Re: newlocale: Linux incompatibility

[Brian Inglis <Brian.Inglis@Shaw.ca>]

On 2023-03-24 06:18, Corinna Vinschen via Cygwin wrote:
> On Mar 23 22:14, Corinna Vinschen via Cygwin wrote:
>> On Mar 23 15:48, Ken Brown via Cygwin wrote:
>>> I'm reporting this here rather than the newlib list because the behavior is
>>> compatible with Posix but not Linux, so I think it's a Cygwin issue.
>>
>> Actually, it's a Windows issue :)
>>
>>> Consider the following test case:
>>>
>>> $ cat locale_test.c
>>> #include <stdio.h>
>>> #include <locale.h>
>>>
>>> int main ()
>>> {
>>>    const char *locale = "en_DE.UTF-8";
>>>    locale_t loc = newlocale (LC_COLLATE_MASK | LC_CTYPE_MASK, locale, 0);
>>>    if (!loc)
>>>      perror ("newlocale");
>>>    else
>>>      printf ("newlocale succeeded on invalid locale %s\n", locale);
>>> }
>>>
>>> $ gcc -o locale_test locale_test.c
>>>
>>> $ ./locale_test.exe
>>> newlocale succeeded on invalid locale en_DE.UTF-8
>>>
>>> On Linux, the newlocale call fails with ENOENT, as is documented on the man
>>> page.  Posix doesn't say what should happen on an invalid locale, so this is
>>> not, strictly speaking, a bug.
>>
>> Three bugs in fact.
>>
>> First, it's a bug in the Emacs testsuite.  The test simply assumes that
>> there's no en_DE locale on any system, but that's just not true.
>> Windows support the RFC 5646 locale "en-DE", which is called "English
>> (Germany)" in the "Region" settings.
>>
>> You can also check with `locale -av | less' and search for en_DE.
>>
>> For the reminder of this mail, I assume you're talking about Cygwin 3.5.
>> I won't fix this for 3.4 anymore, given how much locale handling has
>> changed for 3.5.
>>
>> The second bug is that Cygwin blindly trusts the Windows function
>> ResolveLocaleName().  That function blatantly converts even vaguely
>> similar locales into something it supports.  E.g., it converts "en-XY"
>> to "en-US".  I. .e., even if you use "en_XY.utf8" as locale, the above
>> testcase will wrongly succeed.  So I have to rethink how I resolve POSIX
>> locales to Windows locales.

Does Windows even consider https://www.rfc-editor.org/rfc/rfc4647 "Matching of 
Language Tags", part of https://www.rfc-editor.org/info/bcp47 "Language Tags", 
and if POSIX only matches exactly, will LANGUAGE be able to be used for fallback?

I currently define LANGUAGE=en_CA:en_GB:en in case en-CA is unsupported by 
anything.
[I use my own en-CA locale not the glibc default created by https://rap.dk/.]

Will "-" be supported like "_" as a separator in values?

>> And the third bug is that Cygwin fails to set errno if it doesn't
>> support a locale, but that's a minor inconvenience in comparison.
>>
>> Thanks for the report, I totally missed the above problem with
>> ResolveLocaleName.
> 
> I pushed a couple of patches which hopefully clean up the code.  It's
> really frustrating how these Windows locale functions work.  Or, rather,
> not work.  I mean, come on...
> 
> - ResolveLocaleName() resolves "ff-BF" to "ff-Latn-SN", not to
>    "ff-Adlm-BF" or "ff-Latn-BF", even though both exist.
> 
> - There's a locale called "sd-Arab-PK" and a locale "sd-Deva-IN".  If
>    you ask for the script used in "sd-IN", the result is "Arab", not
>    "Deva".
 >
> I had to create a replacement function for ResolveLocaleName which
> doesn't return totally screwy and unexpected results, and special case
> two more locales in /proc/locales output so the output makes sense.

Aha - a nice new 3.5.0 feature - as well as /proc/codesets - is that charsets 
e.g. ISO-10646, etc. rather than encodings e.g. UTF-8, etc.!

FYI Google fixed their English L14N falling back to en-GB except US territories:

https://developer.android.com/guide/topics/resources/multilingual-support#postN
https://issuetracker.google.com/issues/64429534#comment6

and there have been similar issues posted for other languages.

> Oh, and I added error handling to the code so newlocale is now able to
> set errno to ENOENT if the locale is not supported.
> 
> If you want to test this, the changes are in test release
> 3.5.0-0.260.gb5b67a65f87c, which is just building.
-- 
Take care. Thanks, Brian Inglis              Calgary, Alberta, Canada

La perfection est atteinte                   Perfection is achieved
non pas lorsqu'il n'y a plus rien à ajouter  not when there is no more to add
mais lorsqu'il n'y a plus rien à retirer     but when there is no more to cut
                                 -- Antoine de Saint-Exupéry