On Mon, Aug 02, 2010 at 04:42:33PM +0200, Mike Hommey <m...@glandium.org> wrote: > On Mon, Aug 02, 2010 at 12:12:53PM +0200, Mike Hommey <m...@glandium.org> > wrote: > > Upstream is partially responsible for the problem. I'd say there are > > actually > > 2 different problems. The first one is that despite using weak symbols, the > > initialization steps don't check if the symbols are resolved before using > > them. That should IMHO be fixed in libxml2, instead of removing weak > > symbols. > > The fix should be straightforward. > > Reading the code more thoroughly, it does the check, but there may be code > paths where the result of that test is not checked before calling pthread > functions.
The attached patch works for me. Mike
diff --git a/parser.c b/parser.c index 85e7599..1db706b 100644 --- a/parser.c +++ b/parser.c @@ -14004,8 +14004,8 @@ xmlInitParser(void) { __xmlGlobalInitMutexLock(); if (xmlParserInitialized == 0) { #endif - xmlInitGlobals(); xmlInitThreads(); + xmlInitGlobals(); if ((xmlGenericError == xmlGenericErrorDefaultFunc) || (xmlGenericError == NULL)) initGenericErrorDefaultFunc(NULL); diff --git a/threads.c b/threads.c index 98fd2c2..1eeac0e 100644 --- a/threads.c +++ b/threads.c @@ -439,7 +439,8 @@ __xmlGlobalInitMutexLock(void) /* Make sure the global init lock is initialized and then lock it. */ #ifdef HAVE_PTHREAD_H /* The mutex is statically initialized, so we just lock it. */ - pthread_mutex_lock(&global_init_lock); + if (pthread_mutex_lock) + pthread_mutex_lock(&global_init_lock); #elif defined HAVE_WIN32_THREADS LPCRITICAL_SECTION cs; @@ -508,7 +509,8 @@ void __xmlGlobalInitMutexUnlock(void) { #ifdef HAVE_PTHREAD_H - pthread_mutex_unlock(&global_init_lock); + if (pthread_mutex_unlock) + pthread_mutex_unlock(&global_init_lock); #elif defined HAVE_WIN32_THREADS if (global_init_lock != NULL) { LeaveCriticalSection(global_init_lock);