marius mikucionis wrote:

We tried this here (upstream) with 2.2.1 on i686, and with latest CVS on
x86_64,
and we don't get the segfault but we do get:

interesting. but I use 32bit machine, you seem to use 64bit
architecture, doesn't it imply 64bit pointers and/or different memory
alignment? i.e. you just might be out of luck to trigger that fault
:-)
Yes, one of the machines we tried was a 64bit x86_64, but the other was a 32 bit i686.

did you try running valgrind? e.g. valgrind dot -o/dev/null compressor.dot
it should detect any memory mishandling.
the valgrind.log says that the problem is somewhere with splines code
in dynamic library:
==9257== Invalid read of size 4
==9257==    at 0x1BAFAFD5: (within /usr/lib/graphviz/libdotgen.so.0.0.0)
==9257==    by 0x1BAFA775: dot_splines (in /usr/lib/graphviz/libdotgen.so.0.0.0)
==9257==    by 0x1BAF2129: dot_layout (in /usr/lib/graphviz/libdotgen.so.0.0.0)
==9257==    by 0x8048929: (within /usr/bin/dot)
==9257==    by 0x1BB6096F: __libc_start_main (in
/lib/tls/i686/cmov/libc-2.3.2.so)

Such things should never appear in any useful software. If code
included debug info you would get the source file and line number in
this log, you'll love valgrind :-)

We had just started looking at valgrind output for another reason, so I'll be working on these
in the next few days.

$ dot compressor.dot >/dev/null
Error: lost IUT_ActionHandler ENV_Action edge
Error: lost IUT_ActionHandler ENV_Action edge
Error: lost IUT_ActionHandler ENV_Action edge

I also get these messages, not sure how to interpret them, seems that those edges are just lost (but why?).
I don't know either.  This is Emden's area.   He is investigating.

So we're not completely innocent, but perhaps the segfault is debian
specific?

could be. I am not debian-expert, I don't know how to rebuild the
package with debug info efficiently, sorry.

marius

I don't have any fast fixes for this, sorry, but thank you very much for the bug report.

John


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to