On Sat, May 23, 2015 at 12:33:10PM +0100, Stuart Henderson wrote: > On 2015/05/23 12:16, Antoine Jacoutot wrote: > > On Sat, May 23, 2015 at 10:59:02AM +0100, Stuart Henderson wrote: > > > Not sure whether this is binutils-related or something else changed > > > recently, but we are seeing some ports picking up libatomic.so from > > > gcc-4.8 on i386. > > > > > > As this is typically not listed in WANTLIB (only present in x11/ogre) > > > build often fails, and of course there will be runtime problems for > > > users who install the package but don't have gcc libs. > > > > > > Some specific examples, > > > > Common thing between initial-setup and epiphany is www/webkitgtk4 (which > > uses the gcc4 MODULE). > > Maybe www/webkitgtk4 needs a WANTLIB on libatomic? > > > > But that would mean adding gcc-4.8 as a LIB_DEPENDS which sucks... or we > > subpackage it? > > I have no fast i386 around so I cannot test my "webkitgtk4 theory" though... > > Looks like webkitgtk4 needs a LIB_DEPENDS+WANTLIB, and other things > depending on webkitgtk4 a WANTLIB. > > Does anyone know the criteria for when gcc decides to use libatomic? > Is it likely that sprinkling -march=i586 in some opportune places might > avoid this? (It's not like these are going to be ports which are useful > on a 486 anyway).
If it's just a matter of adding -march=i586 to webkitgtk4 for i386 archs, I am all for it. That at least covers most recent fallout we've been seeing. -- Antoine
