Re: tls.c function glthread_tls_get is void not void * as in tls.h

[Bruno Haible]

Hello,

Gerard Tromp wrote:
> I was building gettext-0.16 on a Sun Sparc Solaris 2.8 sun4u box and found
> that the build croaked on tls.c. The gcc compiler gave an error indicating
> that there was a mismatch between the declaration of glthread_tls_get in
> tls.h and the code in tls.c. Changing tls.c to have the function as "void *
> glthread_tls_get ..." (patch file attached)

Thanks for the patch. This fix will be contained in gettext 0.16.2.

> Failed test output.
> recode-sr-latin: input is not valid in "ASCII" encoding: Illegal byte
> sequence
> FAIL: recode-sr-latin-2

The test will be fixed in the next gettext version: The test must take
care to run itself in a Japanese locale.

> Building for 64-bit mode did not work as well. 3 tests failed with no
> explanation.
> msg-fmt-11
> msg-fmt-13
> xgettext-java-2

This looks more serious.

> Could someone enlighten me as to what code is exercised and tested in these
> tests?

The common point of the three tests is that they exercise the
UTF-8 <--> UCS-4 conversion. I don't remember if it is via iconv or directly
(via ucs4-utf8.h, utf16-ucs4.h, utf8-ucs4.h).

Anyway, there is some possibility that the problem has already been fixed
in the current gettext CVS. You find that at
  http://savannah.gnu.org/cvs/?group=gettext
After checking that out, take a look at the autogen.sh script.

Bruno