Re: test-lock compilation failure on mingw

[Paul Eggert]

On 06/08/11 07:43, Eric Blake wrote:
> test-lock.c: In function lock_mutator_thread:
> test-lock.c:146:7: error: cannot convert to a pointer type
> 
> and many more like it; apparently, the new cygwin cross-compiler for mingw:
> 
> i686-pc-mingw32-gcc (GCC) 4.5.2
> 
> doesn't like converting gl_thread_self() into void* as a printf() argument.

POSIX doesn't guarantee that pthread_t is a pointer type, so
test-lock.c shouldn't be trying to print gl_thread_self ()
with %p; that's not portable.  Perhaps all instances of
code like this:

      dbgprintf ("Mutator %p before lock\n", gl_thread_self ());

should be rewritten like this:

      dbgprintf ("Mutator %p before lock\n", gl_thread_self_pointer ());

where gl_thread_self_pointer is a new primitive that is guaranteed to
yield a pointer value?  The value could be null on platforms where
pthread_t doesn't fit into void *.

Also, isn't there another problem here?  thread.h assumes
that NULL is an 'int', which surely isn't a portable assumption.
Here's a proposed patch:

diff --git a/lib/glthread/thread.h b/lib/glthread/thread.h
index 5d72040..b52646d 100644
--- a/lib/glthread/thread.h
+++ b/lib/glthread/thread.h
@@ -322,7 +322,7 @@ typedef int gl_thread_t;
 # define glthread_create(THREADP, FUNC, ARG) ENOSYS
 # define glthread_sigmask(HOW, SET, OSET) 0
 # define glthread_join(THREAD, RETVALP) 0
-# define gl_thread_self() NULL
+# define gl_thread_self() 0
 # define gl_thread_exit(RETVAL) 0
 # define glthread_atfork(PREPARE_FUNC, PARENT_FUNC, CHILD_FUNC) 0