Summary: | www-client/mozilla-firefox-3.5.2-r1 seqsev on launch | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Boney McCracker <brendlerjg> |
Component: | Current packages | Assignee: | Mozilla Gentoo Team <mozilla> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | jlec, toolchain, zima |
Priority: | High | ||
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: |
'strace -o strace.txt firefox'
'strace -o strace.txt firefox' 'strace -o strace.txt firefox' 'strace -o strace.txt firefox' |
Description
Boney McCracker
2009-08-30 00:03:50 UTC
Created attachment 202676 [details]
'strace -o strace.txt firefox'
Created attachment 202677 [details]
'strace -o strace.txt firefox'
Created attachment 202679 [details]
'strace -o strace.txt firefox'
Created attachment 202680 [details]
'strace -o strace.txt firefox'
*** Bug 283187 has been marked as a duplicate of this bug. *** *** Bug 283186 has been marked as a duplicate of this bug. *** Sorry about the duplication (both bugs and attachments, it seems). Some sort of browser mishap. Confirmed here, even I suspect something related to gtk+, as is the only upgrade that I can think of tha can be had some effects since I've noticed the issue. Investigating... I have a similar issue with mozilla-firefox-3.5.2-r3, mozilla-firefox-3.5.3 and even with mozilla-firefox-bin-3.5.3. After reinstalling whole world with gcc-4.4.1 -O3, all the foxes segfault upon start before showing the window. I guess it is specific for x86, for amd64 seems ok, but I am not 100 % sure. The problem is with zlib (in my case 1.2.3-r1) when compiled with gcc-4.4.1 -O3. Just recompiling zlib with either gcc-4.3.4 -O3 or gcc-4.4.1 -O2 fixes the problem and I can enjoy the "Thank you for downloading Firefox" video :) In particular, the tree-vectorize optimization causes the bug, hence zlib still compiles fine even with CFLAGS="... -O3 -fno-tree-vectorize ...". Maybe this helps, but probably it is a different issue. I can see -O2 in your CFLAGS, but you may have compiled zlib with -O3 earlier (???). You also mention sqlite based on strace observation. Try >=mozilla-firefox-3.5.2-r2 (with >=xulrunner-1.9.1.2-r2), they now use internal bundled sqlite instead of the system one. You can also try just simply running /usr/lib/mozilla-firefox/firefox in gdb and wait for the crash. Even with minimal debug information you may see the library where the segfault appears. So should I file a separate bug report concerning firefox and zlib? Ok, I just got more crashes, this time caused by fontconfig-2.7.3 in exactly the same way as with zlib. Again the same solution (-O2 or -fno-tree-vectorize), tree-vectorize is really bitch on x86! This is likely a gcc-4.4 bug with x86 + -fforce-addr For gcc-4.4 + x86 + -ftree-vectorise, see bug 270120 No, I've always had -O2 on that machine. I will remove -fforce-addr and rebuild. (In reply to comment #12) > No, I've always had -O2 on that machine. > I will remove -fforce-addr and rebuild. > Yes, this fixed it. Thanks, Nirbheek. Please do not close, we either need to filter it via the ebuilds are explode in portage. you would have to filter flags in half a dozen ebuilds (that we know of). just don't use -ftree-vectorize on x86 until bug #270120 is fixed. yeah it's a pain in the ass. (In reply to comment #15) > you would have to filter flags in half a dozen ebuilds (that we know of). just > don't use -ftree-vectorize on x86 until bug #270120 is fixed. yeah it's a pain > in the ass. > Well, it seems besides -ftree-vectorize, -fforce-addr is also problematic. vapier was talking about a big phat warning in gcc in case -ftree-vectorize is enabled, maybe this should be added to that list Closing!!! |