Created attachment 434180 [details] emerge --info I am trying to merge latest version of xf86-video-intel with USE="sna". > [ebuild R ~] x11-drivers/xf86-video-intel-2.99.917_p20160507::gentoo USE="dri dri3 sna* udev uxa -debug -xvmc" 0 KiB It fails with the following error message. > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c: In function 'emit_span_primitive': > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2044:1: error: inlining failed in call to always_inline 'emit_span_vertex': target specific option mismatch emit_span_vertex(struct sna *sna, ^ > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2058:2: error: called from here emit_span_vertex(sna, op, box->x2, box->y2); ^ > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2044:1: error: inlining failed in call to always_inline 'emit_span_vertex': target specific option mismatch emit_span_vertex(struct sna *sna, ^ > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2061:2: error: called from here emit_span_vertex(sna, op, box->x1, box->y2); ^ > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2044:1: error: inlining failed in call to always_inline 'emit_span_vertex': target specific option mismatch emit_span_vertex(struct sna *sna, ^ > /var/tmp/portage/x11-drivers/xf86-video-intel-2.99.917_p20160507/work/88733a7874f7c9b45da5d612802947a9de12893a/src/sna/gen4_vertex.c:2064:2: error: called from here emit_span_vertex(sna, op, box->x1, box->y1); ^
Created attachment 434182 [details] build log
Still reproducible for version 2.99.917_p20160621
> CFLAGS="-Os -pipe -march=native" What if you change this to -O2 and/or -march=haswell ?
Compiled with options -O2. Wow. So it is the only package (in my system) which is broken (not with all USE flags) with -Os. I prefer -Os because I have SSD with limited free space. Is it possible to force one package in ebuild to build always with -O2 (no matter what are MAKEOPTS ?) Firefox is built that way as far as I remember, if I am not wrong. Or better to just add this package to /etc/portage/package.env ?
-Os does occasionally cause breakage. To work around this issue, we can inherit flag-o-matic and call replace-flags -Os -O2 But reporting the problem upstream so it can be addressed there is probably necessary.
(In reply to Pavel Kozlov from comment #4) > I prefer -Os because I have SSD with limited free space. FYI, there are known cases where -Os actually creates bigger binaries than -O2. The space savings are usually not worth the trouble one usually has with -Os, though the situation is much better than it was with older versions of gcc.
guess I'm going to change to -O2 as well...
also, 2.99.917_p20160829 is effected.
not really sure why no one else could do this, but I added flag-o-matic and replace-flags to the latest stable and testing to stop the breakage. thanks for the report, I'll leave this bug open in case someone wants to send it upstream.
We were waiting for someone who sees this problem to report it upstream, so upstream can comment on the correct way to address this issue.
Please try the unstable version 2.99.917_p20161206 and report back.
