after the given
       time has passed.  asyncio.TimeoutError is only raised if some futures
       have not completed and none have raised exceptions, even if timeout
       is zero."""

    def check_futures(futures: T.Iterable[asyncio.Future]) -> None:
        # Raise exceptions if needed
        left = False
        for f in futures:
            if not f.done():
                left = True
            elif not f.cancelled():
                f.result()
        if left:
            raise asyncio.TimeoutError

    # Python is silly and does not have a variant of asyncio.wait with an
    # absolute time as deadline.
    deadline = None if timeout is None else asyncio.get_event_loop().time() + timeout
    while futures and (timeout is None or timeout > 0):
        done, futures = await asyncio.wait(futures, timeout=timeout,
                                           return_when=asyncio.FIRST_EXCEPTION)
        check_futures(done)
        if deadline:
            timeout = deadline - asyncio.get_event_loop().time()

    check_futures(futures)


class TestSubprocess:
    def __init__(self, p: asyncio.subprocess.Process,
                 stdout: T.Optional[int], stderr: T.Optional[int],
                 postwait_fn: T.Callable[[], None] = None):
        self._process = p
        self.stdout = stdout
        self.stderr = stderr
        self.stdo_task = None            # type: T.Optional[asyncio.Future[str]]
        self.stde_task = None            # type: T.Optional[asyncio.Future[str]]
        self.postwait_fn = postwait_fn   # type: T.Callable[[], None]
        self.all_futures = []            # type: T.List[asyncio.Future]

    def stdout_lines(self, console_mode: ConsoleUser) -> T.AsyncIterator[str]:
        q = asyncio.Queue()              # type: asyncio.Queue[T.Optional[str]]
        decode_coro = read_decode_lines(self._process.stdout, q, console_mode)
        self.stdo_task = asyncio.ensure_future(decode_coro)
        return queue_iter(q)

    def communicate(self,
                    console_mode: ConsoleUser,
                    line_handler: T.Callable[[str], None] = None) -> T.Tuple[T.Optional[T.Awaitable[str]], T.Optional[T.Awaitable[str]]]:
        # asyncio.ensure_future ensures that printing can
        # run in the background, even before it is awaited
        if self.stdo_task is None and self.stdout is not None:
            decode_coro = read_decode(self._process.stdout, line_handler)
            self.stdo_task = asyncio.ensure_future(decode_coro)
            self.all_futures.append(self.stdo_task)
        if self.stderr is not None and self.stderr != asyncio.subprocess.STDOUT:
            decode_coro = read_decode(self._process.stderr, line_handler)
            self.stde_task = asyncio.ensure_future(decode_coro)
            self.all_futures.append(self.stde_task)

        return self.stdo_task, self.stde_task

    async def _kill(self) -> T.Optional[str]:
        # Python does not provide multiplatform support for
        # killing a process and all its children so we need
        # to roll our own.
        p = self._process
        try:
            if is_windows():
                subprocess.run(['taskkill', '/F', '/T', '/PID', str(p.pid)])
            else:
                # Send a termination signal to the process group that setsid()
                # created - giving it a chance to perform any cleanup.
                os.killpg(p.pid, signal.SIGTERM)

                # Make sure the termination signal actually kills the process
                # group, otherwise retry with a SIGKILL.
                await try_wait_one(p.wait(), timeout=0.5)
                if p.returncode is not None:
                    return None

                os.killpg(p.pid, signal.SIGKILL)

            await try_wait_one(p.wait(), timeout=1)
            if p.returncode is not None:
                return None

            # An earlier kill attempt has not worked for whatever reason.
            # Try to kill it one last time with a direct call.
            # If the process has spawned children, they will remain around.
            p.kill()
            await try_wait_one(p.wait(), timeout=1)
            if p.returncode is not None:
                return None
            return 'Test process could not be killed.'
        except ProcessLookupError:
            # Sometimes (e.g. with Wine) this happens.  There's nothing
            # we can do, probably the process already died so just wait
            # for the event loop to pick that up.
            await p.wait()
            return None
        finally:
            if self.stdo_task:
                self.stdo_task.cancel()
            if self.stde_task:
                self.stde_task.cancel()

    async def wait(self, timeout: T.Optional[int]) -> T.Tuple[int, TestResult, T.Optional[str]]:
        p = self._process
        result = None
        additional_error = None

        self.all_futures.append(asyncio.ensure_future(p.wait()))
        try:
            await complete_all(self.all_futures, timeout=timeout)
        except asyncio.TimeoutError:
            additional_error = await self._kill()
            result = TestResult.TIMEOUT
        except asyncio.CancelledError:
            # The main loop must have seen Ctrl-C.
            additional_error = await self._kill()
            result = TestResult.INTERRUPT
        finally:
            if self.postwait_fn:
                self.postwait_fn()

        return p.returncode or 0, \
            result, \
            additional_error + '\n' if additional_error else ''

class SingleTestRunner:

    def __init__(self, test: TestSerialisation, env: T.Dict[str, str], name: str,
                 options: argparse.Namespace):
        self.test = test
        self.options = options
        self.cmd = self._get_cmd()

        if self.cmd and self.test.extra_paths:
            env['PATH'] = os.pathsep.join(self.test.extra_paths + ['']) + env['PATH']
            winecmd = []
            for c in self.cmd:
                winecmd.append(c)
                if os.path.basename(c).startswith('wine'):
                    env['WINEPATH'] = get_wine_shortpath(
                        winecmd,
                        ['Z:' + p for p in self.test.extra_paths] + env.get('WINEPATH', '').split(';')
                    )
                    break

        # If MALLOC_PERTURB_ is not set, or if it is set to an empty value,
        # (i.e., the test or the environment don't explicitly set it), set
        # it ourselves. We do this unconditionally for regular tests
        # because it is extremely useful to have.
        # Setting MALLOC_PERTURB_="0" will completely disable this feature.
        if ('MALLOC_PERTURB_' not in env or not env['MALLOC_PERTURB_']) and not options.benchmark:
            env['MALLOC_PERTURB_'] = str(random.randint(1, 255))

        if self.options.gdb or self.test.timeout is None or self.test.timeout <= 0:
            timeout = None
        elif self.options.timeout_multiplier is None:
            timeout = self.test.timeout
        elif self.options.timeout_multiplier <= 0:
            timeout = None
        else:
            timeout = self.test.timeout * self.options.timeout_multiplier

        is_parallel = test.is_parallel and self.options.num_processes > 1 and not self.options.gdb
        self.runobj = TestRun(test, env, name, timeout, is_parallel)

        if self.options.gdb:
            self.console_mode = ConsoleUser.GDB
        elif self.options.verbose and not is_parallel and not self.runobj.needs_parsing:
            self.console_mode = ConsoleUser.STDOUT
        else:
            self.console_mode = ConsoleUser.LOGGER

    def _get_test_cmd(self) -> T.Optional[T.List[str]]:
        if self.test.fname[0].endswith('.jar'):
            return ['java', '-jar'] + self.test.fname
        elif not self.test.is_cross_built and run_with_mono(self.test.fname[0]):
            return ['mono'] + self.test.fname
        elif self.test.cmd_is_built and self.test.is_cross_built and self.test.needs_exe_wrapper:
            if self.test.exe_runner is None:
                # Can not run test on cross compiled executable
                # because there is no execute wrapper.
                return None
            elif self.test.cmd_is_built:
                # If the command is not built (ie, its a python script),
                # then we don't check for the exe-wrapper
                if not self.test.exe_runner.found():
                    msg = ('The exe_wrapper defined in the cross file {!r} was not '
                           'found. Please check the command and/or add it to PATH.')
                    raise TestException(msg.format(self.test.exe_runner.name))
                return self.test.exe_runner.get_command() + self.test.fname
        return self.test.fname

    def _get_cmd(self) -> T.Optional[T.List[str]]:
        test_cmd = self._get_test_cmd()
        if not test_cmd:
            return None
        return TestHarness.get_wrapper(self.options) + test_cmd

    @property
    def is_parallel(self) -> bool:
        return self.runobj.is_parallel

    @property
    def visible_name(self) -> str:
        return self.runobj.name

    @property
    def timeout(self) -> T.Optional[int]:
        return self.runobj.timeout

    async def run(self, harness: 'TestHarness') -> TestRun:
        if self.cmd is None:
            skip_stdout = 'Not run because can not execute cross compiled binaries.'
            harness.log_start_test(self.runobj)
            self.runobj.complete_skip(skip_stdout)
        else:
            cmd = self.cmd + self.test.cmd_args + self.options.test_args
            self.runobj.start(cmd)
            harness.log_start_test(self.runobj)
            await self._run_cmd(harness, cmd)
        return self.runobj

    async def _run_subprocess(self, args: T.List[str], *,
                              stdout: int, stderr: int,
                              env: T.Dict[str, str], cwd: T.Optional[str]) -> TestSubprocess:
        # Let gdb handle ^C instead of us
        if self.options.gdb:
            previous_sigint_handler = signal.getsignal(signal.SIGINT)
            # Make the meson executable ignore SIGINT while gdb is running.
            signal.signal(signal.SIGINT, signal.SIG_IGN)

        def preexec_fn() -> None:
            if self.options.gdb:
                # Restore the SIGINT handler for the child process to
                # ensure it can handle it.
                signal.signal(signal.SIGINT, signal.SIG_DFL)
            else:
                # We don't want setsid() in gdb because gdb needs the
                # terminal in order to handle ^C and not show tcsetpgrp()
                # errors avoid not being able to use the terminal.
                os.setsid()

        def postwait_fn() -> None:
            if self.options.gdb:
                # Let us accept ^C again
                signal.signal(signal.SIGINT, previous_sigint_handler)

        p = await asyncio.create_subprocess_exec(*args,
                                                 stdout=stdout,
                                                 stderr=stderr,
                                                 env=env,
                                                 cwd=cwd,
                                                 preexec_fn=preexec_fn if not is_windows() else None)
        return TestSubprocess(p, stdout=stdout, stderr=stderr,
                              postwait_fn=postwait_fn if not is_windows() else None)

    async def _run_cmd(self, harness: 'TestHarness', cmd: T.List[str]) -> None:
        if self.console_mode is ConsoleUser.GDB:
            stdout = None
            stderr = None
        else:
            stdout = asyncio.subprocess.PIPE
            stderr = asyncio.subprocess.STDOUT \
                if not self.options.split and not self.runobj.needs_parsing \
                else asyncio.subprocess.PIPE

        extra_cmd = []  # type: T.List[str]
        if self.test.protocol is TestProtocol.GTEST:
            gtestname = self.test.name
            if self.test.workdir:
                gtestname = os.path.join(self.test.workdir, self.test.name)
            extra_cmd.append(f'--gtest_output=xml:{gtestname}.xml')

        p = await self._run_subprocess(cmd + extra_cmd,
                                       stdout=stdout,
                                       stderr=stderr,
                                       env=self.runobj.env,
                                       cwd=self.test.workdir)

        parse_task = None
        if self.runobj.needs_parsing:
            parse_coro = self.runobj.parse(harness,
                                           p.stdout_lines(self.console_mode))
            parse_task = asyncio.ensure_future(parse_coro)

        if self.console_mode == ConsoleUser.STDOUT:
            prefix = harness.get_test_num_prefix(self.runobj.num)

            def printer(line: str) -> None:
                ConsoleLogger.print_test_details_line(prefix,
                                                      line,
                                                      flush=True,
                                                      end='')
            ConsoleLogger.print_test_details_header(prefix, 'output')
            stdo_task, stde_task = p.communicate(self.console_mode, printer)
        else:
            stdo_task, stde_task = p.communicate(self.console_mode)
        additional_output = ''
        returncode, result, additional_error = await p.wait(self.runobj.timeout)

        if parse_task is not None:
            res, additional_output, error = await parse_task
            if error:
                additional_error = join_lines(additional_error, error)
            result = result or res

        stdo = await stdo_task if stdo_task else ''
        stde = await stde_task if stde_task else ''
        self.runobj.complete(returncode,
                             result,
                             stdo.strip(),
                             stde.strip(),
                             additional_output.strip(),
                             additional_error.strip())


class TestHarness:
    def __init__(self, options: argparse.Namespace):
        self.options = options
        self.collected_failures = []  # type: T.List[TestRun]
        self.fail_count = 0
        self.expectedfail_count = 0
        self.unexpectedpass_count = 0
        self.success_count = 0
        self.skip_count = 0
        self.timeout_count = 0
        self.test_count = 0
        self.name_max_len = 0
        self.is_run = False
        self.loggers = []         # type: T.List[TestLogger]
        self.loggers.append(ConsoleLogger())
        self.need_console = False

        self.logfile_base = None  # type: T.Optional[str]
        if self.options.logbase and not self.options.gdb:
            namebase = None
            self.logfile_base = os.path.join(self.options.wd, 'meson-logs', self.options.logbase)

            if self.options.wrapper:
                namebase = os.path.basename(self.get_wrapper(self.options)[0])
            elif self.options.setup:
                namebase = self.options.setup.replace(":", "_")

            if namebase:
                self.logfile_base += '-' + namebase.replace(' ', '_')

        startdir = os.getcwd()
        try:
            os.chdir(self.options.wd)
            self.build_data = build.load(os.getcwd())
            if not self.options.setup:
                self.options.setup = self.build_data.test_setup_default_name
            if self.options.benchmark:
                self.tests = self.load_tests('meson_benchmark_setup.dat')
            else:
                self.tests = self.load_tests('meson_test_setup.dat')
        finally:
            os.chdir(startdir)

        ss = set()
        for t in self.tests:
            for s in t.suite:
                ss.add(s)
        self.suites = list(ss)

    def load_tests(self, file_name: str) -> T.List[TestSerialisation]:
        datafile = Path('meson-private') / file_name
        if not datafile.is_file():
            raise TestException(f'Directory {self.options.wd!r} does not seem to be a Meson build directory.')
        with datafile.open('rb') as f:
            objs = check_testdata(pickle.load(f))
        return objs

    def __enter__(self) -> 'TestHarness':
        return self

    def __exit__(self, exc_type: T.Any, exc_value: T.Any, traceback: T.Any) -> None:
        self.close_logfiles()

    def close_logfiles(self) -> None:
        for l in self.loggers:
            l.close()

    def get_test_setup(self, test: T.Optional[TestSerialisation]) -> build.TestSetup:
        if ':' in self.options.setup:
            if self.options.setup not in self.build_data.test_setups:
                sys.exit(f"Unknown test setup '{self.options.setup}'.")
            return self.build_data.test_setups[self.options.setup]
        else:
            full_name = test.project_name + ":" + self.options.setup
            if full_name not in self.build_data.test_setups:
                sys.exit(f"Test setup '{self.options.setup}' not found from project '{test.project_name}'.")
            return self.build_data.test_setups[full_name]

    def merge_setup_options(self, options: argparse.Namespace, test: TestSerialisation) -> T.Dict[str, str]:
        current = self.get_test_setup(test)
        if not options.gdb:
            options.gdb = current.gdb
        if options.gdb:
            options.verbose = True
        if options.timeout_multiplier is None:
            options.timeout_multiplier = current.timeout_multiplier
    #    if options.env is None:
    #        options.env = current.env # FIXME, should probably merge options here.
        if options.wrapper is None:
            options.wrapper = current.exe_wrapper
        elif current.exe_wrapper:
            sys.exit('Conflict: both test setup and command line specify an exe wrapper.')
        return current.env.get_env(os.environ.copy())

    def get_test_runner(self, test: TestSerialisation) -> SingleTestRunner:
        name = self.get_pretty_suite(test)
        options = deepcopy(self.options)
        if self.options.setup:
            env = self.merge_setup_options(options, test)
        else:
            env = os.environ.copy()
        test_env = test.env.get_env(env)
        env.update(test_env)
        if (test.is_cross_built and test.needs_exe_wrapper and
                test.exe_runner and test.exe_runner.found()):
            env['MESON_EXE_WRAPPER'] = join_args(test.exe_runner.get_command())
        return SingleTestRunner(test, env, name, options)

    def process_test_result(self, result: TestRun) -> None:
        if result.res is TestResult.TIMEOUT:
            self.timeout_count += 1
        elif result.res is TestResult.SKIP:
            self.skip_count += 1
        elif result.res is TestResult.OK:
            self.success_count += 1
        elif result.res in {TestResult.FAIL, TestResult.ERROR, TestResult.INTERRUPT}:
            self.fail_count += 1
        elif result.res is TestResult.EXPECTEDFAIL:
            self.expectedfail_count += 1
        elif result.res is TestResult.UNEXPECTEDPASS:
            self.unexpectedpass_count += 1
        else:
            sys.exit(f'Unknown test result encountered: {result.res}')

        if result.res.is_bad():
            self.collected_failures.append(result)
        for l in self.loggers:
            l.log(self, result)

    @property
    def numlen(self) -> int:
        return len(str(self.test_count))

    @property
    def max_left_width(self) -> int:
        return 2 * self.numlen + 2

    def get_test_num_prefix(self, num: int) -> str:
        return '{num:{numlen}}/{testcount} '.format(numlen=self.numlen,
                                                    num=num,
                                                    testcount=self.test_count)

    def format(self, result: TestRun, colorize: bool,
               max_left_width: int = 0,
               left: T.Optional[str] = None,
               middle: T.Optional[str] = None,
               right: T.Optional[str] = None) -> str:
        if left is None:
            left = self.get_test_num_prefix(result.num)

        # A non-default max_left_width lets the logger print more stuff before the
        # name, while ensuring that the rightmost columns remain aligned.
        max_left_width = max(max_left_width, self.max_left_width)

        if middle is None:
            middle = result.name
        extra_mid_width = max_left_width + self.name_max_len + 1 - uniwidth(middle) - uniwidth(left)
        middle += ' ' * max(1, extra_mid_width)

        if right is None:
            right = '{res} {dur:{durlen}.2f}s'.format(
                res=result.res.get_text(colorize),
                dur=result.duration,
                durlen=self.duration_max_len + 3)
            if not (result.res.is_bad() and self.options.print_errorlogs) \
                    and not self.options.verbose \
                    and (result.res.is_bad() or result.needs_parsing):
                detail = result.detail()
                if detail:
                    right += '   ' + detail
        return left + middle + right

    def summary(self) -> str:
        return textwrap.dedent('''\

            Ok:                 {:<4}
            Expected Fail:      {:<4}
            Fail:               {:<4}
            Unexpected Pass:    {:<4}
            Skipped:            {:<4}
            Timeout:            {:<4}
            ''').format(self.success_count, self.expectedfail_count, self.fail_count,
                        self.unexpectedpass_count, self.skip_count, self.timeout_count)

    def total_failure_count(self) -> int:
        return self.fail_count + self.unexpectedpass_count + self.timeout_count

    def doit(self) -> int:
        if self.is_run:
            raise RuntimeError('Test harness object can only be used once.')
        self.is_run = True
        tests = self.get_tests()
        if not tests:
            return 0
        if not self.options.no_rebuild and not rebuild_deps(self.options.wd, tests):
            # We return 125 here in case the build failed.
            # The reason is that exit code 125 tells `git bisect run` that the current
            # commit should be skipped.  Thus users can directly use `meson test` to
            # bisect without needing to handle the does-not-build case separately in a
            # wrapper script.
            sys.exit(125)

        self.name_max_len = max([uniwidth(self.get_pretty_suite(test)) for test in tests])
        startdir = os.getcwd()
        try:
            os.chdir(self.options.wd)
            runners = []             # type: T.List[SingleTestRunner]
            for i in range(self.options.repeat):
                runners.extend(self.get_test_runner(test) for test in tests)
                if i == 0:
                    self.duration_max_len = max([len(str(int(runner.timeout or 99)))
                                                 for runner in runners])
                    # Disable the progress report if it gets in the way
                    self.need_console = any(runner.console_mode is not ConsoleUser.LOGGER
                                            for runner in runners)

            self.test_count = len(runners)
            self.run_tests(runners)
        finally:
            os.chdir(startdir)
        return self.total_failure_count()

    @staticmethod
    def split_suite_string(suite: str) -> T.Tuple[str, str]:
        if ':' in suite:
            split = suite.split(':', 1)
            assert len(split) == 2
            return split[0], split[1]
        else:
            return suite, ""

    @staticmethod
    def test_in_suites(test: TestSerialisation, suites: T.List[str]) -> bool:
        for suite in suites:
            (prj_match, st_match) = TestHarness.split_suite_string(suite)
            for prjst in test.suite:
                (prj, st) = TestHarness.split_suite_string(prjst)

                # the SUITE can be passed as
                #     suite_name
                # or
                #     project_name:suite_name
                # so we need to select only the test belonging to project_name

                # this if handle the first case (i.e., SUITE == suite_name)

                # in this way we can run tests belonging to different
                # (sub)projects which share the same suite_name
                if not st_match and st == prj_match:
                    return True

                # these two conditions are needed to handle the second option
                # i.e., SUITE == project_name:suite_name

                # in this way we select the only the tests of
                # project_name with suite_name
                if prj_match and prj != prj_match:
                    continue
                if st_match and st != st_match:
                    continue
                return True
        return False

    def test_suitable(self, test: TestSerialisation) -> bool:
        if TestHarness.test_in_suites(test, self.options.exclude_suites):
            return False

        if self.options.include_suites:
            # Both force inclusion (overriding add_test_setup) and exclude
            # everything else
            return TestHarness.test_in_suites(test, self.options.include_suites)

        if self.options.setup:
            setup = self.get_test_setup(test)
            if TestHarness.test_in_suites(test, setup.exclude_suites):
                return False

        return True

    def tests_from_args(self, tests: T.List[TestSerialisation]) -> T.Generator[TestSerialisation, None, None]:
        '''
        Allow specifying test names like "meson test foo1 foo2", where test('foo1', ...)

        Also support specifying the subproject to run tests from like
        "meson test subproj:" (all tests inside subproj) or "meson test subproj:foo1"
        to run foo1 inside subproj. Coincidentally also "meson test :foo1" to
        run all tests with that name across all subprojects, which is
        identical to "meson test foo1"
        '''
        for arg in self.options.args:
            if ':' in arg:
                subproj, name = arg.split(':', maxsplit=1)
            else:
                subproj, name = '', arg
            for t in tests:
                if subproj and t.project_name != subproj:
                    continue
                if name and t.name != name:
                    continue
                yield t

    def get_tests(self) -> T.List[TestSerialisation]:
        if not self.tests:
            print('No tests defined.')
            return []

        tests = [t for t in self.tests if self.test_suitable(t)]
        if self.options.args:
            tests = list(self.tests_from_args(tests))

        if not tests:
            print('No suitable tests defined.')
            return []

        return tests

    def flush_logfiles(self) -> None:
        for l in self.loggers:
            l.flush()

    def open_logfiles(self) -> None:
        if not self.logfile_base:
            return

        self.loggers.append(JunitBuilder(self.logfile_base + '.junit.xml'))
        self.loggers.append(JsonLogfileBuilder(self.logfile_base + '.json'))
        self.loggers.append(TextLogfileBuilder(self.logfile_base + '.txt', errors='surrogateescape'))

    @staticmethod
    def get_wrapper(options: argparse.Namespace) -> T.List[str]:
        wrap = []  # type: T.List[str]
        if options.gdb:
            wrap = [options.gdb_path, '--quiet', '--nh']
            if options.repeat > 1:
                wrap += ['-ex', 'run', '-ex', 'quit']
            # Signal the end of arguments to gdb
            wrap += ['--args']
        if options.wrapper:
            wrap += options.wrapper
        return wrap

    def get_pretty_suite(self, test: TestSerialisation) -> str:
        if len(self.suites) > 1 and test.suite:
            rv = TestHarness.split_suite_string(test.suite[0])[0]
            s = "+".join(TestHarness.split_suite_string(s)[1] for s in test.suite)
            if s:
                rv += ":"
            return rv + s + " / " + test.name
        else:
            return test.name

    def run_tests(self, runners: T.List[SingleTestRunner]) -> None:
        try:
            self.open_logfiles()
            # Replace with asyncio.run once we can require Python 3.7
            loop = asyncio.get_event_loop()
            loop.run_until_complete(self._run_tests(runners))
        finally:
            self.close_logfiles()

    def log_subtest(self, test: TestRun, s: str, res: TestResult) -> str:
        rv = ''
        for l in self.loggers:
            tmp = l.log_subtest(self, test, s, res)
            if tmp:
                rv += tmp
        return rv

    def log_start_test(self, test: TestRun) -> None:
        for l in self.loggers:
            l.start_test(self, test)

    async def _run_tests(self, runners: T.List[SingleTestRunner]) -> None:
        semaphore = asyncio.Semaphore(self.options.num_processes)
        futures = deque()  # type: T.Deque[asyncio.Future]
        running_tests = dict() # type: T.Dict[asyncio.Future, str]
        interrupted = False
        ctrlc_times = deque(maxlen=MAX_CTRLC) # type: T.Deque[float]

        async def run_test(test: SingleTestRunner) -> None:
            async with semaphore:
                if interrupted or (self.options.repeat > 1 and self.fail_count):
                    return
                res = await test.run(self)
                self.process_test_result(res)

        def test_done(f: asyncio.Future) -> None:
            if not f.cancelled():
                f.result()
            futures.remove(f)
            try:
                del running_tests[f]
            except KeyError:
                pass

        def cancel_one_test(warn: bool) -> None:
            future = futures.popleft()
            futures.append(future)
            if warn:
                self.flush_logfiles()
                mlog.warning('CTRL-C detected, interrupting {}'.format(running_tests[future]))
            del running_tests[future]
            future.cancel()

        def cancel_all_tests() -> None:
            nonlocal interrupted
            interrupted = True
            while running_tests:
                cancel_one_test(False)

        def sigterm_handler() -> None:
            if interrupted:
                return
            self.flush_logfiles()
            mlog.warning('Received SIGTERM, exiting')
            cancel_all_tests()

        def sigint_handler() -> None:
            # We always pick the longest-running future that has not been cancelled
            # If all the tests have been CTRL-C'ed, just stop
            nonlocal interrupted
            if interrupted:
                return
            ctrlc_times.append(asyncio.get_event_loop().time())
            if len(ctrlc_times) == MAX_CTRLC and ctrlc_times[-1] - ctrlc_times[0] < 1:
                self.flush_logfiles()
                mlog.warning('CTRL-C detected, exiting')
                cancel_all_tests()
            elif running_tests:
                cancel_one_test(True)
            else:
                self.flush_logfiles()
                mlog.warning('CTRL-C detected, exiting')
                interrupted = True

        for l in self.loggers:
            l.start(self)

        if sys.platform != 'win32':
            asyncio.get_event_loop().add_signal_handler(signal.SIGINT, sigint_handler)
            asyncio.get_event_loop().add_signal_handler(signal.SIGTERM, sigterm_handler)
        try:
            for runner in runners:
                if not runner.is_parallel:
                    await complete_all(futures)
                future = asyncio.ensure_future(run_test(runner))
                futures.append(future)
                running_tests[future] = runner.visible_name
                future.add_done_callback(test_done)
                if not runner.is_parallel:
                    await complete(future)
                if self.options.repeat > 1 and self.fail_count:
                    break

            await complete_all(futures)
        finally:
            if sys.platform != 'win32':
                asyncio.get_event_loop().remove_signal_handler(signal.SIGINT)
                asyncio.get_event_loop().remove_signal_handler(signal.SIGTERM)
            for l in self.loggers:
                await l.finish(self)

def list_tests(th: TestHarness) -> bool:
    tests = th.get_tests()
    for t in tests:
        print(th.get_pretty_suite(t))
    return not tests

def rebuild_deps(wd: str, tests: T.List[TestSerialisation]) -> bool:
    def convert_path_to_target(path: str) -> str:
        path = os.path.relpath(path, wd)
        if os.sep != '/':
            path = path.replace(os.sep, '/')
        return path

    if not (Path(wd) / 'build.ninja').is_file():
        print('Only ninja backend is supported to rebuild tests before running them.')
        return True

    ninja = environment.detect_ninja()
    if not ninja:
        print("Can't find ninja, can't rebuild test.")
        return False

    depends = set()            # type: T.Set[str]
    targets = set()            # type: T.Set[str]
    intro_targets = dict()     # type: T.Dict[str, T.List[str]]
    for target in load_info_file(get_infodir(wd), kind='targets'):
        intro_targets[target['id']] = [
            convert_path_to_target(f)
            for f in target['filename']]
    for t in tests:
        for d in t.depends:
            if d in depends:
                continue
            depends.update(d)
            targets.update(intro_targets[d])

    ret = subprocess.run(ninja + ['-C', wd] + sorted(targets)).returncode
    if ret != 0:
        print(f'Could not rebuild {wd}')
        return False

    return True

def run(options: argparse.Namespace) -> int:
    if options.benchmark:
        options.num_processes = 1

    if options.verbose and options.quiet:
        print('Can not be both quiet and verbose at the same time.')
        return 1

    check_bin = None
    if options.gdb:
        options.verbose = True
        if options.wrapper:
            print('Must not specify both a wrapper and gdb at the same time.')
            return 1
        check_bin = 'gdb'

    if options.wrapper:
        check_bin = options.wrapper[0]

    if sys.platform == 'win32':
        loop = asyncio.ProactorEventLoop()
        asyncio.set_event_loop(loop)

    if check_bin is not None:
        exe = ExternalProgram(check_bin, silent=True)
        if not exe.found():
            print(f'Could not find requested program: {check_bin!r}')
            return 1

    with TestHarness(options) as th:
        try:
            if options.list:
                return list_tests(th)
            return th.doit()
        except TestException as e:
            print('Meson test encountered an error:\n')
            if os.environ.get('MESON_FORCE_BACKTRACE'):
                raise e
            else:
                print(e)
            return 1

def run_with_args(args: T.List[str]) -> int:
    parser = argparse.ArgumentParser(prog='meson test')
    add_arguments(parser)
    options = parser.parse_args(args)
    return run(options)