196245 2007-10-18 07:20 0000 gcc 4.2.x + glibc inline fix 2007-10-24 06:10:57 0000 1 1 1 Unclassified Gentoo Linux Core system unspecified All Linux RESOLVED FIXED http://sourceware.org/ml/libc-alpha/2006-05/msg00017.html P2 critical --- 1 tgall@gentoo.org toolchain@gentoo.org ppc64@gentoo.org ppc@gentoo.org tgall@gentoo.org 2007-10-18 07:20:45 0000 gcc 4.2.x is a bit more agressive on inlining and as a result breaks the glibc build. The manifestation is when glibc tries to run rpcgen which is the first binary that is run as the build is progressing. The fix is not a gcc fix but a glibc fix. It makes sure call_gmon_start is not inlined. (It isn't inlined with 4.1.x so the past behavior here isn't being changed with the fix. Supposedly this hits all architectures but I only have validated this on ppc64 Reproducible: Always tgall@gentoo.org 2007-10-18 07:23:31 0000 The patch from the following post I have validated does fix the problem. http://sourceware.org/ml/libc-alpha/2006-05/msg00017.html tgall@gentoo.org 2007-10-18 07:26:08 0000 I do believe with this fix that gcc 4.2.2 could be viable for 2007.1. (least on ppc64) :-) vapier@gentoo.org 2007-10-18 07:59:17 0000 less common arches are certainly free to stabilize gcc-4.2.x, but i dont plan on recommending stabilizing <=4.2.2 for any architecture ... tgall@gentoo.org 2007-10-18 18:16:44 0000 Understood. We haven't decided if we're going to 4.2.2 or not yet. That'll be a team discussion. On our part we have motivations for power6. Is the toolchain team ok with pulling in this patch to glibc then ? vapier@gentoo.org 2007-10-19 06:38:45 0000 yeah, patch looks fine ... i'll queue it up vapier@gentoo.org 2007-10-24 06:10:57 0000 added to glibc-2.7