[Opensim-dev] 0.5.9 Problems: Linking prims via libomv

[Shack Dougall]

As part of Prim Composer for 3ds Max, I'm writing a libomv-based 
importer for SL and OpenSim.  The importer, called maxport, performs 
this set of basic operations:

* rezzes prims, sets their position, name, and permissions
* uploads images, puts them in inventory, sets their permissions, 
applies them to prims
* links prims

This week I started using OpenSim's tags/0.5.9-released branch in SVN.  
Previously, I was running the 0.5.8 binary release.  All of my testing 
so far has been in standalone mode, basicphysics, and SQLite

0.5.8 seemed rock solid.  It took just about anything I could throw at 
it.  Sometimes, I had to slow things down for it, but it worked 
predictably.  0.5.9 seems less solid and in particular, I'm encountering 
huge problems with linking.

I've created a series of data sets with 100, 200, and 400 prims.  And 
also different amounts of linking, from no linking to 100 prim 
linksets.  The naming convention is "link_x_boxes_y" which indicates 
that it contains y boxes that are linked into x linksets.  So, 
link_10_boxes_100 means that there are 100 prims that have been linked 
into 10 linksets of 10 prims each.  link_2_boxes_100 means 100 prims in 
2 linksets of 50 prims each.

I started seeing problems immediately in 0.5.9.  link_2_boxes_4 should 
produce 4 prims that are linked into two pairs of 2.  And it did this in 
0.5.8.  But in 0.5.9, it produced 3 prims that were linked together with 
a 4th prim that was unlinked and out of position.

Larger linksets do even crazier things.  Linksets with 50 prims or more 
will spontaneously link to other prims in the sim that were never 
specified in the link request.  And I'm seeing duplicate prims that get 
linked in strange ways.  Some of the duplicates get their position set 
to half a sim away from where the originals were created.  It is bizarre.

I would probably think that I'd done something wrong, except that the 
same program works perfectly when run against 0.5.8.

Seems like there is a multi-threading issue or something similar going 
on that is causing things to go haywire inside of OpenSim 0.5.9 when 
linking is done.  It's probably a small bug, but it is doing serious damage.

So, here's the thing.  I have more information that I can provide.  I've 
made a lot of observations.  I'd be glad to give you a compiled copy of 
the program I'm using and the data I use to perform the tests.  Just let 
me know what you'd like.

--Shack