import errno
import io
import logging
import os
import random
import re
import socket
import stat
import sys
import threading
import time
import traceback
import unittest

try:
    import configparser
except ImportError:
    import ConfigParser as configparser

import psutil
if sys.version_info < (3, 0):
    import subprocess32 as subprocess
else:
    import subprocess

import mitogen.core
import mitogen.fork
import mitogen.master
import mitogen.utils

try:
    import faulthandler
except ImportError:
    faulthandler = None

try:
    import urlparse
except ImportError:
    import urllib.parse as urlparse

try:
    from cStringIO import StringIO
except ImportError:
    from io import StringIO

try:
    BaseException
except NameError:
    BaseException = Exception


LOG = logging.getLogger(__name__)

DISTRO = os.environ.get('MITOGEN_TEST_DISTRO', 'debian9')
IMAGE_TEMPLATE = os.environ.get(
    'MITOGEN_TEST_IMAGE_TEMPLATE',
    'public.ecr.aws/n5z0e8q9/%(distro)s-test',
)

TESTS_DIR =                     os.path.join(os.path.dirname(__file__))
ANSIBLE_LIB_DIR =               os.path.join(TESTS_DIR, 'ansible', 'lib')
ANSIBLE_MODULE_UTILS_DIR =      os.path.join(TESTS_DIR, 'ansible', 'lib', 'module_utils')
ANSIBLE_MODULES_DIR =           os.path.join(TESTS_DIR, 'ansible', 'lib', 'modules')
DATA_DIR =                      os.path.join(TESTS_DIR, 'data')
MODS_DIR =                      os.path.join(TESTS_DIR, 'data', 'importer')

sys.path.append(DATA_DIR)
sys.path.append(MODS_DIR)


if mitogen.is_master:
    mitogen.utils.log_to_file()

if faulthandler is not None:
    faulthandler.enable()


#
# Temporary hack: Operon changed logging somewhat, and this broke LogCapturer /
# log_handler_test.
#

mitogen.core.LOG.propagate = True


def base_executable(executable=None):
    '''Return the path of the Python executable used to create the virtualenv.
    '''
    # https://docs.python.org/3/library/venv.html
    # https://github.com/pypa/virtualenv/blob/main/src/virtualenv/discovery/py_info.py
    # https://virtualenv.pypa.io/en/16.7.9/reference.html#compatibility-with-the-stdlib-venv-module
    if executable is None:
        executable = sys.executable

    if not executable:
        raise ValueError

    try:
        base_executable = sys._base_executable
    except AttributeError:
        base_executable = None

    if base_executable and base_executable != executable:
        return base_executable

    # Python 2.x only has sys.base_prefix if running outside a virtualenv.
    try:
        sys.base_prefix
    except AttributeError:
        # Python 2.x outside a virtualenv
        return executable

    # Python 3.3+ has sys.base_prefix. In a virtualenv it differs to sys.prefix.
    if sys.base_prefix == sys.prefix:
        return executable

    while executable.startswith(sys.prefix) and stat.S_ISLNK(os.lstat(executable).st_mode):
        dirname = os.path.dirname(executable)
        target = os.path.join(dirname, os.readlink(executable))
        executable = os.path.abspath(os.path.normpath(target))
        print(executable)

    if executable.startswith(sys.base_prefix):
        return executable

    # Virtualenvs record details in pyvenv.cfg
    parser = configparser.RawConfigParser()
    with io.open(os.path.join(sys.prefix, 'pyvenv.cfg'), encoding='utf-8') as f:
        content = u'[virtualenv]\n' + f.read()
    try:
        parser.read_string(content)
    except AttributeError:
        parser.readfp(io.StringIO(content))

    # virtualenv style pyvenv.cfg includes the base executable.
    # venv style pyvenv.cfg doesn't.
    try:
        return parser.get(u'virtualenv', u'base-executable')
    except configparser.NoOptionError:
        pass

    basename = os.path.basename(executable)
    home = parser.get(u'virtualenv', u'home')
    return os.path.join(home, basename)


def data_path(suffix):
    path = os.path.join(DATA_DIR, suffix)
    if path.endswith('.key'):
        # SSH is funny about private key permissions.
        os.chmod(path, int('0600', 8))
    return path


def retry(fn, on, max_attempts, delay):
    for i in range(max_attempts):
        try:
            return fn()
        except on:
            if i >= max_attempts - 1:
                raise
            else:
                time.sleep(delay)


def threading__thread_is_alive(thread):
    """Return whether the thread is alive (Python version compatibility shim).

    On Python >= 3.8 thread.isAlive() is deprecated (removed in Python 3.9).
    On Python <= 2.5 thread.is_alive() isn't present (added in Python 2.6).
    """
    try:
        return thread.is_alive()
    except AttributeError:
        return thread.isAlive()


def threading_thread_name(thread):
    try:
        return thread.name  # Available in Python 2.6+
    except AttributeError:
        return thread.getName()  # Deprecated in Python 3.10+


def wait_for_port(
        host,
        port,
        pattern=None,
        connect_timeout=0.5,
        receive_timeout=0.5,
        overall_timeout=5.0,
        sleep=0.1,
        ):
    """Attempt to connect to host/port, for upto overall_timeout seconds.
    If a regex pattern is supplied try to find it in the initial data.
    Return None on success, or raise on error.
    """
    start = mitogen.core.now()
    end = start + overall_timeout
    addr = (host, port)

    while mitogen.core.now() < end:
        sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        sock.settimeout(connect_timeout)
        try:
            sock.connect(addr)
        except socket.error:
            # Failed to connect. So wait then retry.
            time.sleep(sleep)
            continue

        if not pattern:
            # Success: We connected & there's no banner check to perform.
            sock.shutdown(socket.SHUT_RDWR)
            sock.close()
            return

        sock.settimeout(receive_timeout)
        data = mitogen.core.b('')
        found = False
        while mitogen.core.now() < end:
            try:
                resp = sock.recv(1024)
            except socket.timeout:
                # Server stayed up, but had no data. Retry the recv().
                continue

            if not resp:
                # Server went away. Wait then retry the connection.
                time.sleep(sleep)
                break

            data += resp
            if re.search(mitogen.core.b(pattern), data):
                found = True
                break

        try:
            sock.shutdown(socket.SHUT_RDWR)
        except socket.error:
            e = sys.exc_info()[1]
            # On Mac OS X - a BSD variant - the above code only succeeds if the
            # operating system thinks that the socket is still open when
            # shutdown() is invoked. If Python is too slow and the FIN packet
            # arrives before that statement can be reached, then OS X kills the
            # sock.shutdown() statement with:
            #
            #    socket.error: [Errno 57] Socket is not connected
            #
            # Protect shutdown() with a try...except that catches the
            # socket.error, test to make sure Errno is right, and ignore it if
            # Errno matches.
            if e.errno == 57:
                pass
            else:
                raise
        sock.close()

        if found:
            # Success: We received the banner & found the desired pattern
            return
    else:
        # Failure: The overall timeout expired
        if pattern:
            raise socket.timeout('Timed out while searching for %r from %s:%s'
                                 % (pattern, host, port))
        else:
            raise socket.timeout('Timed out while connecting to %s:%s'
                                 % (host, port))


def sync_with_broker(broker, timeout=10.0):
    """
    Insert a synchronization barrier between the calling thread and the Broker
    thread, ensuring it has completed at least one full IO loop before
    returning.

    Used to block while asynchronous stuff (like defer()) happens on the
    broker.
    """
    sem = mitogen.core.Latch()
    broker.defer(sem.put, None)
    sem.get(timeout=timeout)


def log_fd_calls():
    mypid = os.getpid()
    l = threading.Lock()
    real_pipe = os.pipe
    def pipe():
        l.acquire()
        try:
            rv = real_pipe()
            if mypid == os.getpid():
                sys.stdout.write('\n%s\n' % (rv,))
                traceback.print_stack(limit=3)
                sys.stdout.write('\n')
            return rv
        finally:
            l.release()

    os.pipe = pipe

    real_socketpair = socket.socketpair
    def socketpair(*args):
        l.acquire()
        try:
            rv = real_socketpair(*args)
            if mypid == os.getpid():
                sys.stdout.write('\n%s -> %s\n' % (args, rv))
                traceback.print_stack(limit=3)
                sys.stdout.write('\n')
                return rv
        finally:
            l.release()

    socket.socketpair = socketpair

    real_dup2 = os.dup2
    def dup2(*args):
        l.acquire()
        try:
            real_dup2(*args)
            if mypid == os.getpid():
                sys.stdout.write('\n%s\n' % (args,))
                traceback.print_stack(limit=3)
                sys.stdout.write('\n')
        finally:
            l.release()

    os.dup2 = dup2

    real_dup = os.dup
    def dup(*args):
        l.acquire()
        try:
            rv = real_dup(*args)
            if mypid == os.getpid():
                sys.stdout.write('\n%s -> %s\n' % (args, rv))
                traceback.print_stack(limit=3)
                sys.stdout.write('\n')
            return rv
        finally:
            l.release()

    os.dup = dup


class CaptureStreamHandler(logging.StreamHandler):
    def __init__(self, *args, **kwargs):
        logging.StreamHandler.__init__(self, *args, **kwargs)
        self.msgs = []

    def emit(self, msg):
        self.msgs.append(msg)
        logging.StreamHandler.emit(self, msg)


class LogCapturer(object):
    def __init__(self, name=None):
        self.sio = StringIO()
        self.logger = logging.getLogger(name)
        self.handler = CaptureStreamHandler(self.sio)
        self.old_propagate = self.logger.propagate
        self.old_handlers = self.logger.handlers
        self.old_level = self.logger.level

    def start(self):
        self.logger.handlers = [self.handler]
        self.logger.propagate = False
        self.logger.level = logging.DEBUG

    def raw(self):
        s = self.sio.getvalue()
        # Python 2.x logging package hard-wires UTF-8 output.
        if isinstance(s, mitogen.core.BytesType):
            s = s.decode('utf-8')
        return s

    def msgs(self):
        return self.handler.msgs

    def __enter__(self):
        self.start()
        return self

    def __exit__(self, _1, _2, _3):
        self.stop()

    def stop(self):
        self.logger.level = self.old_level
        self.logger.handlers = self.old_handlers
        self.logger.propagate = self.old_propagate
        return self.raw()


class TestCase(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
        # This is done in setUpClass() so we have a chance to run before any
        # Broker() instantiations in setUp() etc.
        mitogen.fork.on_fork()
        cls._fds_before = psutil.Process().open_files()
        # Ignore children started by external packages - in particular
        # multiprocessing.resource_tracker.main()`, started when some Ansible
        # versions instantiate a `multithreading.Lock()`.
        cls._children_before = frozenset(psutil.Process().children())
        super(TestCase, cls).setUpClass()

    ALLOWED_THREADS = set([
        'MainThread',
        'mitogen.master.join_thread_async'
    ])

    def _teardown_check_threads(self):
        counts = {}
        for thread in threading.enumerate():
            name = threading_thread_name(thread)
            # Python 2.4: enumerate() may return stopped threads.
            assert \
                not threading__thread_is_alive(thread) \
                or name in self.ALLOWED_THREADS, \
                'Found thread %r still running after tests.' % (name,)
            counts[name] = counts.get(name, 0) + 1

        for name in counts:
            assert counts[name] == 1, \
                'Found %d copies of thread %r running after tests.' % (
                    counts[name], name
                )

    def _teardown_check_fds(self):
        mitogen.core.Latch._on_fork()
        fds_after = psutil.Process().open_files()
        fds_leaked = len(self._fds_before) != len(fds_after)
        if not fds_leaked:
            return
        else:
            if sys.platform == 'linux':
                subprocess.check_call(
                    'lsof +E -w -p %i | grep -vw mem' % (os.getpid(),),
                    shell=True,
                )
            else:
                subprocess.check_call(
                    'lsof -w -p %i | grep -vw mem' % (os.getpid(),),
                    shell=True,
                )
            assert 0, "%s leaked FDs: %s\nBefore:\t%s\nAfter:\t%s" % (
                self, fds_leaked, self._fds_before, fds_after,
            )

    # Some class fixtures (like Ansible MuxProcess) start persistent children
    # for the duration of the class.
    no_zombie_check = False

    def _teardown_check_zombies(self):
        if self.no_zombie_check:
            return

        # pid=0: Wait for any child process in the same process group as us.
        # WNOHANG: Don't block if no processes ready to report status.
        try:
            pid, status = os.waitpid(0, os.WNOHANG)
        except OSError as e:
            # ECHILD: there are no child processes in our group.
            if e.errno == errno.ECHILD:
                return
            raise

        if pid:
            assert 0, "%s failed to reap subprocess %d (status %d)." % (
                self, pid, status
            )

        children_after = frozenset(psutil.Process().children())
        children_leaked = children_after.difference(self._children_before)
        if not children_leaked:
            return

        print('Leaked children of unit test process:')
        subprocess.check_call(
            ['ps', '-o', 'user,pid,%cpu,%mem,vsz,rss,tty,stat,start,time,command', '-ww', '-p',
             ','.join(str(p.pid) for p in children_leaked),
            ],
        )
        if self._children_before:
            print('Pre-existing children of unit test process:')
            subprocess.check_call(
                ['ps', '-o', 'user,pid,%cpu,%mem,vsz,rss,tty,stat,start,time,command', '-ww', '-p',
                 ','.join(str(p.pid) for p in self._children_before),
                ],
            )
        assert 0, "%s leaked still-running subprocesses." % (self,)

    def tearDown(self):
        self._teardown_check_zombies()
        self._teardown_check_threads()
        self._teardown_check_fds()
        super(TestCase, self).tearDown()

    def assertRaises(self, exc, func, *args, **kwargs):
        """Like regular assertRaises, except return the exception that was
        raised. Can't use context manager because tests must run on Python2.4"""
        try:
            func(*args, **kwargs)
        except exc:
            e = sys.exc_info()[1]
            return e
        except BaseException:
            LOG.exception('Original exception')
            e = sys.exc_info()[1]
            assert 0, '%r raised %r, not %r' % (func, e, exc)
        assert 0, '%r did not raise %r' % (func, exc)


def get_docker_host():
    # Duplicated in ci_lib
    url = os.environ.get('DOCKER_HOST')
    if url in (None, 'http+docker://localunixsocket'):
        return 'localhost'

    parsed = urlparse.urlparse(url)
    return parsed.netloc.partition(':')[0]


class DockerizedSshDaemon(object):
    PORT_RE = re.compile(
        # e.g. 0.0.0.0:32771, :::32771, [::]:32771'
        r'(?P<addr>[0-9.]+|::|\[[a-f0-9:.]+\]):(?P<port>[0-9]+)',
    )

    @classmethod
    def get_port(cls, container):
        s = subprocess.check_output(['docker', 'port', container, '22/tcp'])
        m = cls.PORT_RE.search(s.decode())
        if not m:
            raise ValueError('could not find SSH port in: %r' % (s,))
        return int(m.group('port'))

    def start_container(self):
        try:
            subprocess.check_output(['docker', '--version'])
        except Exception:
            raise unittest.SkipTest('Docker binary is unavailable')

        self.container_name = 'mitogen-test-%08x' % (random.getrandbits(64),)
        args = [
            'docker',
            'run',
            '--detach',
            '--privileged',
            '--publish-all',
            '--name', self.container_name,
            self.image,
        ]
        subprocess.check_output(args)

    def __init__(self, distro=DISTRO, image_template=IMAGE_TEMPLATE):
        # Code duplicated in ci_lib.py, both should be updated together
        distro_pattern = re.compile(r'''
            (?P<distro>(?P<family>[a-z]+)[0-9]+)
            (?:-(?P<py>py3))?
            (?:\*(?P<count>[0-9]+))?
            ''',
            re.VERBOSE,
        )
        d = distro_pattern.match(distro).groupdict(default=None)

        if d.pop('py') == 'py3':
            self.python_path = '/usr/bin/python3'
        else:
            self.python_path = '/usr/bin/python'

        self.image = image_template % d
        self.start_container()
        self.host = self.get_host()
        self.port = self.get_port(self.container_name)

    def get_host(self):
        return get_docker_host()

    def wait_for_sshd(self):
        wait_for_port(self.get_host(), self.port, pattern='OpenSSH')

    def check_processes(self):
        # Get Accounting name (ucomm) & command line (args) of each process
        # in the container. No truncation (-ww). No column headers (foo=).
        ps_output = subprocess.check_output([
            'docker', 'exec', self.container_name,
            'ps', '-w', '-w', '-o', 'ucomm=', '-o', 'args=',
        ])
        ps_lines = ps_output.decode().splitlines()
        processes = [tuple(line.split(None, 1)) for line in ps_lines]
        counts = {}
        for ucomm, _ in processes:
            counts[ucomm] = counts.get(ucomm, 0) + 1

        if counts != {'ps': 1, 'sshd': 1}:
            assert 0, (
                'Docker container %r contained extra running processes '
                'after test completed: %r' % (
                    self.container_name,
                    processes,
                )
            )

    def close(self):
        args = ['docker', 'rm', '-f', self.container_name]
        subprocess.check_output(args)


class BrokerMixin(object):
    broker_class = mitogen.master.Broker

    # Flag for tests that shutdown the broker themself
    # e.g. unix_test.ListenerTest
    broker_shutdown = False

    def setUp(self):
        super(BrokerMixin, self).setUp()
        self.broker = self.broker_class()

    def tearDown(self):
        if not self.broker_shutdown:
            self.broker.shutdown()
        self.broker.join()
        del self.broker
        super(BrokerMixin, self).tearDown()

    def sync_with_broker(self):
        sync_with_broker(self.broker)


class RouterMixin(BrokerMixin):
    router_class = mitogen.master.Router

    def setUp(self):
        super(RouterMixin, self).setUp()
        self.router = self.router_class(self.broker)

    def tearDown(self):
        del self.router
        super(RouterMixin, self).tearDown()


class DockerMixin(RouterMixin):
    @classmethod
    def setUpClass(cls):
        super(DockerMixin, cls).setUpClass()
        if os.environ.get('SKIP_DOCKER_TESTS'):
            raise unittest.SkipTest('SKIP_DOCKER_TESTS is set')

        # we want to be able to override test distro for some tests that need a different container spun up
        daemon_args = {}
        if hasattr(cls, 'mitogen_test_distro'):
            daemon_args['mitogen_test_distro'] = cls.mitogen_test_distro

        cls.dockerized_ssh = DockerizedSshDaemon(**daemon_args)
        cls.dockerized_ssh.wait_for_sshd()

    @classmethod
    def tearDownClass(cls):
        retry(
            cls.dockerized_ssh.check_processes,
            on=AssertionError,
            max_attempts=5,
            delay=0.1,
        )
        cls.dockerized_ssh.close()
        super(DockerMixin, cls).tearDownClass()

    @property
    def docker_ssh_default_kwargs(self):
        return {
            'hostname': self.dockerized_ssh.host,
            'port': self.dockerized_ssh.port,
            'check_host_keys': 'ignore',
            'ssh_debug_level': 3,
            # https://www.openssh.com/legacy.html
            # ssh-rsa uses SHA1. Least worst available with CentOS 7 sshd.
            # Rejected by default in newer ssh clients (e.g. Ubuntu 22.04).
            # Duplicated cases in
            #   - tests/ansible/ansible.cfg
            #   - tests/ansible/integration/connection_delegation/delegate_to_template.yml
            #   - tests/ansible/integration/connection_delegation/stack_construction.yml
            #   - tests/ansible/integration/process/unix_socket_cleanup.yml
            #   - tests/ansible/integration/ssh/variables.yml
            #   - tests/testlib.py
            'ssh_args': [
                '-o', 'HostKeyAlgorithms +ssh-rsa',
                '-o', 'PubkeyAcceptedKeyTypes +ssh-rsa',
            ],
            'python_path': self.dockerized_ssh.python_path,
        }

    def docker_ssh(self, **kwargs):
        for k, v in self.docker_ssh_default_kwargs.items():
            kwargs.setdefault(k, v)
        return self.router.ssh(**kwargs)

    def docker_ssh_any(self, **kwargs):
        return self.docker_ssh(
            username='mitogen__has_sudo_nopw',
            password='has_sudo_nopw_password',
        )