The root cause of this problem was that multiple ObjectUpdates were being sent on attachment which differed enough to confuse the client.
Sometimes these would eliminate each other and sometimes not, depending on whether the scheduler looked at the queued updates.
The solution here is to only schedule the ObjectUpdate once the attachment code has done all it needs to do.
Backport from head.
Fixes: Undo, T-pose of others on login, modifiedBulletX works again, feet now stand on the ground instead of in the ground, adds checks to CombatModule. Adds: Redo, Land Undo, checks to agentUpdate (so one can not fall off of a region), more vehicle parts. Finishes almost all of LSL (1 function left, 2 events).
Direct flames and kudos to Revolution, please
Signed-off-by: Melanie <melanie@t-data.com>
* Initialize the LLClientView prim full update queue to the number of prims in the scene for a big performance boost
* Reordered some comparisons on hot code paths for a minor speed boost
* Removed an unnecessary call to the expensive DateTime.Now function (if you *have* to get the current time as opposed to Environment.TickCount, always use DateTime.UtcNow)
* Don't fire the queue empty callback for the Resend category
* Run the outgoing packet handler thread loop for each client synchronously. It seems like more time was being spent doing the execution asynchronously, and it made deadlocks very difficult to track down
* Rewrote some expensive math in LandObject.cs
* Optimized EntityManager to only lock on operations that need locking, and use TryGetValue() where possible
* Only update the attachment database when an object is attached or detached
* Other small misc. performance improvements
* Removed two redundant parameters from SceneObjectPart
* Changed some code in terse update sending that was meant to work with references to work with value types (since Vector3 and Quaternion are structs)
* Committing a preview of a new method for sending object updates efficiently (all commented out for now)
During the heartbeat loop, Update() is called on every SceneObjectGroup which in turn checks if any SceneObjectPart has changed. For large regions (> 100k prims) this work consumes 20-30% of a CPU even though there are only a few objects updating each frame.
There is only one other reason to check every object on every frame, and that is the case where a script has registered the object with an "at target" listener. We can easily track when an object is registered or unregistered with an AtTarget, so this is not a reason to check every object every heartbeat.
In the attached patch, I have added a dictionary to the scene which tracks the objects which have At Targets. Each heartbeat, the AtTarget() function will be called on every object registered with a listener for that event. Also, I added a dictionary to SceneGraph which stores references to objects which have been queued for updates during the heartbeat. At each heartbeat, Update() is called only on the objects which have generated updates during that beat.
