-fno-omit-frame-pointer should be in ALLOWED_FLAGS. Since quite some time gcc defaults to omitting frame pointers with -O2 on systems where omitting frame pointers does not disturb with debugging thanks to the location lists features since gcc4/gdb-6.7ish. However it disturbs sample based profiling with sysprof (and I believe also oprofile) - without frame pointers they are not able to construct a function call tree in the end (sampling must be low resource usage and only frame pointers get saved for later processing I think). But -O2 is in ALLOWED_FLAGS, so a typical optimizing developer (like me, who routinely profiles whole desktop and system to find the bottlenecks..) machine is built with CFLAGS="-O2 -fno-omit-frame-pointer -ggdb ...", and packages that use strip-flags get frame pointers omitted because gcc gets -O2 passed but not -fno-omit-frame-pointer. -O2 is used because profiling makes not much sense when not profiling the typical -O2 optimized system. Please add -fno-omit-frame-pointer to the ALLOWED_FLAGS list in flag-o-matic.eclass
sounds fine to me if you feel like making the change yourself
(In reply to comment #1) > sounds fine to me if you feel like making the change yourself > Same here....or file bugs for where we actually call strip-flags so we can get rid of it :) I'd love to actually fix the compiler bugs instead of ignoring them.
I was merely trying to find out how more than 50% of my CPU resources (dual-core) was being used with mplayer -benchmark and found with sysprof that both mplayer and mozilla-firefox were missing -fno-omit-frame-pointer in its CFLAGS file in VDB (mplayer had no CFLAGS VDB file at all), so investigated it to be reminded again that strip-flags was causing it. I believe both are using strip-flags to get more support from upstream or some such. So I'd basically just add -fno-omit-frame-pointer to the short line with -fno-ident in it already? :)
ive added -fno-omit-frame-pointer to the allowed list http://sources.gentoo.org/eclass/flag-o-matic.eclass?r1=1.134&r2=1.135 but i dont think this is what you actually want. if you're saying the packages are building with -fno-omit-frame-pointer but you *dont* want them to, then this strip-flags change wouldnt matter. you actually want to be able to add -fomit-frame-pointer to your flags and/or have the build system of those two packages *not* add -fno-omit-frame-pointer themselves. try USE=custom-cflags in that case.
Thanks for making the change. I hadn't done it yet because I wasn't able to take the time yet to rebuild one of these big packages affected yet to verify it. As for the comment, please read what I actually wrote in comment #0. gcc is adding -fomit-frame-pointer itself implicitly with -O2 automatically on many architectures these days, so for sample based profiling I need to disable that with a -fno-*, but strip-flags filters that out. The ebuilds are not adding -fomit-frame-pointer -- GCC is.
your comment #3 seemed to imply you thought the speed hit was due to the FP being omitted, but whatever
