print_wcwidth.py

# -*- coding: utf-8 -*-
# Copyright (c) 2019 Ansible Project
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt)

# Make coding more python3-ish
from __future__ import (absolute_import, division, print_function)
__metaclass__ = type

import ctypes.util
import locale

from six import text_type


libc_path = ctypes.util.find_library('c')
libc = ctypes.cdll.LoadLibrary(libc_path)
libc.wcwidth.argtypes = (ctypes.c_wchar,)
libc.wcwidth.restype = ctypes.c_int
libc.wcswidth.argtypes = (ctypes.c_wchar_p, ctypes.c_int)
libc.wcswidth.restype = ctypes.c_int


locale.setlocale(locale.LC_ALL, '')


def width(u_text):
    """This function is slower than just using libc directly.
    I recommend not using this and just using ``libc.wcswidth``
    on a full string
    
    A helper may still be useful, to do the isinstance check still
    """
    if not isinstance(u_text, text_type):
        raise ValueError('Value must be text type')
    length = 0
    for c in u_text:
        width = libc.wcwidth(c)
        if width < 0:
            raise ValueError('Something bad happened')
        length += width
    return length


if __name__ == '__main__':
    print(libc.wcswidth(u'コンニチハ', 1024))

    l = width(u'コンニチハ')
    print(u'コンニチハ')
    print(l * '-')

Wild ass guess is that the reason this doesn't work on python2 is that python isn't translating the unicode string into wchar_t correctly. Maybe comparing ctypes code for python-3.0 vs python-2.7 will show if that's correct.

This gets it working on Python2:

--- print_wcwidth.py.orig       2019-10-03 15:56:34.498229625 -0700
+++ print_wcwidth.py    2019-10-03 15:57:17.929191395 -0700
@@ -9,6 +9,9 @@
 from six import text_type
 
 import ctypes.util
+import locale
+
+locale.setlocale(locale.LC_ALL, ('en_US', 'UTF-8'))
 libc_path = ctypes.util.find_library('c')
 libc = ctypes.cdll.LoadLibrary(libc_path)
 libc.wcwidth.argtypes = (ctypes.c_wchar,)

Unfortunately we can't just go setting locale in our code. But perhaps it gives us a hint as to how we can fix it.

Okay, I think this works and we can use it:

`locale.setlocale(locale.LC_ALL, '')

Perf comparison between the custom width function, and just using libc.wcswidth directly. wcswidth is much faster, which is to be expected.

ansibledev ▶ In [2]: %timeit print_width.width(u'コンニチハ')
3.91 µs ± 184 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

ansibledev ▶ In [3]: %timeit print_width.libc.wcswidth(u'コンニチハ', 1024)
963 ns ± 95.2 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

From the wcswidth man page: "If a nonprintable wide character occurs among these characters, -1 is returned." That might be part of the difference with kitchen.text.display.textual_width. Maybe we should have a width() function as a front end but it first tries to run wcswidth() on the string, then, if -1 is returned, have some slower code that steps through each character to determine what the width is.