As summary says, it's at keywordreq state in bug 272046. Depending on sys-devel/gcc[libffi] isn't a solution, because that gets broken when user does a gcc switch using gcc-config.
Eclass (and gnustep apps) does not care about libffi or ffcall enabled, so we only have to deal with gnustep-base :) I'll take a look and test it (hoping I don't have to immediatly rebuild all my gcc's with USE=-libffi!)
gnustep-base-1.18.0-r4 added under the virtual/libffi package.mask entry (gnustep-base has keywords for alpha ppc and x86-fbsd, not there yet in libffi). To get correct include path, I've supposed all libffi matching the virtual will have a pkg-config file, and used pkg-config to get the includedir
(In reply to comment #2) > gnustep-base-1.18.0-r4 added under the virtual/libffi package.mask entry > (gnustep-base has keywords for alpha ppc and x86-fbsd, not there yet in > libffi). It's now unmasked. > > To get correct include path, I've supposed all libffi matching the virtual will > have a pkg-config file, and used pkg-config to get the includedir > Correct.