Re: RFM: transition of flint

[Julien Puydt <julien.puydt@laposte.net>]

Hi,

Le 06/09/2015 20:19, Tobias Hansen a écrit :
> Am 06.09.2015 um 20:14 schrieb Julien Puydt:
> > Hi,
> >
> > Le 06/09/2015 19:14, Tobias Hansen a écrit :
> > > Am 06.09.2015 um 13:27 schrieb Julien Puydt:
> > > > Hi,
> > > >
> > > > Le 05/09/2015 10:11, Julien Puydt a écrit :
> > > > > Le 04/09/2015 21:08, Julien Puydt a écrit :
> > > > > > Hmmmm... bad. If I get nothing tomorrow morning, I'll decide it's
> > > > > > stuck
> > > > > > and notify upstream.
> > > > >
> > > > > It got stuck.
> > > > >
> > > > > But I noticed it doesn't get stuck when I build upstream's sources
> > > > > directly (ie: without the debian flags). I'm now bissecting to find out
> > > > > exactly what triggers the problem.
> > > > >
> > > > > The transition is on hold until the matter is settled.
> > > >
> > > > There's a good chance it's a compiler bug... I reported:
> > > > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=798149
> > > >
> > > > In the mean time, what should I do? There's the option of waiting, or I
> > > > could disable the hardening-flags and override the lintian warnings and
> > > > still try to push the transition...
> > > >
> > > > Snark on #debian-science
> > >
> > > Hi,
> > >
> > > I think it's unlikely that the Debian gcc maintainers will debug this,
> > > because it is still totally unclear what the problem is. It's more
> > > likely that the flint maintainers would be interested to debug this
> > > further. Maybe they even know about this and can tell you which
> > > optimization may cause this, then one could disable only that one. It
> > > would be also ok to use the optimization level that flint uses by
> > > default if that solves the problem.
> > >
> > > Best,
> > > Tobias
> >
> > Flint's upstream doesn't have access to an i386 box and thinks it's a
> > compiler issue, so they're not very interested in debugging it.
> >
> > The problem is the -O2 flag, and we get it from dpkg-buildflags... which
> > is strange since I remember upstream also uses it...
> >
> > I might have another idea to test!
> >
> > Cheers,
> >
> > Snark
>
> -O2 turns on a list of additional optimization flags compared to -O1,
> see gcc(1). If one would know which one causes the problem one could
> enable -O2 and diable only that one optimization. Also it would give
> some hint to gcc people where the problem lies.

In fact -O1 is enough to trigger the issue... I tried to bisect a 
little, but since every step takes hours, I still haven't concluded 
anything.

In the meantime, I prepared a flint package version 2.5.2-2, compiled 
without optimisation : I checked it compiles on amd64 and i386, so it 
should be good to go...

Snark on #debian-science