Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Tuesday 10 February 2009: Any chance that FlightFear is modifying the ParticleSystem/ParticleSystemUpdate within the cull traversal? No idea. Have to check with one of our osg experts. What happens we you try differnt threading models like CullDrawThreadPerContext, SingleThreaded and DrawThreadPerContext? Will try later. For now I've recompiled everything with -O0, and have run fgfs for hours, without crash. I don't really think that the slightly changed compiler/linker flags have removed the cause for the segfault, but the now much lower frame rate has probably changed the timing such that it's now less likely to occur. I'll continue the tests, and will give some more useful info about my next crash (if any). Of coursem, I agree that there was no more reason to delay the 2.8 release. :-) m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hey, * Mathias Fröhlich -- Tuesday 10 February 2009: I can see no connection between the helgrind warning and a segfault. What helgrind tells you, is that there is a possible dealock situation that can potentially happen due to that inverted lock order. But the effect of a deadlock is a hanging application or thread or something like that - not a segfault ... Yes, that's what I even assumed. I was surprised to get a segfault on the lock. But then again, a segfault on the lock, a reported threading error there, ... and lack of knowledge about threading. ;-) * Mathias Fröhlich -- Tuesday 10 February 2009: For the nvidia drivers on linux you need the --smc-check=all valgrind argument. The thing appears to be that the nvidia drivers dynamically build code on the cpu that is not tracked correctly by valgrind without that argument Oh, that's news to me. Thanks a lot. Will try. When I tried to explore the core dump, I didn't get far. I had built with optimization, and several variables were not accessible. Will rebuild everything with -O0 today and re-run valgrind --smc-check=all. Good luck ... Thanks. :-) m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
[osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Robert Osfield wrote: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Did you see the message about rev 9722 not being in the branch yet? Paul Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
On Mon, Feb 9, 2009 at 2:10 PM, Paul Melis p...@science.uva.nl wrote: Did you see the message about rev 9722 not being in the branch yet? It arrived just after I pressed send :-) So yes have spotted it, currently waiting on the server. Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Robert Osfield wrote: On Mon, Feb 9, 2009 at 2:10 PM, Paul Melis p...@science.uva.nl wrote: Did you see the message about rev 9722 not being in the branch yet? It arrived just after I pressed send :-) So yes have spotted it, currently waiting on the server. Good ;-) BTW, you asked Stephan if he had committed it to the trunk, but I thought only you had permission to commit there? Paul ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi, * Robert Osfield -- Monday 09 February 2009: Do you agree/disagree that we are ready to tag? you'll probably not be surprised if I say: I disagree. :-) I had reported a segfault and the diagnose output of valgrind/helgrind, which affirms that there is a threading bug (maybe in the particles code?). It's very likely that this is exactly the bug which causes the segfaults. m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Robert, I did not test many things but IMHO, and seeing what was posted on mailing lists, I agree to tag the 2.8.0. Sukender PVLE - Lightweight cross-platform game engine - http://pvle.sourceforge.net/ Le Mon, 09 Feb 2009 15:08:40 +0100, Robert Osfield robert.osfi...@gmail.com a écrit: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Robert, On Monday 09 February 2009 15:08, Robert Osfield wrote: I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. I successfully compiled today morning's 2.8 branch on solaris, hpux and two linux variants. Looking forward to the release. Mathias -- Dr. Mathias Fröhlich, science + computing ag, Software Solutions Hagellocher Weg 71-75, D-72070 Tuebingen, Germany Phone: +49 7071 9457-268, Fax: +49 7071 9457-511 -- Vorstand/Board of Management: Dr. Bernd Finkbeiner, Dr. Roland Niemeier, Dr. Arno Steitz, Dr. Ingrid Zech Vorsitzender des Aufsichtsrats/ Chairman of the Supervisory Board: Michel Lepert Sitz/Registered Office: Tuebingen Registergericht/Registration Court: Stuttgart Registernummer/Commercial Register No.: HRB 382196 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Melchior, On Mon, Feb 9, 2009 at 2:24 PM, Melchior FRANZ melchior.fr...@gmail.com wrote: you'll probably not be surprised if I say: I disagree. :-) I had reported a segfault and the diagnose output of valgrind/helgrind, which affirms that there is a threading bug (maybe in the particles code?). It's very likely that this is exactly the bug which causes the segfaults. The segfault I saw with your example was because the example was broken as the viewer wasn't ref counted, this wasn't an OSG bug. I haven't had a chance to investigate the valgrind report though, have you had a chance? Has anyone else reproduce/investigated this issue? Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009: The segfault I saw with your example was because the example was broken as the viewer wasn't ref counted, this wasn't an OSG bug. But valgrind/helgrind reported a bug in *OSG*, in the same area where FlightGear crashes. I didn't let it run over FlightGEar, but over osgparticleeffects. No broken example involved here. (Unless osgparticleeffects is broken, too, of course. :-) I haven't had a chance to investigate the valgrind report though, have you had a chance? I'm sorry. I know that patches are more welcomed than bug reports. But it would probably have taken me many hours, if not days, to understand how the code is supposed to work, before I could even start to search for the bug. Time that I better spend on (OSS) projects where I know the code (and that need as much bugfixing). m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009: I've just run valgrind and it generates a number of what it thinks are errors, You are aware that I didn't use valgrind's standard tool mecheck, but its threading checker helgrind? $ valgrind --tool=helgrind ./osgparticleeffects ==31276== Thread #1: lock order 0x58F499C before 0x58F494C violated ==31276== Thread #1: lock order 0x58F6A9C before 0x58F6AEC violated OpenThreads and osgParticle have not changed in any substantial way for quite some while, so these errors have likely been there for a long time. Sure, it may well be that the broken threading is old and has caused segfaults before. :-P Do you get a crash in osgparticleeffects? If so how quickly? None. But OSG is a library. Whether examples crash or not is IMHO not really relevant, *if* there are bugs in the lib. Sure, we can just assume/pretend that helgrind is buggy and osg is bugfree. ;-) m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Melchior, El Lunes 09 Febrero 2009ES 16:34:45 Melchior FRANZ escribió: You are aware that I didn't use valgrind's standard tool mecheck, but its threading checker helgrind? $ valgrind --tool=helgrind ./osgparticleeffects ==31276== Thread #1: lock order 0x58F499C before 0x58F494C violated ==31276== Thread #1: lock order 0x58F6A9C before 0x58F6AEC violated Did you know that you can also attach a debugger so you can inspect directly every error reported by Valgrind? It is done adding --db-attach=yes to the command-line. Maybe the error could be more evident this way without having to dive into the code. Just my 2 cents, Alberto ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
HI Melchoir, On Mon, Feb 9, 2009 at 3:34 PM, Melchior FRANZ melchior.fr...@gmail.com wrote: You are aware that I didn't use valgrind's standard tool mecheck, but its threading checker helgrind? In my testing here I'm using the tool=helgrind Do you get a crash in osgparticleeffects? If so how quickly? None. But OSG is a library. Whether examples crash or not is IMHO not really relevant, *if* there are bugs in the lib. Sure, we can just assume/pretend that helgrind is buggy and osg is bugfree. ;-) 3rd party debugging tools aren't infaluable. The can and do produce false positives. They can and do also miss real problems. How much we can rely upon valgraind/helgrind is not something I can vouch for as this is the first time I'm attempted to use tool=helgrind. If we get a one to one mapping between a crash site and the helgrind warnings then we have something very tangible to go on. Right now I have an osgparticleeffects and osgviewer that doesn't crash, but have a valgraind/helgrind suggesting errors, but it's rather unclear on how to interpret those errors. We don't have this one to one mapping. Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Alberto, On Mon, Feb 9, 2009 at 3:54 PM, Alberto Luaces alua...@udc.es wrote: Did you know that you can also attach a debugger so you can inspect directly every error reported by Valgrind? It is done adding --db-attach=yes to the command-line. Maybe the error could be more evident this way without having to dive into the code. Could you provide an example command line example of what you mean? Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Robert, El Lunes 09 Febrero 2009ES 16:56:40 Robert Osfield escribió: Hi Alberto, On Mon, Feb 9, 2009 at 3:54 PM, Alberto Luaces alua...@udc.es wrote: Did you know that you can also attach a debugger so you can inspect directly every error reported by Valgrind? It is done adding --db-attach=yes to the command-line. Maybe the error could be more evident this way without having to dive into the code. Could you provide an example command line example of what you mean? Yes. By issuing valgrind --tool=helgrind --db-attach=yes osgparticleeffects valgrind stops on every error it reports with the question: Attach to debugger ? --- [Return/N/n/Y/y/C/c] if you answer yes, gdb is automatically attached and you can inspect thread's stacks. You can also quit the debugger without quitting the program and continue valgrind's run. Alberto ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Alberto Luaces -- Monday 09 February 2009: Did you know that you can also attach a debugger Yes, of course. But the helgrind output already contained *exact* line numbers where it found the locking order to be wrong. There wasn't much point in staring at the lines. Those by themselves aren't the bug, just the order in which they are called. And for that I would have needed a lot more (or any ;-) insight into the osg workings. m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi, * Robert Osfield -- Monday 09 February 2009: In my testing here I'm using the tool=helgrind Oh, sorry then. 3rd party debugging tools aren't infaluable. The can and do produce false positives. Yes, of course. But if I get a segfault in FlightGear which gdb traces back to some lines of code in OSG, and then helgrind points out a bug in pretty much exactly the same lines, then my first assumption is not that helgrind is buggy. And helgrind's report was IMHO very precise. How much we can rely upon valgraind/helgrind is not something I can vouch for as this is the first time I'm attempted to use tool=helgrind. Yeah, same here. I'm a total helgrind newbie. But still: segfault and threading bug reported by two independent tools in the same code. But before I annoy the last person here, I stop bitching. Just release OSG as it is. Maybe gdb and helgrind and FlightGear get fixed and the problem goes away all by itself. ;-) m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
HI Melchior, On Mon, Feb 9, 2009 at 4:11 PM, Melchior FRANZ melchior.fr...@gmail.com wrote: 3rd party debugging tools aren't infaluable. The can and do produce false positives. Yes, of course. But if I get a segfault in FlightGear which gdb traces back to some lines of code in OSG, and then helgrind points out a bug in pretty much exactly the same lines, then my first assumption is not that helgrind is buggy. And helgrind's report was IMHO very precise. How easy is it to build FlightGear these days? I was hoping to get it up and running on my machine so I could you it as another test platform for the OSG. I don't want to spend several days chasing up stuff though... have my own release to make... How much we can rely upon valgraind/helgrind is not something I can vouch for as this is the first time I'm attempted to use tool=helgrind. Yeah, same here. I'm a total helgrind newbie. But still: segfault and threading bug reported by two independent tools in the same code. But before I annoy the last person here, I stop bitching. Just release OSG as it is. Maybe gdb and helgrind and FlightGear get fixed and the problem goes away all by itself. ;-) As much as having bugs reported is annoying... if it's a real bug in the OSG then I want to see it fixed. A crash is enough to warrant holding back the release. At my end I do have the helgrind output that suggests a problem, but I don't have a crash, which makes it a bit more difficult for me to fully diagnose the problem and confirm a fix of the problem. The helgrind output is just a hint that somewhere in access of the ReadWriteMutex there is an issue, it doesn't actually pinpoint the actual failure mechanism. I am current reviewing the ReadWriterMutex implementation, and think I have a handle on a possible failure mechanism of the read and write threads changing the locks/unlocks in an unsafe way. I am looking into ways of refactoring this class to fix the potential problems. While it seems likely right now this this could be the cause of the seg faults you are seeing, I can't say this for certain. Also as I can't reproduce the crash the best I can hope is the valigrind will give the refactored code the all clear and FlighGear crash will be solved - but I'll need others to test this for me as I don't have FlightGear to test against. Are you around for the rest of the day? Can I push a testing question of new code in your direction? Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009: How easy is it to build FlightGear these days? Shouldn't be too hard for you, as you probably have a lot of the necessary components installed already (osg, libpng, libjpg, openal, alut etc.) So all you'll need is a decent scene graph lib:;-) $ svn co https://plib.svn.sourceforge.net/svnroot/plib/trunk plib We still use some parts of that -- networking, joystick, GUI. Then you need SimGear: $ cvs -d :pserver:cvsgu...@cvs.simgear.org:/var/cvs/SimGear-0.3 login CVS passwd: guest $ cvs -d :pserver:cvsgu...@cvs.simgear.org:/var/cvs/SimGear-0.3 co source And FlightGear itself: $ cvs -d :pserver:cvsgu...@cvs.flightgear.org:/var/cvs/FlightGear-0.9 login CVS passwd: guest $ cvs -d :pserver:cvsgu...@cvs.flightgear.org:/var/cvs/FlightGear-0.9 co source It's the usual ./autogen.sh ./configure make sudo make install everywhere, IIRC. At my end I do have the helgrind output that suggests a problem, but I don't have a crash, [...] I cannot reproduce the crash every time either. Happens once in a while, but *if* it happens, then the origin is always the same. There's nothing fuzzy about it. I am looking into ways of refactoring this class to fix the potential problems. That sounds much too dangerous shortly before a release. I'd rather look into that afterwards. Crashes aren't nice, but messing up a lib before a release is far worse. :-) Are you around for the rest of the day? Can I push a testing question of new code in your direction? Yes, and sure. But note that I can't reproduce it with every fgfs run. I'd have to run fgfs dozens of times and hope that it segfaults. m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Melchior, Thanks for the instructions on getting flight gear I'll have bash. On Mon, Feb 9, 2009 at 5:00 PM, Melchior FRANZ melchior.fr...@gmail.com wrote: I am looking into ways of refactoring this class to fix the potential problems. That sounds much too dangerous shortly before a release. I'd rather look into that afterwards. Crashes aren't nice, but messing up a lib before a release is far worse. :-) I think we have three choices: 1) Refactor OpenThreads::ReadWriteMutex to fix it. 2) Refactor osgParticle::ParticleSystem to use a Mutex instead of a ReadWriteMutext. 3) Ignore it and fix it after the release. There are pros and cons of each of these routes. My preference would be 1) fix the problem, but would fixing it also introduce the risk of breaking it even more? Solution 2) would introduce a performance bottleneck. Solution 3) isn't a solution :-) Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009: Thanks for the instructions on getting flight gear I'll have bash. You'd also need the data. And ~2GB free disk space, though you could leave most aircraft away, which saves a lot: $ cvs -d :pserver:cvsgu...@cvs.flightgear.org:/var/cvs/FlightGear-0.9 login CVS passwd: guest $ cvs -d :pserver:cvsgu...@cvs.flightgear.org:/var/cvs/FlightGear-0.9 co data I wrote that I can't reliably reproduce, but what I can do is put lots of particle-using aircraft in the scenery, which should increase the crashiness. Will try that now ... m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
At my end I do have the helgrind output that suggests a problem, but I don't have a crash, which makes it a bit more difficult for me to fully diagnose the problem and confirm a fix of the problem. The helgrind output is just a hint that somewhere in access of the ReadWriteMutex there is an issue, it doesn't actually pinpoint the actual failure mechanism. Just my 2 eurocents: We were using openthreads to lock a crew of worker threads which were accessing a shared file cache locked with a mutex. We got a crash every once in a while. To check for the problem we changed the code to use tbb (thread building blocks). After that it did not crash anymore. It is too long ago to recall the details (not much of a help to nail it down either, sorry :), but I would agree to point my finger to core openthreads and not valgrind/helgrind/ osgparticle etc. Stefan ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
On Mon, Feb 9, 2009 at 5:16 PM, Robert Osfield robert.osfi...@gmail.com wrote: I think we have three choices: 1) Refactor OpenThreads::ReadWriteMutex to fix it. 2) Refactor osgParticle::ParticleSystem to use a Mutex instead of a ReadWriteMutext. 3) Ignore it and fix it after the release. There are pros and cons of each of these routes. My preference would be 1) fix the problem, but would fixing it also introduce the risk of breaking it even more? Solution 2) would introduce a performance bottleneck. Solution 3) isn't a solution :-) Over dinner I reflected on the above options and feel that solution 2 is probably the most practical at this late stage in the release. I am now converting osgParticle::ParticleSystem across to all optional compile of use of Mutex or ReadWriteMutex. Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Stefan, On Mon, Feb 9, 2009 at 5:37 PM, Stefan Roettger ste...@stereofx.org wrote: Just my 2 eurocents: We were using openthreads to lock a crew of worker threads which were accessing a shared file cache locked with a mutex. We got a crash every once in a while. To check for the problem we changed the code to use tbb (thread building blocks). After that it did not crash anymore. It is too long ago to recall the details (not much of a help to nail it down either, sorry :), but I would agree to point my finger to core openthreads and not valgrind/helgrind/osgparticle etc. Did you use an OpenThreads::Mutex or ReadWriteMutex in your shared file cache? As far as I know OpenThreads::Mutex is sound, but ReadWriteMutex does look unsound as it stands. Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Melchoir et al. On Mon, Feb 9, 2009 at 5:00 PM, Melchior FRANZ melchior.fr...@gmail.com wrote: Are you around for the rest of the day? Can I push a testing question of new code in your direction? Yes, and sure. But note that I can't reproduce it with every fgfs run. I'd have to run fgfs dozens of times and hope that it segfaults. I have now checked in my changes to osgParticle::ParticleSystem to have an optional compile path for either the use of the OpenThreads::ReadWriteMutex or a OpenThreads::Mutex, with the standard Mutex now being the default. An svn update to the OSG-2.8 branch will get you these changes. Changing to mutex solves the valigrind/helgrind warnings when you run osgparticleeffects in SingleThreaded. I still see helgrind errors if run osgviewer cow.osg multi-threaded with lots of warnings in GL or osgViewer. I suspect most of these errors are benign though, the design of osgViewer is that it can accomodate not completely guarded data structures as the high level frame synchronisation of threads should avoid most issues. Could you test out these changes and let me know how you get on, Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Did you use an OpenThreads::Mutex or ReadWriteMutex in your shared file cache? As far as I know OpenThreads::Mutex is sound, but ReadWriteMutex does look unsound as it stands. Could have been ReadWriteMutex, but not 100% sure. The only sure thing is that tbb fixed it. Stefan ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009: An svn update to the OSG-2.8 branch will get you these changes. OK. I updated and recompiled, and will now test fgfs intensively. One evening won't be enough to confirm that the problem is fixed, but the fact that helgrind no longer complains is a good sign. I'll continue testing ... you can see me on our live-multiplayer map as cptf: http://mpmap01.flightgear.org/ Thanks for your efforts so far! m. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Robert, It seems that the osgautocapture example has stopped working along the way. It seems that DatabasePager::getRequestsInProgress() returns zero even after the terrain model has started loading. Please, test to reproduce with the osgautocapture example: ./osgautocapture --window 0 0 1024 576 --camera 55 10 30 0 30 0 --filename out.jpg terrain.ive Best regards, John Vidar Robert Osfield wrote: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
HI John, Could you please look into, I have enough on my plate already. Robert. On Mon, Feb 9, 2009 at 9:29 PM, John Vidar Larring john_larr...@sil.org wrote: Hi Robert, It seems that the osgautocapture example has stopped working along the way. It seems that DatabasePager::getRequestsInProgress() returns zero even after the terrain model has started loading. Please, test to reproduce with the osgautocapture example: ./osgautocapture --window 0 0 1024 576 --camera 55 10 30 0 30 0 --filename out.jpg terrain.ive Best regards, John Vidar Robert Osfield wrote: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
HI John, I've finished the other tasks then had a look at this issue. I was able to track it down to an incorrect for loop end iterator value in the DatabasePager::getRequestsInProgress(). This is now fixed and the osgautocapture example works once more. Robert. On Mon, Feb 9, 2009 at 9:33 PM, Robert Osfield robert.osfi...@gmail.com wrote: HI John, Could you please look into, I have enough on my plate already. Robert. On Mon, Feb 9, 2009 at 9:29 PM, John Vidar Larring john_larr...@sil.org wrote: Hi Robert, It seems that the osgautocapture example has stopped working along the way. It seems that DatabasePager::getRequestsInProgress() returns zero even after the terrain model has started loading. Please, test to reproduce with the osgautocapture example: ./osgautocapture --window 0 0 1024 576 --camera 55 10 30 0 30 0 --filename out.jpg terrain.ive Best regards, John Vidar Robert Osfield wrote: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
* Robert Osfield -- Monday 09 February 2009:
An svn update to the OSG-2.8 branch will get you these changes.
Tried with that ...
Changing to mutex solves the valigrind/helgrind warnings when you run
osgparticleeffects in SingleThreaded.
Confirmed. However, in fgfs I got again a segfault with particles
and threading ...
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xb67d16f0 (LWP 32467)]
osgParticle::ParticleSystemUpdater::traverse (this=0x96be668, n...@0x9843150)
at /home/m/fgfs/osg/include/OpenThreads/ScopedLock:31
31 explicit ScopedLock(M m):m_lock(m) {m_lock.lock();}
Can't say more, as I had to throw the core file away (don't ask :-).
Now I'm helgrinding fgfs+osg, and after that I'll make another
fgfs run and will then explore the next core dump ...
m.
PS: the ParticleSystemUpdater is a beast!
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi Melchoir,
On Mon, Feb 9, 2009 at 10:38 PM, Melchior FRANZ
melchior.fr...@gmail.com wrote:
Confirmed. However, in fgfs I got again a segfault with particles
and threading ...
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xb67d16f0 (LWP 32467)]
osgParticle::ParticleSystemUpdater::traverse (this=0x96be668, n...@0x9843150)
at /home/m/fgfs/osg/include/OpenThreads/ScopedLock:31
31 explicit ScopedLock(M m):m_lock(m) {m_lock.lock();}
Can't say more, as I had to throw the core file away (don't ask :-).
Now I'm helgrinding fgfs+osg, and after that I'll make another
fgfs run and will then explore the next core dump ...
I've just reviewed the ParticleSystemUpdater code and it looks
correct. Accessing mutex shouldn't just cause a crash. The only time
I could see it causing a crash is if the mutex itself is deleted and
the pointer being used to access isn't dangling. The
ParticleSystemUpdate has keeps ref_ptrParticleSystem to the list of
particle systems it's updating so dangling pointers shouldn't be an
issue.
Any chance that FlightFear is modifying the
ParticleSystem/ParticleSystemUpdate within the cull traversal?
What happens we you try differnt threading models like
CullDrawThreadPerContext, SingleThreaded and DrawThreadPerContext?
I'm turning in catch up with you in the morning.
Robert.
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
I tried on my application and everything seems good to me Robert Osfield wrote: Hi All, I'm ready to tag the 2.8.0. Feedback on the 2.8 branch and 2.8.0-rc2 has been positive so far, so it looks like the code is pretty ready. Do you agree/disagree that we are ready to tag? Now is your last opportunity to report a show stopper! Unless there is a show stopper reported I'll tag 2.8.0 at 1500hrs GMT today. Cheers, Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org -- +33 (0) 6 63 20 03 56 Cedric Pinson mailto:morni...@plopbyte.net http://www.plopbyte.net ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
Hi, On Monday 09 February 2009 23:38, Melchior FRANZ wrote: * Robert Osfield -- Monday 09 February 2009: An svn update to the OSG-2.8 branch will get you these changes. Tried with that ... Changing to mutex solves the valigrind/helgrind warnings when you run osgparticleeffects in SingleThreaded. Confirmed. However, in fgfs I got again a segfault with particles and threading ... Melchior, I can see no connection between the helgrind warning and a segfault. What helgrind tells you, is that there is a possible dealock situation that can potentially happen due to that inverted lock order. But the effect of a deadlock is a hanging application or thread or something like that - not a segfault ... Can't say more, as I had to throw the core file away (don't ask :-). Now I'm helgrinding fgfs+osg, and after that I'll make another fgfs run and will then explore the next core dump ... Good luck ... Mathias -- Dr. Mathias Fröhlich, science + computing ag, Software Solutions Hagellocher Weg 71-75, D-72070 Tuebingen, Germany Phone: +49 7071 9457-268, Fax: +49 7071 9457-511 -- Vorstand/Board of Management: Dr. Bernd Finkbeiner, Dr. Roland Niemeier, Dr. Arno Steitz, Dr. Ingrid Zech Vorsitzender des Aufsichtsrats/ Chairman of the Supervisory Board: Michel Lepert Sitz/Registered Office: Tuebingen Registergericht/Registration Court: Stuttgart Registernummer/Commercial Register No.: HRB 382196 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Please give approval if you feel OpenSceneGraph-2.8 is ready ; -)
An other sidenote regarding valgrind: For the nvidia drivers on linux you need the --smc-check=all valgrind argument. The thing appears to be that the nvidia drivers dynamically build code on the cpu that is not tracked correctly by valgrind without that argument. Greetings Mathias -- Dr. Mathias Fröhlich, science + computing ag, Software Solutions Hagellocher Weg 71-75, D-72070 Tuebingen, Germany Phone: +49 7071 9457-268, Fax: +49 7071 9457-511 -- Vorstand/Board of Management: Dr. Bernd Finkbeiner, Dr. Roland Niemeier, Dr. Arno Steitz, Dr. Ingrid Zech Vorsitzender des Aufsichtsrats/ Chairman of the Supervisory Board: Michel Lepert Sitz/Registered Office: Tuebingen Registergericht/Registration Court: Stuttgart Registernummer/Commercial Register No.: HRB 382196 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
