Hi,
On Thu, 2015-03-05 at 10:47 +0100, Johannes wrote: > Hello Gerrit, > > good to here from you :-) > > >>>>> 2. OSGShaderCacheTree.inl: The ShaderCacheTreeV3::sub() method is imho > >>>>> incorrect. The ShaderCache::clear() function currently does not work > >>>>> properly. I was facing references to already destroyed ShaderCache > >>>>> objects from ShaderProgramVariableChunks. > > > > for my understanding, what happens is you sub some cache entries and > > later on Cache::clear the system tries to delete them again. Your patch > > looks ok, though I'm still not sure where the subtle difference is ;). > > > > Is it easy to dump a dot file for this case. > > No it is not :-( The model with which I have worked on the issue was > rather big and I did have a complicated procedure list to reproduce the > error. I was able to find a way to get imho to the cause of the problem > in function 'void ShaderCacheTreeV3<ObjectT, LevelBits>::sub(UInt32 > uiIdx)' but can't remember the steps which have taken me to this place > in detail. The problem is the missing '0x0000 != (i & uiCurrBits) &&' > part in the second 'if' statement. This leads to remaining entries in > the cache which do pop up later on. ok, I see, that was the bit I missed ;(. > I do have some dot files and logging of instrumented code of the > original case. But I'm not sure that this information is of any help for > you. > > Attached you find what I do have. The session goes from program start to > crash. I did have the code somehow instrumented by adding member > '_oDbg_store' in the ShaderCache. I fill this set in the > 'updateRemoveCallback()' function for each 'addDestroyedFunctor' call. > The crash happens since this set is not empty in the ShaderCache destructor: > > ss << "ShaderCache::dtor = " << std::hex << this << " still in > _oDbg_store = " << *iter; > > For this reason, the living objects carrying the callback > 'ShaderCache::removeShaderVar' due to the 'addDestroyedFunctor' calls, > might execute the callback in the already destroyed ShaderCache object. > After inspection of the code I have found that the 'removeShaderVar' > function do not remove all entries in the '_oVarTree.sub(uiVarId);' call. > > I hope I could give you enough information to tackle the issue. thanks, I committed the patch as it looks fine. I'll continue to look at your test cases, just in case there is something else lurking around ;). kind regards gerrit ------------------------------------------------------------------------------ Dive into the World of Parallel Programming The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ Opensg-users mailing list Opensg-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/opensg-users
