The error states I called ‘state flags’ earlier currently leave some room for problems. Oz says they have found that state flags were not getting set until late in the login process. Therefore, crashes early in the login process were not getting reported as crashes, but as disconnects only. This would have them thinking connection was more of a problem than it really was.
Another window for the numbers to get off is at log off. Some crashes can occur after the viewer says ‘bye’ but has not completely shut down. There is no disconnect but some of these late crashes don’t show up as a crash either.
The Google Breakpad candidate is where they are working to make the windows for this types of misrepresented reports as small as possible and also improve other error reporting.
The crash reporting process is changing too. Previously when the viewer crashed, a new reporting process was started, memory was allocated to the process, lots of data was gathered, network connections were started, and the data was transmitted to the Lab. It is not surprising that attempting all this in the middle of a crash is going to be fragile, tend to break, and never get the data to the Lab.
With the new Breakpad, the minimum that can be done is all that will be attempted. Basically meaning they will try to get everything they can written to a crash file. The next time the viewer starts, it will send in the crash file. It can do this when it is not in a panic state trying get off a message before it dies. This should be much more robust and give the Lab more information.
New Linden stats System
Oz tells us the Lab has changed over to the new stats system. This is the stats system that will allow them to gather better hardware stats.
Old reports are being ported over to the new system. This is slow going because users of the reports are helping decide which reports to keep, change, discard, and what new ones to add. Within the Lab hardware stats and things like how many people actually have ALM enabled is low priority. Which tells me this system is probably handling all the user stats too. Like how many people logged in today and how many times, etc., etc. Management comes ahead of engineering.
Oz tells us this system is going to be reporting statistics for all the Lab’s products. Those developers, like Patterns dev’s, also need stats. So, there is lots of competition for report designer time.
Nyx Linden was at the meeting to update TPV Dev’s on SSA (Server Side Appearance (Baking)). They were getting news that most of us got from the Thursday Beta Server meeting.
What was made clear or maybe I should say: more explicitly, is that SSA is a backend service. Upgrading it does not affect the host machines running simulators (the region servers). That means they can be updated without having to restart regions.
There are some broken spots in the SSA code. Apparently some of those are in the viewer. One is both server and viewer side and has to do with appearance failing due to Inventory fails. The viewer side code is likely to make it out well before the server side code gets to the main grid.
There is starting to be a goodly amount of SSA & Inventory fix-it-code that has not made it out to TPV Dev’s. Nyx feels it’s not stable enough to allow TPV Dev’s to be working on. This means the Lindens are unsure of the changes and may redo them, thus wasting TPV Dev time by giving them the current stuff.
Some of these fixes involve Inventory. There are changes to Inventory that presumably have to do with COF (Current Outfit Folder). There are likely some changes that have to do with how Inventory is downloaded via HTTP and how the COF is handled. There is apparently lots of stuff changing, at least backside. I doubt we will see any UI changes.
There will be an Interest List (IL) viewer some time soon. We will see it as either a release or project viewer. Andrew is pretty much done with the server side. Now the viewer side guys and gals have to get all the IL stuff working in the viewer.
I’m am guessing no sooner than week 34.
Baker Linden says he is making good progress on Group Bans. But, there is no viewer side code out to turn over to TPV Dev’s. Nor to put into a release candidate viewer. So, we continue to wait.
The Advanced Lighting Model (ALM) is causing slowdowns in TPV’s. It produces a significant hit in the SL Viewer. But, the SL Viewer is still getting 25+ FPS. Some TPV users are reporting it takes them down to 10 FPS. There isn’t much the Lab can do until it can be reproduced in an SL Viewer.
Monty Linden is working on HTTP communication in the viewer and servers. He has been doing performance testing for Windows and Mac. That will finish up shortly. He says there isn’t much performance improvement but things work much better. So, I think efficiency and robust.
Monty does say that the changes that reduce the connection to SL by 75% should be in a release candidate shortly… with shortly being like a Linden real soon™. I’ll guess probably this month.
This is something that would have a significant impact on performance and bandwidth use. Several people would like to see improved viewer caching. But, the project is meeting some resistance. Not because it isn’t seen as a good thing, but because the fruit is so far up the tree.
CHUI and SSA both were changes that touch a large number of the SL systems and processes. Changes like that are extremely complex. The Lindens expect a change to caching to be something similar. Remember. The Firestorm Team has been trying to get CHUI integrated into the FS Viewer for months and it took the Lab a year+ to get it out. There is still a project viewer undergoing changes for CHUI now.