Hi Richard, This morning I awoke with the thought that the new atomic ref counting code most likely has broken the thread safe addParent()/removeParent(). To fix this we'll need to reintroduce another mutex for this task, either one per Object/Referenced or perhaps a shared one. Mutex's are quite costly memory/allocation/deletion wise so it might be good to share a single one.
Robert. On Tue, Oct 14, 2008 at 8:14 AM, Schmidt, Richard <[EMAIL PROTECTED]> wrote: > Hi Robert, > I am using osg 2.6.0 and you're right, osg::Node as well as > osg::Stateset use getRefMutex for protecting the access to the parent > list. However getRefMutex isn't used in the atomic refcounted case and > osg::StateAttribute doesn't use a mutex at all for protecting the > parents. > > The method (Node|Stateset|StateAttribute) ::getParent, ::getParents, > ::getNumParents doesn't seem to be protected by a mutex as well. I > wonder if that's intented. > > Richard > > >>HI Richard, >> >>Which version of the OSG are you using? The OSG has had mutex >>protection of addParents/removeParent() for quite some while. >> >>Robert. > > On Mon, Oct 13, 2008 at 7:59 AM, Schmidt, Richard > <[EMAIL PROTECTED]> wrote: >> Osg has the option to cache nodes by using osgDB:Registry. Retrieving > a node >> from the cache and attaching them to your subgraph is not threadsafe, >> because Node::addParent, Node::removeParent and Node::getParents is > not >> threadsafe. >> >> >> >> Consider this scenario: A node gets 'disposed' in the main thread, > because >> the 'user' deatches the node from the graph. At the same time the node > may >> be loaded from the cache by the databasepager. While the refcounting > will >> work perfectly, the calls to Node::addParent and Node::removeParent is > not >> threadsafe, causing the node to get an invalid parent setup. >> >> >> >> The same holds for StateSet and StateAttribute (because they have >> parentlinks as well). Example: The openflight loader stores static ref >> pointers to various StateAttributes. The StateAttribute::addParent and >> StateAttribute::removeParent may be called from multiple threads, i.e > the >> main thread releases a model, while another thread (the databasepager) > loads >> the model. >> >> >> >> Possible solution: Make the parent links threadsafe. For Statesets and >> StateAttributes I would remove them and implementing an "immutable" > pattern >> instead. >> >> >> >> Richard >> >> >> >> Richard Schmidt >> >> System Designer >> >> EADS Deutschland GmbH >> >> Organisationseinheit (SDGE1) >> >> >> >> EADS Deutschland GmbH >> >> Registered Office: Ottobrunn >> >> District Court of Munich HRB107648 >> >> Chairman of the Supervisory Board: Dr. Thomas Enders Managing > Directors: Dr. >> Stefan Zoller (chairman), Michael Hecht >> >> This E-mail And any attachment(s) to it are for the addressee's use > only. It >> is strictly confidential and may contain legally privileged > information. No >> confidentiality Or privilege is waived or lost by any mistransmission. > If >> you are not the intended addressee, then please delete it from your > system >> and notify the sender immediately. You are hereby notified that any > use, >> disclosure, copying or any action taken in reliance on it is strictly >> prohibited and may be unlawful. - Thank you. >> >> >> >> _______________________________________________ >> osg-users mailing list >> [email protected] >> > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.or > g >> >> > _______________________________________________ > osg-users mailing list > [email protected] > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.or > g > _______________________________________________ > 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
