realloc-gnu: Fix wrong configure test result with optimizing ibm-clang

2023-11-17 Thread Bruno Haible
On AIX 7.3, with CC="ibm-clang", I see different configure test results depending on optimization level: With -O0: checking whether realloc (0, 0) returns nonnull... no With -O2: checking whether realloc (0, 0) returns nonnull... yes This can be reproduced standalone: $ cat > foo.c #include in

mbrtowc, mbrtoc32, nl_langinfo, setlocale-null: Obey --disable-threads

2023-11-17 Thread Bruno Haible
The option --disable-threads from the 'threadlib' module was only meant to have an effect on the gnulib modules that use the Gnulib multithreading API internally. But from the 'configure --help' output, David Edelsohn got the impression that this option should apply also to the gnulib code that tes

Re: Test for pthread_rwlock_init() always passes, regardless

2023-11-17 Thread Sevan Janiyan
On 16/11/2023 20:41, Sevan Janiyan wrote: Hi, Though the issue is fixed in pthread_rwlock_rdlock.m4, there's an edge case where if we only rely on pthread_h.m4, things will fail since test-pthread.c uses 'PTHREAD_RWLOCK_INITIALIZER'. So, if I install all pthread modules & run make check, all

gettext: Recognize a statically built libintl on macOS and AIX

2023-11-17 Thread Bruno Haible
This patch fixes a failure of the "checking for GNU gettext in libintl..." test to recognize a libintl built with --disable-shared. It's a long-standing bug (certainly 10 years or older). It was not noticed because hardly anyone builds with --disable-shared nowadays. Only the GCC people have a habi