Some progress is being made in trying to resolve the crash issues with the beta viewer code, and some of the changes made have been merged back into the development viewer branch. An update to the beta viewer was made on October 31st (22.214.171.1246400). This was followed on November 2nd, with release 126.96.36.1996511, which will remain available for testing over the next few days. Both of these releases have tcmalloc re-enabled and realloc disabled, with other updates reported in the release notes remaining the same as the previous two releases.
As it stands, tcmalloc is liable to remain enabled in further beta viewer builds for the time being, as work to completely disable tcmalloc from within the viewer code is now being carried out separately to the work on dealing with crash issues.Dealing with Future Bottlenecks
Given the degree with which this issue has impacted viewer releases, LL are looking into how to avoid similar situations from developing in the future. Details on this aren’t finalised or available as yet, but they have stated that they will update TPV developers with relevant information as to how they’ll be addressing matters once they have determined the best means of handling similar problems, should they ever occur.Project Code Merges Once the Crash Issue is Sorted
There is still no definitive statement or decision on the precise order in which code will be merged into the SL beta viewer once the crash issues are resolved, but all HTTP-related project code is on a high priority, and is expected to be within the initial merges and releases. This work includes Baker Linden’s viewer-side Group Services code, and Monty Linden’s own ongoing HTTP work, notably the new texture fetching code.
As it is, the code for both of these projects is available in repositories available to TPVs, and there have been moves to start merging the code into TPV code repositories. The Cool VL viewer has already merged the HTTP texture fetch project code, and the Group Services code has been merged into Firestorm so it will be included as a part of an upcoming release.
While not an actual SL project, this has been raised as a discussion point by Linden Lab during the week.
Those who have been in Second Life a long time (i.e. prior to 2007) may remember the old Proposal tab within the Viewer 1.x Group information floater (see right).
This allowed group moderators / owners to put forward proposals and ideas to the membership with a group and have them vote on the motions put forward, with results being returned on the basis of certain thresholds being set.
The server-side element of this functionality was in fact turned off in 2007 for a combination of performance issues related to the back-end database processing and on the simulators themselves.
However, various elements of the capability were not completely removed from either the server side of things or from the viewer. This means that with some viewers (Singularity, Phoenix and the old official Viewer 1.23.5 (which is still being used) for example), it is possible to still raise a proposal, have it added to a back-end database table and have it circulated to members of a group – the actual voting and reporting of results just no longer happens.
LL are currently engaged in various database optimisation projects, and as a result of this – and given the fact that the proposals system has not actually worked for some five years without user complaint (and Proposals tab itself has been long removed from Viewer 2 and Viewer 3 and TPVs based upon them) – this optimisation work is going to further break the proposals code. This means that in the future, any proposal raised will even be recorded in any back-end database table, but will simply be ignored as an unrecognised command.
However, rather than assume taking this action will not lead to issues within any viewer, LL have established two test regions on Aditi – Mesh Sandbox 22 and Kissling – which have the server-side code disabled. These can be used to test those viewers which still have still have the proposals code and tab included in them and ensure nothing untoward happens when a proposal is raised and then fails silently as a result of the server code being disabled.
It has been reported that Oskar Linden from the server team has departed Linden Lab.
If correct, the exact circumstances of his departure are unclear, although I have been told he did contact a number of people in-world on Friday November 2nd to inform them of the news. Oskar has always been one of the most approachable Linden staff, even when sometimes hampered by company policy on communications. He will be missed.