dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Dec 02, 2009 at 01:53:04PM +0100, Johannes Ring wrote:
> On Wed, Dec 2, 2009 at 1:34 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> > On Wed, Dec 02, 2009 at 01:20:44PM +0100, Johannes Ring wrote:
> >> On Wed, Dec 2, 2009 at 12:46 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> >> > On Wed, Dec 02, 2009 at 12:10:22PM +0100, Johannes Ring wrote:
> >> >> On Wed, Dec 2, 2009 at 11:58 AM, Anders Logg <logg@xxxxxxxxx> wrote:
> >> >> > On Wed, Dec 02, 2009 at 11:24:56AM +0100, Johannes Ring wrote:
> >> >> >> On Wed, Dec 2, 2009 at 11:02 AM, Johannes Ring <johannr@xxxxxxxxx> wrote:
> >> >> >> > On Wed, Dec 2, 2009 at 10:58 AM, Anders Logg <logg@xxxxxxxxx> wrote:
> >> >> >> >> On Wed, Dec 02, 2009 at 10:24:18AM +0100, Johannes Ring wrote:
> >> >> >> >>> Hi,
> >> >> >> >>>
> >> >> >> >>> Someone pushed the Force Build button on the winxp-mingw32 slave. I
> >> >> >> >>> don't see a good reason for this since the demo.log shows it quite
> >> >> >> >>> clearly why it fails:
> >> >> >> >>>
> >> >> >> >>> http://fenics.org:8080/builders/dolfin-winxp-mingw32/builds/11/steps/dolfin%20check/logs/demo.log
> >> >> >> >>>
> >> >> >> >>> The problem is the same as the hardy-i386 slave reports, that is, the
> >> >> >> >>> following two demos exits with a segmentation fault:
> >> >> >> >>>
> >> >> >> >>> demo\pde\stokes\stabilized\cpp
> >> >> >> >>> demo\pde\stokes\taylor-hood\cpp
> >> >> >> >>
> >> >> >> >> I can't figure out why these fail. Valgrind does not complain on my
> >> >> >> >> machine. Does anyone have a 32 bit system to test on?
> >> >> >> >
> >> >> >> > Yes, I have my laptop. I will try to do some debugging.
> >> >> >>
> >> >> >> I got a segfault too. Here is the output from running
> >> >> >> demo\pde\stokes\stabilized\cpp in gdb:
> >> >> >>
> >> >> >> johannr@simula-x61:cpp$ gdb ./demo
> >> >> >> GNU gdb 6.8-debian
> >> >> >> Copyright (C) 2008 Free Software Foundation, Inc.
> >> >> >> License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
> >> >> >> This is free software: you are free to change and redistribute it.
> >> >> >> There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
> >> >> >> and "show warranty" for details.
> >> >> >> This GDB was configured as "i486-linux-gnu"...
> >> >> >> (gdb) run
> >> >> >> Starting program:
> >> >> >> /home/johannr/src/bzr/dolfin/demo/pde/stokes/stabilized/cpp/demo
> >> >> >> [Thread debugging using libthread_db enabled]
> >> >> >> [New Thread 0xb536a6c0 (LWP 21306)]
> >> >> >> Solving linear variational problem
> >> >> >>   Matrix of size 8604 x 8604 has 174636 nonzero entries.
> >> >> >>   Applying boundary conditions to linear system.
> >> >> >>   Applying boundary conditions to linear system.
> >> >> >>
> >> >> >> Program received signal SIGSEGV, Segmentation fault.
> >> >> >> [Switching to Thread 0xb536a6c0 (LWP 21306)]
> >> >> >> 0xb7c265c8 in dolfin::Data::clear () from
> >> >> >> /home/johannr/src/bzr/dolfin/local/lib/libdolfin.so.0
> >> >> >> Current language:  auto; currently asm
> >> >> >> (gdb) where
> >> >> >> #0  0xb7c265c8 in dolfin::Data::clear () from
> >> >> >
> >> >> > The line
> >> >> >
> >> >> >  x.resize(0);
> >> >> >
> >> >> > in the function Data::clear() in dolfin/function/Data.cpp looks
> >> >> > suspicious. Could you try replacing that line with
> >> >> >
> >> >> >  x.clear();
> >> >> >
> >> >> > and see if that helps?
> >> >>
> >> >> No, it didn't help.
> >> >
> >> > ok, can you some other things, like not calling clear() from the
> >> > previous location in the trace. And if that works, then comment out
> >> > the lines in clear() one by one.
> >>
> >> When I comment out line 62 in dolfin/function/GenericFunction.cpp, the
> >> stokes/stabilized demo seems to be running "forever" (I killed it
> >> after ten minutes).
> >
> > Strange...
> >
> > Is there some virtual machine or host I could get access to so I can
> > debug it?
>
> Yes, for instance gogmagog.simula.no or kangchenjunga.simula.no. Or I
> can give you access to the hardy-i386 buildbot slave if you prefer.
>
> Johannes

I've been spending a few hours on this so for without any luck. It's
difficult to spot what goes wrong. It's one of those cases where it
suddenly works if I print too much debug information...

I'll keep searching but any suggestions appreciated. This is the last
remaining issue before the release.

--
Anders

Attachment: signature.asc
Description: Digital signature