cygwin-3.1.0 and mintty from desktop shortcut

[Brian Inglis]

On 2020-06-08 04:27, Takashi Yano via Cygwin wrote:
> On Sun, 7 Jun 2020 21:20:09 -0600
> Brian Inglis wrote:
>> On 2020-06-07 03:23, Takashi Yano via Cygwin wrote:
>>> On Sun, 7 Jun 2020 16:42:52 +0900
>>> Takashi Yano via Cygwin <cygwin@cygwin.com> wrote:
>>>> On Sun, 7 Jun 2020 00:15:59 -0600
>>>> Brian Inglis wrote:
>>>>> On 2020-06-06 19:35, Takashi Yano via Cygwin wrote:
>>>>>> On Sat, 06 Jun 2020 13:20:16 +0200
>>>>>> ASSI wrote:
>>>>>>> Takashi Yano via Cygwin writes:
>>>>>>>>> 1. Rename /usr/share/locale to something else, like local.bak.
>>>>>>>>> 2. Start mintty in the usual way.
>>>>>>>>> 3. Rename the directory from step 1 back to /usr/share/local.
>>>>>>>>> 4. Work just like the problem never existed either in the shell running
>>>>>>>>> inside the mintty or even start a new mintty.
>>>>>>>>
>>>>>>>> I guess renaming /usr/share/locale/locale.alias is enough.
>>>>>>>
>>>>>>> I've had some time to look into this problem again after having updated
>>>>>>> Cygwin to the latest and greatest.  Indeed, when
>>>>>>>
>>>>>>> /usr/share/locale/locale.alias
>>>>>>>
>>>>>>> gets renamed, the problem also goes away.  This is great because I don't
>>>>>>> really need the locale aliases for anything.  Btw, my laptop got
>>>>>>> upgraded to Win10 1909 (Enterprise) in the meantime, so the issue isn't
>>>>>>> specific to 1803 as was supected earlier.
>>>>>>>
>>>>>>> I then tried to figure out what exactly causes the problem and it turns
>>>>>>> out that it's the _presence_ of this file with the additional condition
>>>>>>> that it must not be owned by the user starting the mintty/shell.  Since
>>>>>>> I install Cygwin on my work laptop with a different (admin) account and
>>>>>>> not my (non-admin) user account, that explains why I am seeing the
>>>>>>> problem there and not on other machines.  Before you are going to
>>>>>>> suggest that it's the admin vs. non-admin rights: no, if I create a
>>>>>>> locale.alias with my user account (either as an empty file or a copy of
>>>>>>> the backup file), then the admin account is unable to start a shell in
>>>>>>> mintty successfully.  I have no idea why the ownership of a file that
>>>>>>> onnly should get read (and is readable by everyone) would have the
>>>>>>> effect I'm seeing, but maybe that gives the clue on where to look for a
>>>>>>> fix.
>>>>>>
>>>>>> Thanks for the additional information. I tested the following steps
>>>>>> to confirm if the problem can be reproduced.
>>>>>>
>>>>>> 1. Enable Administrator account.
>>>>>> 2. Remove all groups from account yano other than Users.
>>>>>> 3. Install cygwin for all with gettext package as Administrator.
>>>>>> 4. Run mintty from desktop shortcut as Administrator.
>>>>>> 5. Run mintty from desktop shortcut as yano.
>>>>>>
>>>>>> Both 4 and 5 successfully open mintty window with shell.
>>>>>>
>>>>>> I wonder what is the difference between my environment and yours.
>>>>>
>>>>> Locale setting?
>>>>>
>>>>> fhandler_tty.cc(fhandler_pty_slave::setup_locale)@2854 calls
>>>>> get_langinfo@2768 calls@2781
>>>>> nlsfuncs.cc(__set_locale_from_locale_alias)@1462
>>>>> which opens /usr/share/locale/locale.alias@1472.
>>>>>
>>>>> One problem I see with that is that __set_locale_from_locale_alias is meant to
>>>>> be called when loadlocale fails with an unrecognized locale, but in
>>>>> get_langinfo@2778 if the locale is not found, it is defaulted to C before
>>>>> __set_locale_from_locale_alias is called, defeating the purpose:
>>>>>
>>>>>   const char *locale = __loadlocale (&loc, LC_CTYPE, new_locale);
>>>>>   if (!locale)
>>>>>     locale = "C";
>>>>>
>>>>>   char tmp_locale[ENCODING_LEN + 1];
>>>>>   char *ret = __set_locale_from_locale_alias (locale, tmp_locale);
>>>>>   if (ret)
>>>>>     locale = tmp_locale;
>>>>
>>>> Hmm..., you are right. Furthermore, __set_locale_from_locale_alias()
>>>> here is completely unnecessary since it is already processed in
>>>> __loadlocale().
>>>
>>> No. I was wrong. If locale alias is used, __loadlocale() returns
>>> aliased locale. Calling __set_locale_from_locale_alias() is
>>> necessary to resolve alias. For example, if locale is set to
>>> "japanese", which is defined in /usr/share/locale/locale.alias,
>>> __loadlocale() returns "japanese", while
>>> __set_locale_from_locale_alias() returns "ja_JP.eucJP".
>>>
>>> __loadlocale() returns NULL when the alias resolution also fails.
>>> So the current code is as designed.
>>
>> But if __loadlocale() returns a non-NULL string, then the locale and/or alias
>> has been resolved and loaded, so it is unnecessary to further call
>> __set_locale_from_locale_alias().
> 
> This is not true. When __loadlocale() returns non-NULL, the third
> argument of __loadlocale() is returned.
> Please see newlib/libc/locale/locale.c.
> 
> So, as for return value of __loadlocale(), alias is not resolved.
> Alias-resolved locale is used in __loadlocale() only internally.

The alias value is returned, so that the library can avoid re-resolving it and
re-loading locale categories if the same value is passed again: they do not
assume that any returned value will be saved and reused by the most callers in
subsequent uses, so avoid having to re-resolve aliases by keeping them around.

Internally to __loadlocale(), the alias is resolved and the resolved value is
used to set the base locale information and load all the associated locale
categories: I have custom locale/regional settings in Linux and Windows, so I
know these work well.

[It would be clearer if the charset handling was refactored into a separate
function, and also doing the language and territory separately would help
declutter __loadlocale(), but who has the time and knowledge?]

-- 
Take care. Thanks, Brian Inglis, Calgary, Alberta, Canada

This email may be disturbing to some readers as it contains
too much technical detail. Reader discretion is advised.
[Data in IEC units and prefixes, physical quantities in SI.]