Bug#667901: iceweasel: FTBFS on s390x: segmentation fault with jemalloc

[Mike Hommey]

On Sat, Apr 07, 2012 at 03:05:20PM +0200, Mike Hommey wrote:
> On Sat, Apr 07, 2012 at 01:06:45PM +0200, Philipp Kern wrote:
> > Source: iceweasel
> > Version: 10.0.3esr-3
> > Severity: important
> > User: debian-s390 at lists.debian.org
> > Usertags: s390x
> > 
> > iceweasel currently fails to build from source:
> > 
> > |   adding: hyphenation/hyph_is.dic (deflated 50%)
> > |   adding: hyphenation/hyph_da.dic (deflated 47%)
> > |   adding: hyphenation/hyph_gl.dic (deflated 69%)
> > |   adding: hyphenation/hyph_sv.dic (deflated 51%)
> > |   adding: hyphenation/hyph_nn.dic (deflated 52%)
> > |   adding: hyphenation/hyph_lt.dic (deflated 51%)
> > |   adding: hyphenation/hyph_bg.dic (deflated 72%)
> > | Segmentation fault
> > | make[2]: *** [install] Error 139
> > | make[2]: Leaving directory `/build/buildd-iceweasel_10.0.3esr-3-s390x-oh0PPR/iceweasel-10.0.3esr/build-xulrunner/xulrunner/installer'
> > | dh_auto_install: make -j1 install DESTDIR=/build/buildd-iceweasel_10.0.3esr-3-s390x-oh0PPR/iceweasel-10.0.3esr/debian/tmp AM_UPDATE_INFO_DIR=no MOZ_APP_BASE_VERSION=10.0 MOZ_PKG_MANIFEST=/build/buildd-iceweasel_10.0.3esr-3-s390x-oh0PPR/iceweasel-10.0.3esr/build-xulrunner/debian/installer/package-manifest MOZ_NONLOCALIZED_PKG_LIST=xpcom xulrunner debian MOZ_LOCALIZED_PKG_LIST=en-US SIGN_NSS= returned exit code 2
> > | make[1]: *** [stamps/install-xulrunner] Error 29
> > | rm build-browser/debian/installer/Makefile build-xulrunner/debian/installer/Makefile
> > | make[1]: Leaving directory `/build/buildd-iceweasel_10.0.3esr-3-s390x-oh0PPR/iceweasel-10.0.3esr'
> > | make: *** [binary-arch] Error 2
> > 
> > This failure is reproducible and happens when xpcshell is invoked.  Aurélien
> > pinned it down to a failure with jemalloc, as the build completes successfully
> > with --disable-jemalloc.  zelenka, the s390 porterbox, does have a s390x chroot
> > as sid_s390x.  Fedora apparently disabled jemalloc in response to this problem.
> > 
> > Aurélien further found this:
> > 
> > 10:25 <aurel32> what i know about jemalloc so far:
> > 10:25 <aurel32> - it works with version 7
> > 10:25 <aurel32> - it doesn't with version 8
> > 10:26 <aurel32> - using the jemalloc from version 8 in firefox 7 still works
> > 10:26 <aurel32> - building with --disable-jemalloc make firefox working
> > 
> > Mike, is disabling jemalloc to make it build an option or would you prefer to
> > dig into the problem?
> 
> I doubt jemalloc is the whole story. I think the js engine is the actual
> problem, and not using jemalloc merely hides it at build time.

So, it is in fact a jemalloc problem, and I have an actual fix for it.
As 10.0.4 is due on tuesday, I'll wait for that.

Mike