Title: Windows also treats full-width spaces as a delimiter when parsing arguments
Type: behavior Stage: resolved
Components: Library (Lib), Windows Versions: Python 3.7
Status: closed Resolution: not a bug
Dependencies: Superseder:
Assigned To: Nosy List: LCY, eryksun, paul.moore, steve.dower, tim.golden, zach.ware
Priority: normal Keywords:

Created on 2017-04-07 11:28 by LCY, last changed 2017-04-07 17:36 by LCY. This issue is now closed.

Pull Requests
URL Status Linked Edit
PR 1029 closed python-dev, 2017-04-07 11:37
Messages (3)
msg291262 - (view) Author: LCY (LCY) * Date: 2017-04-07 11:28
Windows also treats full-width spaces as a delimiter when parsing command line arguments.

Therefore, and subprocess.Popen() also need to quote the arg in the sequence of arguments if there is any full-width spaces in it.


>>['foo', 'half-width space', 'full-width space'])

should be executed as 
>> foo "half-width space" "full-width space"
Windows will treat it as 3 arguments

but now it is incorrectly executed as 
>> foo "half-width space" full-width space
Windows will treat it as 4 arguments
msg291277 - (view) Author: Eryk Sun (eryksun) * (Python triager) Date: 2017-04-07 16:56
> Windows also treats full-width spaces as a delimiter when parsing 
> command line arguments.

CreateProcess has to parse the beginning of the command-line string if the lpApplicationName parameter is omitted. According to the documentation, it treats "white space" as a delimiter, but it doesn't actually say which characters are in that set. We know for an unquoted name like "python{character}spam" that it will try to execute python.exe if "{character}" is parsed as a space. Otherwise we expect CreateProcess to fail with ERROR_FILE_NOT_FOUND because it's looking for the non-existent file "python{character}spam". Here's a test that checks all characters that Unicode considers to be whitespace, which includes "ideographic space" (U+3000):

    import os
    import sys
    import subprocess

    space_chars = [chr(c) for c in range(sys.maxunicode) if chr(c).isspace()]
    assert '\N{IDEOGRAPHIC SPACE}' in space_chars # U+3000

    def get_create_delims():
        assert not os.path.exists('spam')
        filename = 'python{}spam'
        basepath = os.path.dirname(sys.executable)
        delims = []
        for space in space_chars:
            path = os.path.join(basepath, filename.format(space))
            assert not os.path.exists(path)
                subprocess.check_output(path, stderr=subprocess.STDOUT)
            except FileNotFoundError:
                pass # not a delimiter
            except subprocess.CalledProcessError:
                assert False, 'python.exe should have failed'
        return delims

    >>> get_create_delims()
    ['\t', ' ']

CreateProcess considers only space and horizontal tab as white-space delimiters, at least on this Windows 10 system.

Otherwise Windows itself doesn't care about the command line. It's up to each application to parse its command line however it wants. subprocess.list2cmdline assumes an application uses argv from Microsoft's C runtime. The Windows shell function CommandLineToArgvW is supposed to follow the same rules. The following calls CommandLineToArgvW on a test command-line string for each character in the space_chars set:

    import ctypes
    from ctypes import wintypes

    shell32 = ctypes.WinDLL('shell32', use_last_error=True) 

    PLPWSTR = ctypes.POINTER(wintypes.LPWSTR)
    shell32.CommandLineToArgvW.restype = PLPWSTR

    def cmdline2argv(cmdline):
        argc = ctypes.c_int()
        pargv = shell32.CommandLineToArgvW(cmdline, ctypes.byref(argc))
        if not pargv:
            raise ctypes.WinError(ctypes.get_last_error())
        return pargv[:argc.value]

    def get_argv_delims():
        cmdline = 'test{}space'
        delims = []
        for space in space_chars:
            if len(cmdline2argv(cmdline.format(space))) > 1:
        return delims

    >>> get_argv_delims()
    ['\t', '\n', '\x0b', '\x0c', '\r', '\x1c', '\x1d', '\x1e', '\x1f', ' ']

In addition to space and horizontal tab, CommandLineToArgvW also considers line feed, vertical tab, form feed, carriage return, file separator, group separator, record separator, and unit separator to be white-space delimiters. This disagrees with [1], which says it should be limited to space and horizontal tab, like CreateProcess. Let's test this as well:

    def get_msvc_argv_delims():
        template = '"{}" -c "import sys;print(len(sys.argv))" test{}space'
        delims = []
        for space in space_chars:
            cmdline = template.format(sys.executable, space)
            out = subprocess.check_output(cmdline)
            argc = int(out)
            if argc > 2:
        return delims

    >>> get_msvc_argv_delims()
    ['\t', ' ']

Apparently CommandLineToArgvW is inconsistent with the C runtime in this case.

On my Windows 10 system, ideographic space (U+3000) is not generally a command-line delimiter. That's not to say that some applications (and maybe localized CRTs?) don't use it that way. But I don't think it's the place of the subprocess module to handle it.

msg291281 - (view) Author: LCY (LCY) * Date: 2017-04-07 17:36
Thank you for testing and explanation.

My test cases are not enough.

I will close this issue.
Date User Action Args
2017-04-07 17:36:32LCYsetstatus: open -> closed
resolution: not a bug
messages: + msg291281

stage: resolved
2017-04-07 16:56:53eryksunsetnosy: + eryksun
messages: + msg291277
2017-04-07 11:37:00python-devsetpull_requests: + pull_request1186
2017-04-07 11:28:03LCYcreate