Migrate to modern win32/winrt API for `localemodule`

[mokurin000]

Feature or enhancement

Proposal:

AFAIK, CPython has dropped support for Windows 7 since 3.9.0, while _localemodule.c contains compatibility fix for Windows NT pre-4.0. Which flag was introduced since Windows 98.

Here I listed some modern way to get locale name in BCP-47 / RFC 1766 format, is it possible to replace the legacy hex code mapping?

• GlobalizationPreferences since Windows 10 (build 10240)
• GetUserDefaultLocaleName since Windows Vista / Windows Server 2008

import ctypes
from ctypes import wintypes
from locale import getdefaultlocale, getlocale, LC_CTYPE
import warnings

# Introduced since Windows 10, BCP-47
# depends on:
# - winrt-windows-system-userprofile
# - winrt-windows-foundation
# - winrt-windows-foundation-collections
from winrt.windows.system.userprofile import GlobalizationPreferences

warnings.filterwarnings("ignore", category=DeprecationWarning)

# Load kernel32.dll
kernel32 = ctypes.WinDLL("kernel32", use_last_error=True)

# Introduced since Windows Vista, RFC 1766
# Define GetUserDefaultLocaleName signature
# int GetUserDefaultLocaleName(LPWSTR lpLocaleName, int cchLocaleName);
kernel32.GetUserDefaultLocaleName.argtypes = [wintypes.LPWSTR, ctypes.c_int]
kernel32.GetUserDefaultLocaleName.restype = ctypes.c_int

# Allocate buffer for locale name
LOCALE_NAME_MAX_LENGTH = 85  # per Windows docs


def user_default_locale() -> str:
    """
    Get user default locale encoded in BCP-47.
    """
    buffer = ctypes.create_unicode_buffer(LOCALE_NAME_MAX_LENGTH)

    # Call the function
    result = kernel32.GetUserDefaultLocaleName(buffer, LOCALE_NAME_MAX_LENGTH)

    if result == 0:
        # Call failed
        error_code = ctypes.get_last_error()
        raise ctypes.WinError(error_code)

    return buffer.value


def bcp47():
    return GlobalizationPreferences.languages[0]

>>> getdefaultlocale()
('zh_CN', 'cp936')
>>> getlocale(LC_CTYPE)
('Chinese (Simplified)_China', '936')
>>> user_default_locale()
'zh-CN'
>>> bcp47()
'zh-Hans-CN'

Has this already been discussed elsewhere?

Yes

Links to previous discussion of this feature:

gh-90817 gh-130796 #123853 (comment)

[mokurin000]

Update: Due to misunderstanding Microsoft style of documentation, I thought the LOCALE_SISO639LANGNAME was removed in Vista, but the Version noted as bold actually means the version the flag was first introduced.

I did write a quick test in rust:

use windows::Win32::Globalization::{
    GetLocaleInfoA, LOCALE_SISO639LANGNAME, LOCALE_SISO3166CTRYNAME, LOCALE_USER_DEFAULT,
};

fn helper(lctype: u32) -> String {
    let mut buf = vec![0u8; 85];
    let return_code = unsafe { GetLocaleInfoA(LOCALE_USER_DEFAULT, lctype, Some(&mut buf)) };
    buf.truncate(return_code as _);

    String::from_utf8_lossy(&buf).into_owned()
}

fn main() {
    // outputs like `zh_CN`
    println!(
        "{}_{}",
        helper(LOCALE_SISO639LANGNAME),
        helper(LOCALE_SISO3166CTRYNAME)
    );
}

So the current code can be expected to work perfectly, but I doubt why there was bug gh-123853