pre-commit/pre_commit/xargs.py

from __future__ import annotations

import concurrent.futures
import contextlib
import math
import multiprocessing
import os
import subprocess
import sys
from collections.abc import Callable
from collections.abc import Generator
from collections.abc import Iterable
from collections.abc import MutableMapping
from collections.abc import Sequence
from typing import Any
from typing import TypeVar

from pre_commit import parse_shebang
from pre_commit.util import cmd_output_b
from pre_commit.util import cmd_output_p

TArg = TypeVar('TArg')
TRet = TypeVar('TRet')


def cpu_count() -> int:
    try:
        # On systems that support it, this will return a more accurate count of
        # usable CPUs for the current process, which will take into account
        # cgroup limits
        return len(os.sched_getaffinity(0))
    except AttributeError:
        pass

    try:
        return multiprocessing.cpu_count()
    except NotImplementedError:
        return 1


def _environ_size(_env: MutableMapping[str, str] | None = None) -> int:
    environ = _env if _env is not None else getattr(os, 'environb', os.environ)
    size = 8 * len(environ)  # number of pointers in `envp`
    for k, v in environ.items():
        size += len(k) + len(v) + 2  # c strings in `envp`
    return size


def _get_platform_max_length() -> int:  # pragma: no cover (platform specific)
    if os.name == 'posix':
        maximum = os.sysconf('SC_ARG_MAX') - 2048 - _environ_size()
        maximum = max(min(maximum, 2 ** 17), 2 ** 12)
        return maximum
    elif os.name == 'nt':
        return 2 ** 15 - 2048  # UNICODE_STRING max - headroom
    else:
        # posix minimum
        return 2 ** 12


def _command_length(*cmd: str) -> int:
    full_cmd = ' '.join(cmd)

    # win32 uses the amount of characters, more details at:
    # https://github.com/pre-commit/pre-commit/pull/839
    if sys.platform == 'win32':
        return len(full_cmd.encode('utf-16le')) // 2
    else:
        return len(full_cmd.encode(sys.getfilesystemencoding()))


class ArgumentTooLongError(RuntimeError):
    pass


def partition(
        cmd: Sequence[str],
        varargs: Sequence[str],
        target_concurrency: int,
        _max_length: int | None = None,
) -> tuple[tuple[str, ...], ...]:
    _max_length = _max_length or _get_platform_max_length()

    # Generally, we try to partition evenly into at least `target_concurrency`
    # partitions, but we don't want a bunch of tiny partitions.
    max_args = max(4, math.ceil(len(varargs) / target_concurrency))

    cmd = tuple(cmd)
    ret = []

    ret_cmd: list[str] = []
    # Reversed so arguments are in order
    varargs = list(reversed(varargs))

    total_length = _command_length(*cmd) + 1
    while varargs:
        arg = varargs.pop()

        arg_length = _command_length(arg) + 1
        if (
                total_length + arg_length <= _max_length and
                len(ret_cmd) < max_args
        ):
            ret_cmd.append(arg)
            total_length += arg_length
        elif not ret_cmd:
            raise ArgumentTooLongError(arg)
        else:
            # We've exceeded the length, yield a command
            ret.append(cmd + tuple(ret_cmd))
            ret_cmd = []
            total_length = _command_length(*cmd) + 1
            varargs.append(arg)

    ret.append(cmd + tuple(ret_cmd))

    return tuple(ret)


@contextlib.contextmanager
def _thread_mapper(maxsize: int) -> Generator[
    Callable[[Callable[[TArg], TRet], Iterable[TArg]], Iterable[TRet]],
]:
    if maxsize == 1:
        yield map
    else:
        with concurrent.futures.ThreadPoolExecutor(maxsize) as ex:
            yield ex.map


def xargs(
        cmd: tuple[str, ...],
        varargs: Sequence[str],
        *,
        color: bool = False,
        target_concurrency: int = 1,
        _max_length: int = _get_platform_max_length(),
        **kwargs: Any,
) -> tuple[int, bytes]:
    """A simplified implementation of xargs.

    color: Make a pty if on a platform that supports it
    target_concurrency: Target number of partitions to run concurrently
    """
    use_pty = color and kwargs.get('input') is None
    cmd_fn = cmd_output_p if use_pty else cmd_output_b
    retcode = 0
    stdout = b''

    try:
        cmd = parse_shebang.normalize_cmd(cmd)
    except parse_shebang.ExecutableNotFoundError as e:
        return e.to_output()[:2]

    # on windows, batch files have a separate length limit than windows itself
    if (
            sys.platform == 'win32' and
            cmd[0].lower().endswith(('.bat', '.cmd'))
    ):  # pragma: win32 cover
        # this is implementation details but the command gets translated into
        # full/path/to/cmd.exe /c *cmd
        cmd_exe = parse_shebang.find_executable('cmd.exe')
        # 1024 is additionally subtracted to give headroom for further
        # expansion inside the batch file
        _max_length = 8192 - len(cmd_exe) - len(' /c ') - 1024

    partitions = partition(cmd, varargs, target_concurrency, _max_length)
    if kwargs.get('input') is not None and len(partitions) != 1:
        raise AssertionError('`input` is only supported with one partition')

    def run_cmd_partition(
            run_cmd: tuple[str, ...],
    ) -> tuple[int, bytes, bytes | None]:
        return cmd_fn(
            *run_cmd, check=False, stderr=subprocess.STDOUT, **kwargs,
        )

    threads = min(len(partitions), target_concurrency)
    with _thread_mapper(threads) as thread_map:
        results = thread_map(run_cmd_partition, partitions)

        for proc_retcode, proc_out, _ in results:
            if abs(proc_retcode) > abs(retcode):
                retcode = proc_retcode
            stdout += proc_out

    return retcode, stdout