|Summary:||sci-physics/cernlib-2006 completely ignores CFLAGS & FFLAGS on sparc and chooses very bad defaults.|
|Product:||Gentoo Linux||Reporter:||Ferris McCormick (RETIRED) <fmccor>|
|Component:||Current packages||Assignee:||Gentoo Science Physics related packages <sci-physics>|
|Package list:||Runtime testing required:||---|
|Bug Depends on:|
Description Ferris McCormick (RETIRED) 2007-08-28 14:28:59 UTC
sci-physics/cernlib-2006 does build on sparc. However, it ignores CFLAGS (and FFLAGS) from make.conf, substituting its own defaults. For CFLAGS, it forces -O3 (in general, not a good idea) ignores -mcpu= from make.conf CFLAGS, and even sometimes FORCES -mcpu=v7. This results in building the C portion for essentially an SS2 and quite suboptimal performance on any sparc system anyone is likely to have. Fortran build is similar, although I did not notice any forced -mcpu=v7 flags. Since it builds, I'm giving it ~sparc for testing, but it really should use my CFLAGS/FFLAGS, not what it thinks I need unless there is some compelling reason to do otherwise. Please see Bug 59506 for an explanation.
Comment 1 Ferris McCormick (RETIRED) 2007-08-28 15:18:55 UTC
Add CC sparc.
Comment 2 Sébastien Fabbro (RETIRED) 2007-08-28 18:52:51 UTC
I will take a deeper look how to enforce users CFLAGS/FFLAGS as well as a decent src_test procedure. Useful for other arches anyway.
Comment 3 Sébastien Fabbro (RETIRED) 2007-09-03 15:58:12 UTC
just commited some cernlib fixes allowing user's flags and a src_test. Give it a try when you can. I will also propagate these changes in other cernlib packages (paw, geant, cernlib-montecarlo).
Comment 4 Ferris McCormick (RETIRED) 2007-09-17 13:57:37 UTC
(In reply to comment #3) > just commited some cernlib fixes allowing user's flags and a src_test. Give it > a try when you can. I will also propagate these changes in other cernlib > packages (paw, geant, cernlib-montecarlo). > That fixes the CFLAGS, thanks. Also, it does run many tests now. Apparently, it runs them successfully (at least, src_test does not take the "die" branch, and as far as I could tell from watching it run, everything reported success). So far as I am concerned, this bug is fixed and you can close it. Thanks for your help.