Kilian, On Sun, Mar 13, 2005 at 12:09:19AM +0100, Kilian Krause wrote: > Am Samstag, den 12.03.2005, 14:54 -0800 schrieb Steve Langasek: > > On Sat, Mar 12, 2005 at 09:09:04PM +0100, Kilian Krause wrote: > > > i've come to find a large problem debugging the FTBFS listed in > > > http://buildd.debian.org/fetch.php?&pkg=gnugk&ver=2% > > > 3A2.2.1-4&arch=alpha&stamp=1110438256&file=log&as=raw > > > Quite apparently i have neither a gdb backtrace (Cannot access memory at > > > address 0x202dbae000000000) nor a serious strace output > > > (http://backend.verfaction.de/~kk/gnugk.strace) to hand in. The only > > > guess i have so far is that either the dynlinker or gcc output code is > > > broken in latest SID. As i have no idea about where to start digging for > > > the issue causing this within the gcc-internals or glibc-changes, i'd > > > welcome a guiding light on how to get this solved. > > > As this seems be the only blocker for the pkg-voip packages to proceed > > > into Sarge, we'd be quite relieved to see this solved. Whoever has a > > > proposal what to try or which patch to add, i'm all ears. > > Well, I'm pleased to see this package does use -Wall, and builds cleanly -- > > but I also see that it's building with -Os, which should not be used in > > Debian packages. Any chance changing that to -O2 helps (in which case, > > there's a compiler bug, but only in a feature you shouldn't be using)? > > I'll grab the sources and see what I can find out on my alpha here. > that -Os comes most probably from pwlib which has size as a preference > for the sheer amount of OS abstraction layer functions which would > elsewise produce a quite larger library. That decision was taken > upstream and we have so far not questioned it, but if that is indeed an > issue, I see no problem putting a fixed version up. If however that's > not really mandantory, i'd prefer doing that post sarge as we're down to > about one left showstopper which is the gnugk build on alpha. (or at > least starting only that once the current pkg-voip debs are safe in > sarge with the current versions). Well, this is definitely a problem with build options. If I add the following two lines to Makefile.in: STDCXXFLAGS = -fno-elide-constructors OPTCCFLAGS = -O2 gnugk builds successfully on alpha, with a gnugk binary that doesn't segfault on startup. I'm running through a few rebuilds now to see which of these two changes is actually needed, and if -fno-elide-constructors is really needed or if it's sufficient to omit the -felide-constructors that was previously being set. Either way, the fix is within reach. Even if pwlib upstream things -Os is appropriate, btw, unless you can concretely justify its use for *this* package, I would still strongly recommend switching to -O2 in gnukg for sarge even if it's not actually needed to fix the alpha failure. Cheers, -- Steve Langasek postmodern programmer
Attachment:
signature.asc
Description: Digital signature