bug#35671: 'glibc-utf8-locales' should include C.UTF-8 locale

[Meiyo Peng]

Hi Ludovic,

Ludovic Courtès writes:

> But anyway, I think we should just set “LC_ALL=en_US.utf8” in the
> .service file: that’s what we do on Guix System, and the choice of a
> locale doesn’t matter since ‘guix substitute’ honors the client’s
> locale.

That's what I did to my guix-daemon.service and guix-publish.service
files on a foreign distro after getting locale warnings.

> Meiyo Peng <address@hidden> skribis:
>
>> Ludovic Courtès writes:
>
> [...]
>
>>> Thus, I think ‘glibc-utf8-locales’ should provide that locale.
>>
>> And the zh_CN.UTF-8, zh_TW.UTF-8 please.
>
> ‘glibc-utf8-locales’ has always been described as an *arbitrary* sample
> of UTF-8 locales (info "(guix) Application Setup").  Initially its main
> purpose was to use it in build processes, but we’ve come to more or less
> recommend it for users, which is not great because we know it only works
> for some users.

I was surprised last week (maybe the week before that) when I found out
that glibc-utf8-locales only provides locales for de_DE, el_GR, en_US,
fr_FR, tr_TR.  And in our manual, we tell people to install
glibc-utf8-locales after installing Guix on a foreign distro in order to
solve locale problems.  The name "glibc-utf8-locales" indicates it
provides ALL UTF-8 locales while in fact it does not.  This caused me a
trouble because I have to set the locale of Emacs to zh_CN.UTF-8 in
order to enable the ibus input method in Emacs.  (This is a known
problem of Emacs.  Locale has to be set to one of CJK locale in order to
enable external input methods.)  Since the zh_CN.UTF-8 locale is not in
$GUIX_LOCPATH, ibus does not work in Emacs.  I ended up installing the
"glibc-locales" package and solved the problem.

> I built a ‘glibc-utf8-locales’ package that provides all the supported
> UTF-8 locales, and only UTF-8 locales:
>
> --8<---------------cut here---------------start------------->8---
> $ guix size 
> /gnu/store/rh7iq1kwma7ir96mvzvqcg1v50yi05yp-glibc-utf8-locales-2.28
> store item                                                       total    self
> /gnu/store/rh7iq1kwma7ir96mvzvqcg1v50yi05yp-glibc-utf8-locales-2.28   855.7   
> 855.7 100.0%
> total: 855.7 MiB
> $ ls -1d 
> /gnu/store/rh7iq1kwma7ir96mvzvqcg1v50yi05yp-glibc-utf8-locales-2.28/lib/locale/2.28/*.utf8
>  |wc -l
> 312
> $ guix size glibc-locales
> store item                                                       total    self
> /gnu/store/acl2wxzzkkcjv74rlqswdf9p8pwddlmk-glibc-locales-2.28     916.7   
> 916.7 100.0%
> total: 916.7 MiB
> --8<---------------cut here---------------end--------------->8---
>
> As you can see, it’s almost the same size as ‘glibc-locales’, so there’s
> no point in having such a ‘glibc-utf8-locales’ package.
>
> In Guix System, the (gnu system locale) module generates just the needed
> locales.  I’m thinking we should do the same at the ‘guix package’
> level.  For example, ‘guix package --install-locales’ would
> automatically install whatever $LANG or $LC_* refers to.
>
> Thoughts?

I think it would be better to simply tell people to install the
"glibc-locales" package.  There may be a case where different
applications are set to different locales.  For example, my system is
set to the en_US.UTF-8 locale but my Emacs is set to the zh_CN.UTF-8
locale.  So $LC_* only refers to en_US.UTF-8 if guix tries to detect the
locale.  If ‘guix package --install-locales’ only install what $LANG or
$LC_* refers to, zh_CN.UTF-8 won't be installed for me.

We are unable to detect what locales a user will need later.  Installing
a full package like "glibc-locales" or letting the user explicitly
specify what locales they need can avoid a lot of special problems.


--
Meiyo Peng
https://www.pengmeiyu.com/