requests to a removed archive file. ReaderWriterTXP::getArchiveName
will create an archive if it doesn't exist. This is causing a crash
if the terrain is removed and there are outstanding database requests
in the database pager. The request will create a new archive, and as
the archive is newly create doesn't have the materials etc to support
the subtile that was just loaded. To fix the problem getArchiveName
will only return existing archives. createArchiveName was added to
create archives.
"
Out application has the ability to switch to different types of terrains on the fly. This problem only exists in this type of situation.
The TXPArchive is held by the ReadWriterTXP class. When the TXPNode, which is the top level node, is released from memory, the archive associated to that TXPNode is also released. The issue is that the reference count on the TXPArchive never gets to zero.
The reason why the reference count never gets to zero is because the TXPParse, which is owned by the TXPArchive, stores a ref_ptr to the TXPArchive. You can then see why this becomes a problem. The TXPParser's ref_ptr cannot be unreferenced since the TXPArchive has not released the TXPParser.
Since the TXPParser is fully contained within the TXPArchive, I don't see the reason to have the TXPParser have a ref_ptr to the TXPArchive. I've made this change locally and have had no problems and our memory leak has been fixed.
"
Therefore I have changed all the occurances of atof by asciiToFloat or
asciiToDouble.
I believe that it is safe to do so at least for all the plugins.
Included here are also asciiToFloat conversion of environment variables. One
might argue that these should be locale dependent. But IMO these should be
set and interpreted by osg independent of the current locale.
"
The first bug is that the terrain tiles will page out to a lower LOD when they are right in front of you. The issue appears to be with the blacklisting heuristic which forces a tile to LOD 1, commenting out the usage of blacklisting with the LOD Nodes fixes our problem. This code change was made to line 29 of TXPPageLOD.cpp.
The second bug we were experiencing is that the database reader options never make it through to the archive loader. The use case for us appeared when the FID codes for the terrain were no longer on the materials. As it turns out the archive was being created twice, once by TXPNode and once by the ReaderWriterTXP on getArchive() so the options never actually got set on the archive that was being loaded. The fix is to first create the archive by calling getArchive on the ReaderWriterTXP, which stores it in a map for reference later, and then passing that archive into the TXPNode for it to set its internal member. With this code change we only create one archive (not sure what creating two did) and our options flags get set properly on the database.
The changes made are in TXPNode.h line 72 where the TXPArchive is now passed in. In the TXPNode.cpp the loadArchive(TXPArchive*) was changed to have the default behavior if NULL is passed in, if an archive is passed in then it does not load it since all the loading is done in the ReaderWriterTXP::getArchive(). The only other place that loadArchive is called is in TXPIO.cpp where a modification was made to pass in NULL which will have the same behavior as it used to. The last change is the little block of code starting on line 57 of ReaderWriterTXP.cpp, this was changed so that it first calls getArchive() which caches the archives in a map does some loading stuff and returns a pointer to it which is then passed in as a parameter to TXPNode::loadArchive().
The performance changes were made to TXPParser.cpp line 163 where we use to osgUtil::Optimizer on the node before passing it off, and on line 1456 we changed the geometry to use display lists. These small changes actually made drastic performance increases for us, as much as 1000% on certain laptops.
As far as testing goes, we have tested these changes with at least 5 txp databases on a variety of different computers including Mac OS and Linux. The base version used is 2.4."
multi-threaded paging, where the Pager manages threads of reading local
and http files via seperate threads. This makes it possible to smoothly
browse large databases where parts of the data are locally cached while
others are on a remote server. Previously with this type of dataset
the pager would stall all paging while http requests were being served,
even when parts of the models are still loadable virtue of being in the
local cache.
Also as part of the refactoring the DatabaseRequest are now stored in the
ProxyNode/PagedLOD nodes to facilitate quite updating in the cull traversal,
with the new code avoiding mutex locks and searches. Previous on big
databases the overhead involved in make database requests could accumulate
to a point where it'd cause the cull traversal to break frame. The overhead
now is negligable.
Finally OSG_FILE_CACHE support has been moved from the curl plugin into
the DatabasePager. Eventually this functionality will be moved out into
osgDB for more general usage.
completed the new registration of the plugin-readerwriters
("REGISTER_OSGPLUGIN") according to your osgstaticviewer-example (see
attachment, based on today's svn)."
