On Saturday, April 19, 2008 at 21:49:42 +0200, Bruno Haible wrote:
> Alain Guibert wrote:
>>>>| FAIL: test-c-strcase.sh
>> LC_ALL=fr_FR setlocale(LC_ALL, "") returns NULL
> Weird. It looks as if some symbols from libintl (or its dependencies:
> libiconv) would be affecting/overriding the behaviour of libc. Can you
> look at which symbols are defined in libintl.so and libiconv.so:
Nothing special in libiconv 1.12. But there are 4 collisions between
libintl.so.8.0.2 and libc.so.5.4.33:
| 00031088 T _nl_expand_alias
| 000302c8 T _nl_explode_name
| 00030514 T _nl_make_l10nflist
| 00030a04 T _nl_normalize_codeset
Only -lintl -lc fails; -liconv -lc works. This NULL setlocale() problem
is present with libintl.so.8 from gettext 0.15, 0.16, and 0.17; absent
with libintl.so.2 or 3 from gettext 0.12.1, 0.14.1 and 0.14.4. The small
test program (only printf() and setlocale(), no other functions)
normally working compiled with -lc, extract from strace:
| open("/usr/share/locale/locale.alias", O_RDONLY) = -1 ENOENT (No such file or
directory)
| open("/usr/share/i18n/locale.alias", O_RDONLY) = -1 ENOENT (No such file or
directory)
| open("/usr/share/locale/fr_FR/LC_MESSAGES", O_RDONLY) = 3
Same program compiled with -lintl -lc, but working, with gettext 0.12.1:
| open("/usr/local/share/locale/locale.alias", O_RDONLY) = -1 ENOENT (No such
file or directory)
| open("/usr/share/locale/fr_FR/LC_MESSAGES", O_RDONLY) = 3
Note setlocale() was already influenced by -lintl first, as the alias
file was then searched in /usr/local instead of /usr... Now same program
failing with -lintl -lc gettext 0.17:
| open("/usr/local/share/locale/locale.alias", O_RDONLY) = -1 ENOENT (No such
file or directory)
| open("/usr/share/locale/fr_FR/|", O_RDONLY) = -1 ENOENT (No such file or
directory)
| open("/usr/share/locale/fr/|", O_RDONLY) = -1 ENOENT (No such file or
directory)
The alias file is also searched in /usr/local, and the sampled category
is "|" (!?), obviously not found.
>>>>| test-snprintf.c:61: assertion failed
>>>>| FAIL: test-snprintf
> You can test whether the attemted workaround actually works by using
> this test dir: http://www.haible.de/bruno/gnu/testdirs.tar.gz
Excellent, thank you. testdir1 all OK:
| checking whether snprintf is declared... yes
| checking for snprintf... yes
| checking whether snprintf respects a size of 1... no
| PASS: test-snprintf
But testdir2 fails:
| checking whether vsnprintf is declared... yes
| checking for vsnprintf... yes
| checking whether snprintf respects a size of 1... no
| ./configure: test: =: unary operator expected
| test-vsnprintf.c:75: assertion failed
| FAIL: test-vsnprintf
I think that is due to a typo, one lacking "v" in
testdir2/glm4/vsnprintf.m4. The attached vsn.patch fixes it, then
all tests PASS (appart the error gltests/test-stdint.c:249: two `l's in
integer constant).
>>>>| test-wcwidth.c:45: assertion failed
>>>>| FAIL: test-wcwidth
>> wcwidth(32)=-1, and the same for all ASCII characters.
> And iswprint(32) = ? And isprint(32) = ?
libc iswprint(x)=0 for *all* possible x. isprint(x)=1, for all ASCII
except 0-31 and 127, and behaviour above 128 correctly follows the
locale.
Alain.
--- testdir2/glm4/vsnprintf.m4 Sat Apr 19 19:05:10 2008
+++ testdir2.mod/glm4/vsnprintf.m4 Mon Apr 21 13:35:19 2008
@@ -17,7 +17,7 @@
;;
esac
fi
- if test $gl_cv_func_snprintf_usable = no; then
+ if test $gl_cv_func_vsnprintf_usable = no; then
gl_REPLACE_VSNPRINTF
fi
AC_CHECK_DECLS_ONCE([vsnprintf])
