here the patch is: http://paste.ubuntu.com/23953329/
The code tested, but maybe it sucks.

1. striping @xx, _TT: when mapping with full name failed,  we check if
there's a '@' in locname. if so, go back to the part of copying catname,
override and skip '@xx'.

Then we check if there's a '_', and if both '@' and '_TT' is there, point
locname to '@xx', set a correct loclen, go back to the part of writing
locname to replace '_TT' with '@xx'. If not both, skip and simply override
'_TT'.

Because there's also '_' in 'LC_xx', we may get into a dead loop of
stripping '_TT'. So locname is checked, it's set to NULL if we used strchr
to skip once.

Same reason, we may get into a dead loop of overriding '_TT'. The first
position of '/' should be front of the '_' if we replaced it once, the name
will like: 'zh@t/LC_xx'.

zh_CN@t (stripped by the first part)-> zh_CN (overrided by the second
part)-> zh@t  (stripped by the first part again)-> zh

2. about rewriting of '.GBK': I agreeded with keeping the original value
 of user, and stripping it in gettext() before. But i thought that someone
may validate if libc set the correct charset by setlocale(). So we should
rewrite .XX to .UTF-8 in setlocale(), we cant return a wrong value in
principle.

2017-01-30 0:33 GMT+08:00 Rich Felker <dalias@libc.org>:

> On Mon, Jan 30, 2017 at 12:14:49AM +0800, He X wrote:
> > I can't wait, can i work on it and make a patch for these issues if
> Masanori
> > Ogino is busy now? I'd like to see that these issues could be solved in
> > official musl repo as soon as possible.
>
> I'm not saying you need to wait, just that you should be aware of past
> discussion of the topic, and if you want to propose patches they
> should either follow the behavior outlined before or come with
> discussion of why you think a different behavior is more appropriate.
>
> > And maybe rejection for NON-UTF-8, since 'LANG=zh_CN.GBK ./a.out(
> > setlocale(LC_*, "") )' showed me a segfault with glibc.
>
> I don't think "it crashes on glibc" is a good justification for
> anything. Rather there should probably be UX discussions of what
> different choices mean for different poor-configuration situations
> that are likely to arise in the wild (from things like LC_* getting
> copied over ssh).
>
> Rich
>