http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59648
--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to David Kredba from comment #10) > (In reply to Markus Trippelsdorf from comment #9) > > So, invalid. > > Xorg not agrees: > > No, it's clearly valid, since PanoramiXNumScreens will always be set > 1 > in PanoramiXExtensionInit() if noPanoramiXExtension is not set. There > is no way that gcc can know this from compiling events.c alone, it would > have to do static analysis across dozens of source files, which it can't > do with the information passed to it during the compilation steps. > > > Setting PanoramiXNumScreens = 1 in Xext/panoramiX.c fixes the issue. > > How would gcc know that when you are compiling a completely unrelated > source file? It knows due -flto.