Martin Münstermann wrote:
OK, we now have thread compile failure reports on Debian and Slackware.
The config/acx_pthread.m4 script basically tests these:
acx_pthread_flags=3D"pthreads none -Kthread -kthread lthread -pthread -pthreads -mthreads pthread --thread-safe -mt pthread-config" =09 in that order and exits once it finds the first one that can compile/link this:
AC_TRY_LINK([#include <pthread.h>], [pthread_t th; pthread_join(th, 0); pthread_attr_init(0); pthread_cleanup_push(0, 0); pthread_create(0,0,0,0); pthread_cleanup_pop(0); ], [acx_pthread_ok=3Dyes])
Now, the big question is why -pthread can compile/link this successfully but -pthread isn't enough to build a library that uses threads.
I suspect that on Solaris, without "-lpthread" the pthread-stubs in /usr/lib/libc.so might be called at runtime instead of the functional ones in libpthread.so.
Could this easily be checked?
I can confirm this suspect for Solaris8/gcc 3.2.3. How to reproduce it: Add a line like printf("### NOTE: check_sigpipe_handler called.\n"); to check_sigpipe_handler() in fe-secure.c and rebuild libpq.so.
The line is only dumped out (indicating that check_sigpipe_handler() is being called), when the library is linked with -lpthread.
Without -lpthread no complains from the compiler or linker, but check_sigpipe_handler() is not called, because the stub of pthread_once() in libc.so does nothing.
--> We need -lpthread on solaris, too.
---------------------------(end of broadcast)--------------------------- TIP 7: don't forget to increase your free space map settings