githubmirror/bash-completion

test/t/conftest.py
1 018 строк33 KB
Koichi Murase
feat: support the "startup{,-core}" directory
23 июн 2026, 01:59
23 июн 2026, 01:59d6169af
import difflib
import os
import re
import shlex
import subprocess
import sys
import time
from enum import Enum
from pathlib import Path
from types import TracebackType
from typing import (
    Callable,
    Dict,
    Iterable,
    Iterator,
    List,
    Optional,
    TextIO,
    Tuple,
    Type,
)

import pexpect  # type: ignore[import]
import pytest

PS1 = "/@"
MAGIC_MARK = "__MaGiC-maRKz-NEtXZVZfKC__"
MAGIC_MARK2 = "Re8SCgEdfN"


def find_unique_completion_pair(
    items: Iterable[str],
) -> Optional[Tuple[str, str]]:
    result = None
    bestscore = 0
    sitems = sorted(set(items))
    for i in range(len(sitems)):
        cur = sitems[i]
        curlen = len(cur)
        prv = sitems[i - 1] if i != 0 else ""
        prvlen = len(prv)
        nxt = sitems[i + 1] if i < len(sitems) - 1 else ""
        nxtlen = len(nxt)
        diffprv = prv == ""
        diffnxt = nxt == ""
        # Analyse each item of the list and look for the minimum length of the
        # partial prefix which is distinct from both nxt and prv. The list
        # is sorted so the prefix will be unique in the entire list.
        for j in range(curlen):
            curchar = cur[j]
            if not diffprv and (j >= prvlen or prv[j] != curchar):
                diffprv = True
            if not diffnxt and (j >= nxtlen or nxt[j] != curchar):
                diffnxt = True
            if diffprv and diffnxt:
                break
        # At the end of the loop, j is the index of last character of
        # the unique partial prefix. The length is one plus that.
        parlen = j + 1
        if parlen >= curlen:
            continue
        # Try to find the most "readable pair"; look for a long pair where
        # part is about half of full.
        if parlen < curlen / 2:
            parlen = int(curlen / 2)
        score = curlen - parlen
        if score > bestscore:
            bestscore = score
            result = (cur[:parlen], cur)
    return result


@pytest.fixture(scope="class")
def output_sort_uniq(bash: pexpect.spawn) -> Callable[[str], List[str]]:
    def _output_sort_uniq(command: str) -> List[str]:
        return sorted(
            set(  # weed out possible duplicates
                assert_bash_exec(bash, command, want_output=True).split()
            )
        )

    return _output_sort_uniq


@pytest.fixture(scope="class")
def part_full_user(
    bash: pexpect.spawn, output_sort_uniq: Callable[[str], List[str]]
) -> Optional[Tuple[str, str]]:
    res = output_sort_uniq("compgen -u")
    pair = find_unique_completion_pair(res)
    if not pair:
        pytest.skip("No suitable test user found")
    return pair


@pytest.fixture(scope="class")
def part_full_group(
    bash: pexpect.spawn, output_sort_uniq: Callable[[str], List[str]]
) -> Optional[Tuple[str, str]]:
    res = output_sort_uniq("compgen -g")
    pair = find_unique_completion_pair(res)
    if not pair:
        pytest.skip("No suitable test user found")
    return pair


@pytest.fixture(scope="class")
def hosts(bash: pexpect.spawn) -> List[str]:
    output = assert_bash_exec(bash, "compgen -A hostname", want_output=True)
    return sorted(set(output.split() + _avahi_hosts(bash)))


@pytest.fixture(scope="class")
def avahi_hosts(bash: pexpect.spawn) -> List[str]:
    return _avahi_hosts(bash)


def _avahi_hosts(bash: pexpect.spawn) -> List[str]:
    output = assert_bash_exec(
        bash,
        "! type avahi-browse &>/dev/null || "
        "avahi-browse -cpr _workstation._tcp 2>/dev/null "
        "| command grep ^= | cut -d';' -f7",
        want_output=None,
    )
    return sorted(set(output.split()))


@pytest.fixture(scope="class")
def known_hosts(bash: pexpect.spawn) -> List[str]:
    output = assert_bash_exec(
        bash,
        '_comp_compgen_known_hosts ""; '
        r'printf "%s\n" "${COMPREPLY[@]}"; unset -v COMPREPLY',
        want_output=True,
    )
    return sorted(set(output.split()))


@pytest.fixture(scope="class")
def user_home(bash: pexpect.spawn) -> Tuple[str, str]:
    user = assert_bash_exec(
        bash, 'id -un 2>/dev/null || echo "$USER"', want_output=True
    ).strip()
    # We used to echo $HOME here, but we expect that it will be consistent with
    # ~user as far as bash is concerned which may not hold.
    home = assert_bash_exec(bash, "echo ~%s" % user, want_output=True).strip()
    return (user, home)


def partialize(
    bash: pexpect.spawn, items: Iterable[str]
) -> Tuple[str, List[str]]:
    """
    Get list of items starting with the first char of first of items.

    Disregard items starting with a COMP_WORDBREAKS character
    (e.g. a colon ~ IPv6 address), they are special cases requiring
    special tests.
    """
    first_char = None
    comp_wordbreaks = assert_bash_exec(
        bash,
        'printf "%s" "$COMP_WORDBREAKS"',
        want_output=True,
        want_newline=False,
    )
    partial_items = []
    for item in sorted(items):
        if first_char is None:
            if item[0] not in comp_wordbreaks:
                first_char = item[0]
                partial_items.append(item)
        elif item.startswith(first_char):
            partial_items.append(item)
        else:
            break
    if first_char is None:
        pytest.skip("Could not generate partial items list from %s" % items)
        # superfluous/dead code to assist mypy; pytest.skip always raises
        assert first_char is not None
    return first_char, partial_items


def get_testdir():
    return os.path.abspath(os.path.join(os.path.dirname(__file__), os.pardir))


@pytest.fixture(scope="session")
def test_session_tmpdir(tmp_path_factory) -> Path:
    tmpdir = tmp_path_factory.mktemp("bash-completion.session.")

    user_dir_1 = tmpdir / "bash-completion"
    user_dir_2 = tmpdir / "bash-completion-fallback"
    user_dir_1.mkdir()
    user_dir_2.mkdir()

    bash_completion_root = Path(get_testdir()).parent

    completions_1 = user_dir_1 / "completions"
    completions_2 = user_dir_2 / "completions"
    completions_1.symlink_to(bash_completion_root / "completions-core")
    completions_2.symlink_to(bash_completion_root / "completions-fallback")

    helpers_1 = user_dir_1 / "helpers-core"
    helpers_2 = user_dir_2 / "helpers-core"
    helpers_1.symlink_to(bash_completion_root / "helpers-core")
    helpers_2.symlink_to(bash_completion_root / "helpers-core")

    compat_dir = tmpdir / "bash_completion.d"
    compat_dir.mkdir()

    # Create an empty temporary file for HISTFILE.
    #
    # To prevent the tested Bash processes from writing to the user's
    # history file or any other files, we prepare an empty temporary
    # file for each test.
    #
    # - Note that HISTFILE=/dev/null may not work.  It results in the
    #   removal of the device /dev/null and the creation of a regular
    #   file at /dev/null when the number of commands reach
    #   HISTFILESIZE due to a bug in bash 4.3.  This causes execution of
    #   garbage through BASH_COMPLETION_USER_FILE=/dev/null.
    # - Note also that "unset -v HISTFILE" in "test/config/bashrc" was not
    #   adopted because "test/config/bashrc" is loaded after the
    #   history is read from the history file.
    #
    histfile = tmpdir / "bash_history"
    histfile.touch()

    return tmpdir


@pytest.fixture(scope="class")
def bash(request, test_session_tmpdir, tmp_path_factory) -> pexpect.spawn:
    logfile: Optional[TextIO] = None
    bash = None

    if os.environ.get("BASH_COMPLETION_TEST_LOGFILE"):
        logfile = open(os.environ["BASH_COMPLETION_TEST_LOGFILE"], "w")
    elif os.environ.get("CI"):
        logfile = sys.stdout

    testdir = get_testdir()

    try:
        env = os.environ.copy()
        env.update(
            dict(
                SRCDIR=testdir,  # TODO needed at least by bashrc
                SRCDIRABS=testdir,
                PS1=PS1,
                INPUTRC="%s/config/inputrc" % testdir,
                TERM="dumb",
                LC_COLLATE="C",  # to match Python's default locale unaware sort
                _comp__test_session_tmpdir=str(test_session_tmpdir),
                HISTFILE=str(test_session_tmpdir / "bash_history"),
            )
        )

        marker = request.node.get_closest_marker("bashcomp")

        # Set up the current working directory
        cwd = None
        if marker:
            if "cwd" in marker.kwargs and marker.kwargs.get("cwd") is not None:
                cwd = os.path.join(
                    testdir, "fixtures", marker.kwargs.get("cwd")
                )
            elif "temp_cwd" in marker.kwargs and marker.kwargs.get("temp_cwd"):
                tmpdir = tmp_path_factory.mktemp("bash-completion.bash.")
                cwd = str(tmpdir)
        if cwd is None:
            cwd = os.path.join(testdir, "fixtures")
        os.chdir(cwd)

        # Start bash
        bash = pexpect.spawn(
            "%s --norc" % os.environ.get("BASH_COMPLETION_TEST_BASH", "bash"),
            maxread=os.environ.get(
                "BASH_COMPLETION_TEST_PEXPECT_MAXREAD", 20000
            ),
            logfile=logfile,
            cwd=cwd,
            env=env,
            encoding="utf-8",  # TODO? or native or...?
            # FIXME: Tests shouldn't depend on dimensions, but it's difficult to
            # expect robustly enough for Bash to wrap lines anywhere (e.g. inside
            # MAGIC_MARK).  Increase window width to reduce wrapping.
            dimensions=(24, 240),
            # TODO? codec_errors="replace",
        )
        # See https://pexpect.readthedocs.io/en/stable/commonissues.html?highlight=delaybeforesend#timing-issue-with-send-and-sendline
        # We don't need the delay, and it makes running the tests much slower.
        bash.delaybeforesend = 0
        bash.expect_exact(PS1)

        # Load bashrc and bash_completion
        bash_completion = os.environ.get(
            "BASH_COMPLETION_TEST_BASH_COMPLETION",
            "%s/../bash_completion" % testdir,
        )
        assert_bash_exec(bash, "source '%s/config/bashrc'" % testdir)
        assert_bash_exec(bash, "source '%s'" % bash_completion)

        # Use command name from marker if set, or grab from test filename
        cmd = None  # type: Optional[str]
        cmd_found = False
        if marker:
            cmd = marker.kwargs.get("cmd")
            cmd_found = "cmd" in marker.kwargs
            # Run pre-test commands, early so they're usable in skipif
            for pre_cmd in marker.kwargs.get("pre_cmds", []):
                assert_bash_exec(bash, pre_cmd, want_output=None)
            # Process skip and xfail conditions
            skipif = marker.kwargs.get("skipif")
            if skipif:
                try:
                    assert_bash_exec(bash, skipif, want_output=None)
                except AssertionError:
                    pass
                else:
                    bash.close()
                    bash = None
                    pytest.skip(skipif)
            xfail = marker.kwargs.get("xfail")
            if xfail:
                try:
                    assert_bash_exec(bash, xfail, want_output=None)
                except AssertionError:
                    pass
                else:
                    pytest.xfail(xfail)
        if not cmd_found:
            match = re.search(
                r"^test_(.+)\.py$", os.path.basename(str(request.fspath))
            )
            if match:
                cmd = match.group(1)
        if (
            marker
            and marker.kwargs
            and marker.kwargs.get("require_cmd", False)
        ):
            if not is_bash_type(bash, cmd):
                pytest.skip("Command not found")

        request.cls.cmd = cmd

        if (cmd_found and cmd is None) or is_testable(bash, cmd):
            before_env = get_env(bash)
            yield bash
            # Not exactly sure why, but some errors leave bash in state where
            # getting the env here would fail and trash our test output. So
            # reset to a good state first (Ctrl+C, expect prompt).
            bash.sendintr()
            bash.expect_exact(PS1)
            diff_env(
                before_env,
                get_env(bash),
                marker.kwargs.get("ignore_env") if marker else "",
            )

        if marker:
            for post_cmd in marker.kwargs.get("post_cmds", []):
                assert_bash_exec(bash, post_cmd, want_output=None)

    finally:
        # Clean up
        if bash:
            bash.close()
        if logfile and logfile != sys.stdout:
            logfile.close()


def is_testable(bash: pexpect.spawn, cmd: Optional[str]) -> bool:
    if not cmd:
        pytest.fail("Could not resolve name of command to test")
        return False
    if not load_completion_for(bash, cmd):
        pytest.skip("No completion for command %s" % cmd)
    return True


def is_bash_type(bash: pexpect.spawn, cmd: Optional[str]) -> bool:
    if not cmd:
        return False
    typecmd = "type %s &>/dev/null && echo -n 0 || echo -n 1" % cmd
    bash.sendline(typecmd)
    bash.expect_exact(typecmd + "\r\n")
    result = bash.expect_exact(["0", "1"]) == 0
    bash.expect_exact(PS1)
    return result


def load_completion_for(bash: pexpect.spawn, cmd: str) -> bool:
    try:
        # Allow _comp_load to fail so we can test completions
        # that are directly loaded in bash_completion without a separate file.
        assert_bash_exec(bash, "_comp_load -- %s || :" % cmd)
        assert_bash_exec(bash, "complete -p -- %s &>/dev/null" % cmd)
    except AssertionError:
        return False
    return True


def assert_bash_exec(
    bash: pexpect.spawn,
    cmd: str,
    want_output: Optional[bool] = False,
    want_newline=True,
) -> str:
    """
    :param want_output: if None, don't care if got output or not
    """

    # Send command
    bash.sendline(cmd)
    bash.expect_exact(cmd)

    # Find prompt, output is before it
    bash.expect_exact("%s%s" % ("\r\n" if want_newline else "", PS1))
    output = bash.before

    # Retrieve exit status
    echo = "echo $?"
    bash.sendline(echo)
    got = bash.expect(
        [
            r"^%s\r\n(\d+)\r\n%s" % (re.escape(echo), re.escape(PS1)),
            PS1,
            pexpect.EOF,
            pexpect.TIMEOUT,
        ]
    )
    status = bash.match.group(1) if got == 0 else "unknown"

    assert status == "0", 'Error running "%s": exit status=%s, output="%s"' % (
        cmd,
        status,
        output,
    )
    if want_output is not None:
        if output:
            assert want_output, (
                'Unexpected output from "%s": exit status=%s, output="%s"'
                % (
                    cmd,
                    status,
                    output,
                )
            )
        else:
            assert not want_output, (
                'Expected output from "%s": exit status=%s, output="%s"'
                % (
                    cmd,
                    status,
                    output,
                )
            )

    return output


class bash_env_saved:
    counter: int = 0

    class saved_state(Enum):
        ChangesDetected = 1
        ChangesIgnored = 2

    def __init__(self, bash: pexpect.spawn, sendintr: bool = False):
        bash_env_saved.counter += 1
        self.prefix: str = "_comp__test_%d" % bash_env_saved.counter

        self.bash = bash
        self.cwd_changed: bool = False
        self.saved_set: Dict[str, bash_env_saved.saved_state] = {}
        self.saved_shopt: Dict[str, bash_env_saved.saved_state] = {}
        self.saved_variables: Dict[str, bash_env_saved.saved_state] = {}
        self.sendintr = sendintr

        self.noexcept: bool = False
        self.captured_error: Optional[Exception] = None

    def __enter__(self):
        return self

    def __exit__(
        self,
        exc_type: Optional[Type[BaseException]],
        exc_value: Optional[BaseException],
        exc_traceback: Optional[TracebackType],
    ) -> None:
        self._restore_env()
        return None

    def _safe_sendintr(self):
        try:
            self.bash.sendintr()
            self.bash.expect_exact(PS1)
        except Exception as e:
            if self.noexcept:
                self.captured_error = e
            else:
                raise

    def _safe_exec(self, cmd: str):
        try:
            self.bash.sendline(cmd)
            self.bash.expect_exact(cmd)
            self.bash.expect_exact("\r\n" + PS1)
        except Exception as e:
            if self.noexcept:
                self._safe_sendintr()
                self.captured_error = e
            else:
                raise

    def _safe_assert(self, cmd: str):
        try:
            assert_bash_exec(self.bash, cmd, want_output=None)
        except Exception as e:
            if self.noexcept:
                self._safe_sendintr()
                self.captured_error = e
            else:
                raise

    def _copy_variable(self, src_var: str, dst_var: str):
        self._safe_exec(
            "if [[ ${%s+set} ]]; then %s=${%s}; else unset -v %s; fi"
            % (src_var, dst_var, src_var, dst_var),
        )

    def _unset_variable(self, varname: str):
        self._safe_exec("unset -v %s" % varname)

    def _save_cwd(self):
        if not self.cwd_changed:
            self.cwd_changed = True
            self._copy_variable("PWD", "%s_OLDPWD" % self.prefix)

    def _check_set(self, name: str):
        if self.saved_set[name] != bash_env_saved.saved_state.ChangesDetected:
            return
        self._safe_assert(
            '[[ $(shopt -po %s) == "${%s_NEWSHOPT_%s}" ]]'
            % (name, self.prefix, name),
        )

    def _unprotect_set(self, name: str):
        if name not in self.saved_set:
            self.saved_set[name] = bash_env_saved.saved_state.ChangesDetected
            self._safe_exec(
                "%s_OLDSHOPT_%s=$(shopt -po %s || true)"
                % (self.prefix, name, name),
            )
        else:
            self._check_set(name)

    def _protect_set(self, name: str):
        self._safe_exec(
            "%s_NEWSHOPT_%s=$(shopt -po %s || true)"
            % (self.prefix, name, name),
        )

    def _check_shopt(self, name: str):
        if (
            self.saved_shopt[name]
            != bash_env_saved.saved_state.ChangesDetected
        ):
            return
        self._safe_assert(
            '[[ $(shopt -p %s) == "${%s_NEWSHOPT_%s}" ]]'
            % (name, self.prefix, name),
        )

    def _unprotect_shopt(self, name: str):
        if name not in self.saved_shopt:
            self.saved_shopt[name] = bash_env_saved.saved_state.ChangesDetected
            self._safe_exec(
                "%s_OLDSHOPT_%s=$(shopt -p %s || true)"
                % (self.prefix, name, name),
            )
        else:
            self._check_shopt(name)

    def _protect_shopt(self, name: str):
        self._safe_exec(
            "%s_NEWSHOPT_%s=$(shopt -p %s || true)"
            % (self.prefix, name, name),
        )

    def _check_variable(self, varname: str):
        if (
            self.saved_variables[varname]
            != bash_env_saved.saved_state.ChangesDetected
        ):
            return
        try:
            self._safe_assert(
                '[[ ${%s-%s} == "${%s_NEWVAR_%s-%s}" ]]'
                % (varname, MAGIC_MARK2, self.prefix, varname, MAGIC_MARK2),
            )
        except Exception:
            self._copy_variable(
                "%s_NEWVAR_%s" % (self.prefix, varname), varname
            )
            raise
        else:
            if self.noexcept and self.captured_error:
                self._copy_variable(
                    "%s_NEWVAR_%s" % (self.prefix, varname), varname
                )

    def _unprotect_variable(self, varname: str):
        if varname not in self.saved_variables:
            self.saved_variables[varname] = (
                bash_env_saved.saved_state.ChangesDetected
            )
            self._copy_variable(
                varname, "%s_OLDVAR_%s" % (self.prefix, varname)
            )
        else:
            self._check_variable(varname)

    def _protect_variable(self, varname: str):
        self._copy_variable(varname, "%s_NEWVAR_%s" % (self.prefix, varname))

    def _restore_env(self):
        self.noexcept = True

        if self.sendintr:
            self._safe_sendintr()

        # We first go back to the original directory before restoring
        # variables because "cd" affects "OLDPWD".
        if self.cwd_changed:
            self._unprotect_variable("OLDPWD")
            self._safe_exec('command cd -- "$%s_OLDPWD"' % self.prefix)
            self._protect_variable("OLDPWD")
            self._unset_variable("%s_OLDPWD" % self.prefix)
            self.cwd_changed = False

        for varname in self.saved_variables:
            self._check_variable(varname)
            self._copy_variable(
                "%s_OLDVAR_%s" % (self.prefix, varname), varname
            )
            self._unset_variable("%s_OLDVAR_%s" % (self.prefix, varname))
            self._unset_variable("%s_NEWVAR_%s" % (self.prefix, varname))
        self.saved_variables = {}

        for name in self.saved_shopt:
            self._check_shopt(name)
            self._safe_exec('eval "$%s_OLDSHOPT_%s"' % (self.prefix, name))
            self._unset_variable("%s_OLDSHOPT_%s" % (self.prefix, name))
            self._unset_variable("%s_NEWSHOPT_%s" % (self.prefix, name))
        self.saved_shopt = {}

        for name in self.saved_set:
            self._check_set(name)
            self._safe_exec('eval "$%s_OLDSHOPT_%s"' % (self.prefix, name))
            self._unset_variable("%s_OLDSHOPT_%s" % (self.prefix, name))
            self._unset_variable("%s_NEWSHOPT_%s" % (self.prefix, name))
        self.saved_set = {}

        self.noexcept = False
        if self.captured_error:
            raise self.captured_error

    def chdir(self, path: str):
        self._save_cwd()
        self.cwd_changed = True
        self._unprotect_variable("OLDPWD")
        self._safe_exec("command cd -- %s" % shlex.quote(path))
        self._protect_variable("OLDPWD")

    def set(self, name: str, value: bool):
        self._unprotect_set(name)
        if value:
            self._safe_exec("set -o %s" % name)
        else:
            self._safe_exec("set +o %s" % name)
        self._protect_set(name)

    def save_set(self, name: str):
        self._unprotect_set(name)
        self.saved_set[name] = bash_env_saved.saved_state.ChangesIgnored

    def shopt(self, name: str, value: bool):
        self._unprotect_shopt(name)
        if value:
            self._safe_exec("shopt -s %s" % name)
        else:
            self._safe_exec("shopt -u %s" % name)
        self._protect_shopt(name)

    def save_shopt(self, name: str):
        self._unprotect_shopt(name)
        self.saved_shopt[name] = bash_env_saved.saved_state.ChangesIgnored

    def write_variable(self, varname: str, new_value: str, quote: bool = True):
        if quote:
            new_value = shlex.quote(new_value)
        self._unprotect_variable(varname)
        self._safe_exec("%s=%s" % (varname, new_value))
        self._protect_variable(varname)

    def save_variable(self, varname: str):
        self._unprotect_variable(varname)
        self.saved_variables[varname] = (
            bash_env_saved.saved_state.ChangesIgnored
        )

    # TODO: We may restore the "export" attribute as well though it is
    #   not currently tested in "diff_env"
    def write_env(self, envname: str, new_value: str, quote: bool = True):
        if quote:
            new_value = shlex.quote(new_value)
        self._unprotect_variable(envname)
        self._safe_exec("export %s=%s" % (envname, new_value))
        self._protect_variable(envname)


def get_env(bash: pexpect.spawn) -> List[str]:
    return [
        x
        for x in assert_bash_exec(
            bash,
            "_comp__test_get_env",
            want_output=True,
        )
        .strip()
        .splitlines()
        # Sometimes there are empty lines in the output due to unknown
        # reasons, e.g. in GitHub Actions' macos-latest OS. Filter them out.
        if x
    ]


def diff_env(before: List[str], after: List[str], ignore: str):
    diff = [
        x
        for x in difflib.unified_diff(before, after, n=0, lineterm="")
        # Remove unified diff markers:
        if not re.search(r"^(---|\+\+\+|@@ )", x)
        # Ignore variables expected to change:
        and not re.search(
            r"^[-+](_|PPID|BASH_REMATCH|(BASH_)?LINENO)=",
            x,
            re.ASCII,
        )
        # Ignore likely completion functions added by us:
        and not re.search(r"^\+declare -f _.+", x)
        # ...and additional specified things:
        and not re.search(ignore or "^$", x)
    ]
    # For some reason, COMP_WORDBREAKS gets added to the list after
    # saving. Remove its changes, and note that it may take two lines.
    for i in range(0, len(diff)):
        if re.match("^[-+]COMP_WORDBREAKS=", diff[i]):
            if i < len(diff) and not re.match(r"^\+[\w]+=", diff[i + 1]):
                del diff[i + 1]
            del diff[i]
            break
    assert not diff, "Environment should not be modified"


class CompletionResult(Iterable[str]):
    """
    Class to hold completion results.
    """

    def __init__(self, output: Optional[str] = None):
        """
        :param output: All completion output as-is.
        """
        self.output = output or ""

    def endswith(self, suffix: str) -> bool:
        return self.output.endswith(suffix)

    def startswith(self, prefix: str) -> bool:
        return self.output.startswith(prefix)

    def _items(self) -> List[str]:
        return [x.strip() for x in self.output.strip().splitlines()]

    def __eq__(self, expected: object) -> bool:
        """
        Returns True if completion contains expected items, and no others.

        Defining __eq__ this way is quite ugly, but facilitates concise
        testing code.
        """
        if isinstance(expected, str):
            expiter = [expected]  # type: Iterable
        elif not isinstance(expected, Iterable):
            return False
        else:
            expiter = expected
        return self._items() == expiter

    def __contains__(self, item: str) -> bool:
        return item in self._items()

    def __iter__(self) -> Iterator[str]:
        return iter(self._items())

    def __len__(self) -> int:
        return len(self._items())

    def __repr__(self) -> str:
        return "<CompletionResult %s>" % self._items()


def assert_complete(
    bash: pexpect.spawn, cmd: str, **kwargs
) -> CompletionResult:
    skipif = kwargs.get("skipif")
    if skipif:
        try:
            assert_bash_exec(bash, skipif, want_output=None)
        except AssertionError:
            pass
        else:
            pytest.skip(skipif)
    xfail = kwargs.get("xfail")
    if xfail:
        try:
            assert_bash_exec(bash, xfail, want_output=None)
        except AssertionError:
            pass
        else:
            pytest.xfail(xfail)

    with bash_env_saved(bash, sendintr=True) as bash_env:
        cwd = kwargs.get("cwd")
        if cwd:
            bash_env.chdir(str(cwd))

        for k, v in kwargs.get("env", {}).items():
            bash_env.write_env(k, v, quote=False)

        for k, v in kwargs.get("shopt", {}).items():
            bash_env.shopt(k, v)

        input_cmd = cmd
        rendered_cmd = kwargs.get("rendered_cmd", cmd)
        re_MAGIC_MARK = re.escape(MAGIC_MARK)

        trail = kwargs.get("trail")
        if trail:
            # \002 = ^B = cursor left
            input_cmd += trail + "\002" * len(trail)
            rendered_cmd += trail + "\b" * len(trail)

            # After reading the results, something weird happens. For most test
            # setups, as expected (pun intended!), MAGIC_MARK follows as
            # is. But for some others (e.g. CentOS 6, Ubuntu 14 test
            # containers), we get MAGIC_MARK one character a time, followed
            # each time by trail and the corresponding number of \b's. Don't
            # know why, but accept it until/if someone finds out.  Or just be
            # fine with it indefinitely, the visible and practical end result
            # on a terminal is the same anyway.
            maybe_trail = "(%s%s)?" % (re.escape(trail), "\b" * len(trail))
            re_MAGIC_MARK = "".join(
                re.escape(x) + maybe_trail for x in MAGIC_MARK
            )

        bash.send(input_cmd + "\t")
        # Sleep a bit if requested, to avoid `.*` matching too early
        time.sleep(kwargs.get("sleep_after_tab", 0))
        bash.expect_exact(rendered_cmd)
        bash.send(MAGIC_MARK)
        got = bash.expect(
            [
                # 0: multiple lines, result in .before
                r"\r\n" + re.escape(PS1 + rendered_cmd) + ".*" + re_MAGIC_MARK,
                # 1: no completion
                r"^" + re_MAGIC_MARK,
                # 2: on same line, result in .match
                r"^([^\r]+)%s$" % re_MAGIC_MARK,
                # 3: error messages
                r"^([^\r].*)%s$" % re_MAGIC_MARK,
                pexpect.EOF,
                pexpect.TIMEOUT,
            ]
        )
        if got == 0:
            output = re.sub(re_MAGIC_MARK + "$", "", bash.before)
            return CompletionResult(output)
        elif got == 2:
            output = bash.match.group(1)
            return CompletionResult(output)
        elif got == 3:
            output = bash.match.group(1)
            raise AssertionError("Unexpected output: [%s]" % output)
        else:
            # TODO: warn about EOF/TIMEOUT?
            return CompletionResult()


@pytest.fixture
def completion(request, bash: pexpect.spawn) -> CompletionResult:
    marker = request.node.get_closest_marker("complete")
    if not marker:
        return CompletionResult()
    for pre_cmd in marker.kwargs.get("pre_cmds", []):
        assert_bash_exec(bash, pre_cmd, want_output=None)
    cmd = getattr(request.cls, "cmd", None)
    if marker.kwargs.get("require_longopt"):
        # longopt completions require both command presence and that it
        # responds something useful to --help
        if "require_cmd" not in marker.kwargs:
            marker.kwargs["require_cmd"] = True
        if "xfail" not in marker.kwargs:
            marker.kwargs["xfail"] = (
                # --help is required to exit with zero in order to not get a
                # positive for cases where it errors out with a message like
                # "foo: unrecognized option '--help'"
                "! %s --help &>/dev/null || "
                "! %s --help 2>&1 | command grep -qF -- --help"
            ) % ((cmd,) * 2)
    if marker.kwargs.get("require_cmd") and not is_bash_type(bash, cmd):
        pytest.skip("Command not found")

    return assert_complete(bash, marker.args[0], **marker.kwargs)


def in_container() -> bool:
    try:
        container = subprocess.check_output(
            "virt-what || systemd-detect-virt --container",
            stderr=subprocess.DEVNULL,
            shell=True,
        ).strip()
    except subprocess.CalledProcessError:
        container = b""
    if container and container != b"none":
        return True
    if os.path.exists("/.dockerenv"):
        return True
    try:
        with open("/proc/1/environ", "rb") as f:
            # LXC, others?
            if any(
                x.startswith(b"container=") for x in f.readline().split(b"\0")
            ):
                return True
    except OSError:
        pass
    return False


def prepare_fixture_dir(
    tmp_path_factory, files: Iterable[str], dirs: Iterable[str]
) -> Path:
    """
    Fixture to prepare a test dir with dummy contents on the fly.

    Tests that contain filenames differing only by case should use this to
    prepare a dir on the fly rather than including their fixtures in git and
    the tarball. This is to work better with case insensitive file systems.
    """
    tempdir = tmp_path_factory.mktemp("bash-completion.fixture_dir.")

    old_cwd = os.getcwd()
    try:
        os.chdir(tempdir)
        _, _ = create_dummy_filedirs(files, dirs)
    finally:
        os.chdir(old_cwd)

    return tempdir


def create_dummy_filedirs(
    files: Iterable[str], dirs: Iterable[str]
) -> Tuple[List[str], List[str]]:
    """
    Create dummy files and directories on the fly in the current directory.

    Tests that contain filenames differing only by case should use this to
    prepare a dir on the fly rather than including their fixtures in git and
    the tarball. This is to work better with case insensitive file systems.
    """
    new_files = []
    new_dirs = []

    for dir_ in dirs:
        path = Path(dir_)
        if not path.exists():
            path.mkdir()
            new_dirs.append(dir_)
    for file_ in files:
        path = Path(file_)
        if not path.exists():
            path.touch()
            new_files.append(file_)

    return sorted(new_files), sorted(new_dirs)


class TestUnitBase:
    def _test_unit(
        self, func, bash, comp_words, comp_cword, comp_line, comp_point, arg=""
    ):
        assert_bash_exec(
            bash,
            "COMP_WORDS=%s COMP_CWORD=%d COMP_LINE=%s COMP_POINT=%d"
            % (comp_words, comp_cword, shlex.quote(comp_line), comp_point),
        )
        output = assert_bash_exec(bash, func % arg, want_output=True)
        return output.strip()