如何从 Python 迭代器提供子进程的标准输入？

Question

我正在尝试使用 Python 中的 subprocess 模块与以流方式读取标准输入并写入标准输出的进程进行通信。我想让子进程从产生输入的迭代器中读取行，然后从子进程中读取输出行。输入和输出线之间可能没有一一对应的关系。如何从返回字符串的任意迭代器中提供子进程？

这里有一些示例代码给出了一个简单的测试用例，以及我尝试过的一些方法由于某种原因不起作用：

#!/usr/bin/python
from subprocess import *
# A really big iterator
input_iterator = ("hello %s\n" % x for x in xrange(100000000))

# I thought that stdin could be any iterable, but it actually wants a
# filehandle, so this fails with an error.
subproc = Popen("cat", stdin=input_iterator, stdout=PIPE)

# This works, but it first sends *all* the input at once, then returns
# *all* the output as a string, rather than giving me an iterator over
# the output. This uses up all my memory, because the input is several
# hundred million lines.
subproc = Popen("cat", stdin=PIPE, stdout=PIPE)
output, error = subproc.communicate("".join(input_iterator))
output_lines = output.split("\n")

那么当我从迭代器的标准输出中逐行读取时，如何让我的子进程逐行读取呢？

Answer 1

简单的方法似乎是从子进程中分叉并提供输入句柄。任何人都可以详细说明这样做的任何可能的缺点吗？或者有没有 python 模块让它更容易和更安全？

#!/usr/bin/python
from subprocess import *
import os

def fork_and_input(input, handle):
    """Send input to handle in a child process."""
    # Make sure input is iterable before forking
    input = iter(input)
    if os.fork():
        # Parent
        handle.close()
    else:
        # Child
        try:
            handle.writelines(input)
            handle.close()
        # An IOError here means some *other* part of the program
        # crashed, so don't complain here.
        except IOError:
            pass
        os._exit()

# A really big iterator
input_iterator = ("hello %s\n" % x for x in xrange(100000000))

subproc = Popen("cat", stdin=PIPE, stdout=PIPE)
fork_and_input(input_iterator, subproc.stdin)

for line in subproc.stdout:
    print line,

Answer 2

从 Python 迭代器提供子进程的标准输入：

#!/usr/bin/env python3 
from subprocess import Popen, PIPE

with Popen("sink", stdin=PIPE, bufsize=-1) as process:
    for chunk in input_iterator:
        process.stdin.write(chunk)

如果您想同时读取输出，则需要threads 或 async.io：

#!/usr/bin/env python3
import asyncio
import sys
from asyncio.subprocess import PIPE
from contextlib import closing

async def writelines(writer, lines):
    # NOTE: can't use writer.writelines(lines) here because it tries to write
    # all at once
    with closing(writer):
        for line in lines:
            writer.write(line)
            await writer.drain()

async def main():
    input_iterator = (b"hello %d\n" % x for x in range(100000000))
    process = await asyncio.create_subprocess_exec("cat", stdin=PIPE, stdout=PIPE)
    asyncio.ensure_future(writelines(process.stdin, input_iterator))
    async for line in process.stdout:
        sys.stdout.buffer.write(line)
    return await process.wait()

if sys.platform == 'win32':
    loop = asyncio.ProactorEventLoop()  # for subprocess' pipes on Windows
    asyncio.set_event_loop(loop)
else:
    loop = asyncio.get_event_loop()
with closing(loop):
    sys.exit(loop.run_until_complete(main()))

Answer 3

关注this recipe 这是一个支持异步 I/O 的子进程的附加组件。不过，这仍然需要您的子流程使用其输出的一部分来响应每个输入行或行组。