On 3/25/19 2:32 PM, Juri Linkov wrote:
> I don't know where to put i18n related code, so since ngettext should
> have C calls anyway, I moved it to editfns.c near the function ‘message’
> where it still just returns the correct plurals without doing any translation.

That stub had some problems:

1. It lacked documentation in the Elisp manual. Important changes like
this should be documented -- to some extent the documentation is even
more important than the code. Can you write something?

2. While you're thinking about (1) here are some other questions. How
will ngettext determine the message catalog? Is the catalog visible to
users as a global variable, or as a hidden part of the global state, or
is it something explicit? How will catalogs from multiple packages be
used? How would a multi-lingual application work in Emacs if the message
catalog is part of global state? This seems to be a crucial issue, I'd
say. For example, should Emacs export dcngettext to Lisp code, instead
of just plain ngettext? (Emacs could then define ngettext in terms of
dcngettext.)

3. User C code is not supposed to inspect the _LIBC macro; that's for
glibc internal use. In Emacs _LIBC should be used only with code shared
with glibc, and we should assume _LIBC is never defined when files are
compiled for Emacs.

4. The stub doesn't work with bignums.

5. When calling the C-level ngettext, strings are not properly recoded.

I fixed (3) and (4), and temporarily worked around (5), by installing
the attached patch. To do a better job with (2) and (5) please see the
gettext manual's instructions for package maintainers, here:

https://www.gnu.org/software/gettext/manual/gettext.html#Maintainers

To my mind (1) and (2) are the most-pressing problems.