Summary: | Mozilla ebuild fails to build with -march=pentium4 | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Garen <garen> |
Component: | New packages | Assignee: | Martin Schlemmer (RETIRED) <azarah> |
Status: | RESOLVED FIXED | ||
Severity: | normal | ||
Priority: | High | ||
Version: | 1.4_rc2 | ||
Hardware: | All | ||
OS: | All | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- |
Description
Garen
2003-01-29 09:38:50 UTC
Yes, its a gcc3 issue. That is why there is the 'strip-flags' and changing -march=pentium4 to -march=pentium3 in the ebuild. If you remove those, what do you want me to do ? Hmm. I tracked the problem down from someone in the forums who had -march=pentium4 in their CFLAGS and reproduced it myself, so I figured the ebuild could just replace the -march with -mcpu or something else. Are you saying it's already supposed to change the -march=pentium4 setting? Yes, in theory. What version of gcc ? Just the other day, I could reproduce the crash with 3.2.1, 3.2.1-r6 and 3.2.1-r7. Now I can't. So I suppose either 1) I'm losing my marbles and did something different before, or 2) The ebuild changed I looked at the ebuild and noticed: if [ "$(gcc-major-version)" -eq "3" ] then # Currently gcc-3.2 or older do not work well if we specify "-march" # and other optimizations for pentium4. export CFLAGS="${CFLAGS/-march=pentium4/-march=pentium3}" export CXXFLAGS="${CXXFLAGS/-march=pentium4/-march=pentium3}" That could be changed to -mcpu=pentium4 instead. The only difference is the -msse2 flag thats silently enabled (not well documented by gcc that it does this either.) Well, it could enable other stuff that makes -march=pentium4 so unstable, and besides, I *know* pentium3 works 99% of the time fine. Yeah, probably a safer bet as its already set that way anyhow. Sorry for the crummy bug report. I'll do better next time. |