Re: Test for pthread_rwlock_init() always passes, regardless

[Bruno Haible]

Sevan Janiyan wrote:
> OS X 10.4 Tiger definitely lacks pthread_rwlock_init()

According to our function database
https://git.savannah.gnu.org/gitweb/?p=gnulib/maint-tools.git;a=tree;f=platforms/various-symlists;h=959af7a410a8416f4b183793ac2dae7e1063024f;hb=HEAD
the function does exist in Mac OS X 10.3, 10.4, and 10.5. In 10.5
they added an alias pthread_rwlock_init$UNIX2003, but that is not
relevant for the autoconf test.

>        dnl On Android 4.3, the pthread_rwlock_* functions are declared in
>        dnl <pthread.h> but don't exist in libc.
>        AC_CACHE_CHECK([for pthread_rwlock_init],
>          [gl_cv_func_pthread_rwlock_init],
>          [saved_LIBS="$LIBS"
>           LIBS="$LIBS $LIBPMULTITHREAD"
>           AC_LINK_IFELSE(
>             [AC_LANG_SOURCE(
>                [[extern
>                  #ifdef __cplusplus
>                  "C"
>                  #endif
>                  int pthread_rwlock_init (void);
>                  int main ()
>                  {
>                    return pthread_rwlock_init ();
>                  }
>                ]])],
>             [gl_cv_func_pthread_rwlock_init=yes],
>             [gl_cv_func_pthread_rwlock_init=no])
>           LIBS="$saved_LIBS"
>          ])
> 
> Compiling the test case with 'cc -Wall -O2 -lpthread' by hand results in 
> Bus Error

What do you mean? You get a Bus Error already at compile time, i.e.
the compiler or linker crashes?

Or do you get a Bus Error by running that test program? I.e. as if
we had an AC_RUN_IFELSE instead of AC_LINK_IFELSE in the above test?

> Otherwise things fail during compilation because it's 
> expecting an implementation from the host.
> test-pthread.c:33: error: ‘PTHREAD_RWLOCK_INITIALIZER’ undeclared here 
> (not in a function).

Ah, but that is a different problem. What you report here is that
no macro PTHREAD_RWLOCK_INITIALIZER is defined in Mac OS X 10.4,
while it exists in 10.5:
  #define PTHREAD_RWLOCK_INITIALIZER {_PTHREAD_RWLOCK_SIG_init, {0}}

It looks like lib/glthread/lock.h does better efforts at handling the
lack of PTHREAD_RWLOCK_INITIALIZER than lib/pthread.in.h. What is the
result if you change lib/pthread.in.h?

diff --git a/lib/pthread.in.h b/lib/pthread.in.h
index 48e0837478..fcba87dd9f 100644
--- a/lib/pthread.in.h
+++ b/lib/pthread.in.h
@@ -298,6 +298,9 @@ typedef unsigned int pthread_rwlockattr_t;
    { PTHREAD_MUTEX_INITIALIZER, PTHREAD_COND_INITIALIZER, 
PTHREAD_COND_INITIALIZER, 0, 0 }
 #else
 # if @HAVE_PTHREAD_T@
+#  if !defined PTHREAD_RWLOCK_INITIALIZER && (defined __APPLE__ && defined 
__MACH__) /* Mac OS X < 10.5 */
+#   define PTHREAD_RWLOCK_INITIALIZER { _PTHREAD_RWLOCK_SIG_init, { 0 } }
+#  endif
 #  if !defined PTHREAD_RWLOCK_INITIALIZER && defined 
PTHREAD_RWLOCK_INITIALIZER_NP /* z/OS */
 #   define PTHREAD_RWLOCK_INITIALIZER PTHREAD_RWLOCK_INITIALIZER_NP
 #  endif