HI Jan, Thanks for the further testing and stack trace.
This is really peculiar - normally one hits threading problems when one has multi-core systems where race conditions and timing issues are much more stressed, on single core systems one typically sees few threading issues in comparison. So... typically if you can get things running solidly on a multi-core system the single core system will just work... In you case it looks like it could just be a single core timing related issue, the fact that using the threading event handler allows you to enter the threading model after construction is also a typical, as this actually requires stopping and starting of threads, reinitializing objects - things that would normally stress things more, not less... So the whole characteristics of the problem feels really odd, I don't really have any idea right now what might be up. Uninitialized variable might be a first candidate to look at given some of the characteristics. I will boot up my old laptop and get things update and see if it reproduce any of these problems as my multi-core systems are painfully rock solid... Robert. Robert. On 8/24/07, Jan Ciger <[EMAIL PROTECTED]> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Robert Osfield wrote: > > I'm pretty well stumped. A range of OS's and machines have now been > > reported to be working, but one very close to the specs of the > > machines I have here exhibits problems... > > > > Is it just particular models or any model? > > I have updated OSG to the current SVN, including your last fixes. I have > tried with cow.osg and few other models - the same hang. > > > Do you have any other systems you could try it out on? > > I do, but not SMP-capable, nor with an SLI config. I have tested the > problem on my T41p laptop, with a FireGL T2 (~ Radeon 9600) card and the > original Pentium M (not Core Duo, no SMP nor multithreading). > > I have used the cow.osg for testing and it also deadlocked. The > backtraces are attached. Sometimes it took few seconds up to half a > minute of moving the cow with the mouse, but ultimately a deadlock > occurred. I had also problem to reproduce the deadlock under a debugger > - - it never occurred while debugging the viewer on that machine. To > produce the backtraces, I had to run it normally and then attach gdb to > the deadlocked process. That suggests to me that the issue is quite > timing sensitive. > > The problem also seems to happen only if the viewer is started using the > given threading mode. If I change the threading mode to > CullThreadPerCameraDrawThreadPerContext after it has started already > using the 'M' key, I get no deadlock on neither machine. > > Regards, > > Jan > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.7 (GNU/Linux) > Comment: Using GnuPG with Mandriva - http://enigmail.mozdev.org > > iD8DBQFGz1ASn11XseNj94gRAqPjAKC5FMS10uLb7ix1u2/XfohdCKHSzgCgsVT3 > Mtlz6Bw+z4HZoBurAv7UzGM= > =wdll > -----END PGP SIGNATURE----- > > _______________________________________________ > osg-users mailing list > [email protected] > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > > _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
