dev-libs/libatomic_ops causes build failures at least on arm, hppa and sh because AO_fetch_and_add1_full is not defined on all arches. This especially concerns x11-libs/libdrm with libdrm_intel enabled. The failure seems to be related to this piece of code in atomic_ops.h: --- /* * The generalization section. * Theoretically this should repeatedly include atomic_ops_generalize.h. * In fact, we observe that this converges after a small fixed number * of iterations, usually one. */ #include "atomic_ops/generalize.h" #ifdef AO_GENERALIZE_TWICE # include "atomic_ops/generalize.h" #endif ---
Do we still care, now that video_cards_intel is use.mask'd on ARM, HPPA, and SH? It's not like xf86-video-intel or any other intel graphics driver component is every going to have any use on these platforms. Can't we just close this?
The AO_fetch_and_add1_full issue is not fixed. It just happens that no ebuild in portage uses it currently in a way which breaks. But we can resolve this as upstream, I guess.
Reopening as more breakage has been reported.
(In reply to comment #2) > The AO_fetch_and_add1_full issue is not fixed. It just happens that no ebuild > in portage uses it currently in a way which breaks. > > But we can resolve this as upstream, I guess. > Take a look at #353240 as it may have some relevant fixes.
Does this happen with =dev-libs/libatomic_ops-7.2_alpha4?