Hi Ryan, I sounds like the lack of a readRefFontFile() is a potential problem. In the case of include/osgDB/ReadFile there are readRefNodeFile() etc. functions that pass back a ref_ptr<> rather than a C pointer, and are all safer in multi-thread situations where object cache is in play. Could you try implementing a readRefFontFile() and convert osgText across to use this to see if it helps.
Cheers, Robert. On Fri, Dec 11, 2009 at 1:04 PM, Ryan H. Kawicki <[email protected]> wrote: > I've been investigating a crash that has plagued us for quite some time. > I've finally gotten some time to sit down and take a look at this. > > Details: > OSG: 2.8.1 > OS: WinXP SP3 32Bit > CPU: 8 Core 2.5GHz Xenon > Graphics: Quadro FX 3700 512 MB > Memory: Plenty > > Description: > Our application embeds OSG into two views. There are multiple threads at > work here, but the two of importance are the Map thread and the DatabasePager > thread. > > At certain points during the execution of our application, the Map thread > will be given a request to create an osgText object. One of the operations > is to request a font for the osgText object. At this point we call down into > osgText::readFontFile( ... ). > > At this same time, the database pager is hard at work reading paged files > from disk on its own thread. At some point, the database pager gets into > Registry::readImplementation( ... ) and begins to copy the object cache to a > temp object cache. There is only one object in the cache and it is an > osgText::Font object. At some point during this stage, an unref of the > object happens. Here is where the oddity comes in. When the object is > getting unreferenced, the code checks to see if the object needs to be > deleted. For some odd reason this value is 0 on the DB thread but 2 on the > Map thread. When the Map thread tries to use the font object, it dies, as > the DB thread has already released it. I would give more details on this, > but the computer I was working on did not come back from standby and > everything I am stating is from memory. Sorry. > > Question: > Now that we have gotten through that, the question I have is related to the > implementation of osgText::readFontFile. Would it not be safer to have this > function use ref_ptr objects internally rather than standard pointers? I'm > quite confused as to the design decisions behind this and hope someone can > help me here. > > > Code: > > osgText::Font* osgText::readFontFile(const std::string& filename, const > osgDB::ReaderWriter::Options* userOptions) > { > if (filename=="") return 0; > > std::string foundFile = findFontFile(filename); > if (foundFile.empty()) return 0; > > OpenThreads::ScopedLock<OpenThreads::Mutex> lock(s_FontFileMutex); > > osg::ref_ptr<osgDB::ReaderWriter::Options> localOptions; > if (!userOptions) > { > localOptions = new osgDB::ReaderWriter::Options; > > localOptions->setObjectCacheHint(osgDB::ReaderWriter::Options::CACHE_OBJECTS); > } > > osg::Object* object = osgDB::readObjectFile(foundFile, userOptions ? > userOptions : localOptions.get()); > > // if the object is a font then return it. > osgText::Font* font = dynamic_cast<osgText::Font*>(object); > if (font) return font; > > // otherwise if the object has zero references then delete it by doing > another unref(). > if (object && object->referenceCount()==0) object->unref(); > return 0; > } > > > > > Thanks, > Ryan > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=21314#21314 > > > > > > _______________________________________________ > 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
