The testing program glib/mutex.exe fails sometimes with the mingw32 port in Windows XP.

The reason is that in g_thread_xp_get_srwlock() there is a race condition by which two different threads executing it concurrently can create two GThreadSRWLock objects. Then one of them will be thrown into oblivion.

Currently the code is as follows, abbreviated:

static GThreadSRWLock * __stdcall
g_thread_xp_get_srwlock (GThreadSRWLock * volatile *lock)
{
  GThreadSRWLock *result;
  result = *lock;
  if (result == NULL)
    {
      EnterCriticalSection (&g_thread_xp_lock);
      result = /* create the GThreadSRWLock */
      *lock = result;
      LeaveCriticalSection (&g_thread_xp_lock);
    }
  return result;
}

If two threads get to EnterCriticalSection() at the same time, one of them will be granted access while the other waits. Then the first one finishes, and the second one... creates other object and overwrites the previous.

My fix would be along the lines of:

static GThreadSRWLock * __stdcall
g_thread_xp_get_srwlock (GThreadSRWLock * volatile *lock)
{
  GThreadSRWLock *result;
  result = *lock;
  if (result == NULL)
    {
      EnterCriticalSection (&g_thread_xp_lock);
      /* Double check! */
      result = *lock;
      if (result == NULL)
        {
          result = /* create the GThreadSRWLock */
          *lock = result;
        }
      LeaveCriticalSection (&g_thread_xp_lock);
    }
  return result;
}

I can format a patch if you think it is worth it.