[Opensim-dev] SDAGUE: Spelunking the bug where avatar appearance reverts across server restarts

Justin Clark-Casey jjustincc at googlemail.com
Wed Oct 29 20:26:49 UTC 2008


I actually started tracking this down myself.  The major cause appeared to be that we weren't keeping a hold of the 
asset appearance textures (which we are now doing, though in a somewhat crude way that could potentially be optimized), 
so this may be resolved in r7003.

However, it's certainly not impossible that there are race conditions out there that are causing problems (I didn't 
investigate this aspect of things too closely).  So further feedback (via Mantis preferably) is welcome.

Pixelcat Schroedinger wrote:
> sdague, justincc suggested you were the best one to give me suggestions 
> on this...
> 
> So I've been doing some spelunking, the last couple of nights, chasing 
> the reason why avatar appearance reverts to default across server 
> restarts.  I haven't nailed it yet but I thought I'd mention what I've 
> seen so that perhaps someone can shed light on the subject for me.
> 
> Through lots of single-stepping and breakpoints, I've convinced myself 
> that the problem is happening at save rather than load.
> 
> I've managed to discover that there is a race condition at 
> appearance-change time and the window of overlap for the race condition 
> is just a couple of seconds long.  In particular a wait during update of 
> the avatar_appearance will cause changes to not be saved.  
> 
> In fact just by repeatedly wearing and  unwearing a shirt and varying 
> the pause time of the breakpoint in update avatar appearance, I was able 
> to get the client to become confused about state, to have the appearance 
> values become reverted and to actually lose several of the assets 
> permanently.  I also managed to get the simulator confused about whether 
> the avatar agent was logged in or not.
> 
> Conversely, I observed no race condition during appearance load time, 
> which leads me to conclude that the the thread contention on update 
> appearance is not intentional.
> 
> Can someone give me a quickie noob-lesson on the intended threading 
> model so I can understand better the behavior I'm tracing?
> 
> Thanks much,
> 
> -- Pixelcat
> 
> 
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-dev
> 


-- 
justincc
Justin Clark-Casey
http://justincc.wordpress.com



More information about the Opensim-dev mailing list