Re: wint_t in cross building, gnulib failing build.

[Bruno Haible]

Hi,

Lluís Batlle wrote:
> I'm trying to cross-build gettext, and with my toolchain (gcc 4.3.4
> and glibc-2.9 for armv5tel-linux) I have problems, while I don't have
> any problem building it not natively.
>
> I found that the gettext-runtime configure has a check like this:
> if test "$cross_compiling" = yes; then
> gl_cv_func_wcwidth_works="guessing no"

This has been improved in gnulib on 2009-07-14: it will now guess "yes"
on glibc systems and "no" elsewhere.

> In native builds (where I use the same compiler and glibc version) I
> have gl_cv_func_wcwidth_works=yes, and having this defined as 'no'
> results in gettext-runtime/gnulib-lib/ use its own 'wchar.h' created
> from 'wchar.h.in'

This should be redundant but harmless.

> and this wchar.h ends up not having wint_t defined
> as required in glibc's wchar.h, included through mbchar.h in
> gnulib-lib.

This has been fixed in gnulib on 2009-01-15 and 2009-01-17.

You might therefore want to try the newest gettext from git, combined
with the newest gnulib from git. Be sure to look at the gettext/HACKING file.

> I'm going to go on using a cache-file, and setting
> gl_cv_func_wcwidth_works=yes there

Yes, this should work. You can also pass gl_cv_func_wcwidth_works=yes as
an environment variable setting while running configure.

> but I would like to know why this
> mbchar build fails for me having no wint_t type defined. Why
> cross-compiling the wcwidth function would not work?

Simply, it had not been tested before the release of gettext 0.17.

Bruno

[1] https://savannah.gnu.org/git/?group=gettext
[2] https://savannah.gnu.org/git/?group=gnulib

[Lluís Batlle]

Hello,

thank you very much for your answer.

2009/11/25 Bruno Haible <br...@clisp.org>:

> You might therefore want to try the newest gettext from git, combined
> with the newest gnulib from git. Be sure to look at the gettext/HACKING file.

I think I will wait for the next release, because by now I have a
solution to get all working. I only wanted to understand a bit the
logic behind the 'configure' decisions I was seeing, in order to learn
from something I was not understanding.
But now I'm happy that I was understanding, and there are fixes in
newer versions of the code.

Regards,
Lluís.