The Beta Preview has been updated with a new version of the Second Life simulator for public testing.
PLEASE NOTE: This is an update of the server-side code that runs at our data centers. This is not a viewer update. Please do not post about viewer problems here, as it distracts from the core discussion about beta testing of the next release of the simulator (server) software.
This version includes fixes for many important issues, including what we hope is a resolution for the llTargetOmega (spinning things stop spontaneously and don’t start again) problem. The Beta Preview is a separate system, and can be entered using the beta viewer, which can be downloaded on our beta software download page. I just noticed that the release notes link on that page is out of date, and will have it removed as soon as possible. The notes below describe the specific changes in this release.
What Has Changed In This Version?
Changes in this build include (SVC items were submitted on the public jira issue tracker / DEV items are internally discovered issues:
DEV-11061: Vehicles no longer deflect severely when crossing prim seam boundaries
This issue has been problematic for vehicles throughout the beta process, and we believe we have a solution that will improve many vehicles substantially. Here are some developer notes from Andrew about this resolution, so that you understand the details and possible implications of this solution:
I’ve just checked in a fix that solves the prim-seams problem for the “#12 Pure Stock Brick (1976)” (a car that has severe prim seam bouncing problems), and I expect it to solve it for all other vehicles.
Note that our solution to the “prim seam problem” involves storing collision locations and normals over time and then filtering incoming new collision normals and sometimes altering them to agree with historical data. This can cause the normals for some collisions to be wrong momentarily until the correct normals win over the old, which can cause some problems, most notably momentary penetration with sudden obstacles or slope changes in the road.
At the end of my most recent changes I could not many really bad effects from these momentary penetrations, but it is something to keep an eye out for when testing.
DEV-13720: Rezzing an object from inventory in edit mode now results in all prims selected (rather than just the root prim as before)
DEV-12952 (and numerous pjira items): llTargetOmega item rotation sync restored.
After lots of research time, given that this was very hard to reproduce on demand, it is believed that the source of this issue is a ‘selection leak’, where the simulator’s view of what is selected (and thus should not be spinning) and the viewer’s view of this end up mismatching under certain scenarios. In particular this can happen if a complex item has been in edit mode or has been otherwise selected, it is shift-dragged, and the edit focus moves back and forth to the complex object and terrain or other objects quickly. There are likely various other ways to trigger the problem, however this was the closest we have gotten to creating the issue on demand to attempt to find the source. The resolution implemented here will, we hope, significantly improve the behavior, but has a few edge cases. A full resolution will take more work as a separate mini-project.
A related symptom that several people have noted is an increase in the number of “selected / sat upon” objects, which reduce the available prim count. After a region restart these phantom selected objects are removed from the list.
In order to ensure that the selection list is cleaned up, on a per agent (avatar) basis, the selection list is now scrubbed after the avatar has left the region and stayed away for several minutes. This timing is to avoid various other edge cases. Please note: If you see “stuck rotation” and just move to an adjacent region, the list may not be cleared, as your avatar will likely still be listed as a “child agent” that the simulator needs to track, to allow seeing across the region boundary.
Please see http://jira.secondlife.com/browse/SVC-2200 if you are a region owner who has encountered this problem, will test for it if we copy your region to the beta preview, and have time to actively test before Monday night. This jira will serve as a region copy request list for regions to be tested with this new code on the Beta Preview, before it is deployed.
DEV-13509: Phantom setting can now be toggled with llVolumeDetect enabled, even if the prims no longer have any scripts in them (a case appeared where a house was phantom – it had parts with llVolumeDetect set, but could not be set unphantom since the parts no longer had scripts in them… the related case of “how did they get llVolumeDetect set” is still unknown)
SVC-1987: Camera no longer hunts severely in boxing ring
SVC-1952 and DEV-12030: llSetLinkPrimitiveParams and llSetLinkPrimitiveParams PRIM_POSITION behavior improved
SVC-2113: Top Colliders list no longer missing entries
SVC-1951: llGroundRepel is now effective over water (parameter limits adjusted to allow negative limits for use with water)
DEV-11326: After some time, in some cases, physics scripts become inactive. This issue, reported by a handful of people, has now been traced to be linked to the llTargetOmega problem, and is believed resolved by that fix.
DEV-13126: Camera now stays on “your side” of a wall when the avatar is backed up to a wall
SVC-1975: Avatar jump height increased to match the earlier Havok1-based simulator’s behavior (adjusted impulse velocity of jump)
Havok4 Program Manager
P.S. Again please note that these fixes have been deployed to the beta preview grid for testing, and are not yet deployed to Second Life. If this simulator version appears solid on the beta preview with your review, it will be deployed to Second Life with these fixes, and timing of that release will be blogged separately.
P.P.S. The snapshot of the database on the Beta Preview grid is by now several months old. If you find that you are getting invalid password errors logging in there, please try an older password…