> I think it's probably to force unwind info to be generated for the > cleanup attribute (note that this test uses -fexceptions). I'm > testing this patch version that uses an undefined function. Ah, that makes sense. I like this version better. Thanks, Roland