sldev-traffic no 13
sldev discussions through May 25, 2007
Memory Leak Patches, Leak in libcurl
Nicholaz Beresford posted patches dealing with memory leaks, and requested peer review. The patches are found attached to . There was no follow-up on the list.
Subsequently, Nicholaz also identified a leak that may be fixed in newer versions of libcurl. Nicholaz tested with 7.16.2. and using that version appears to fix an issued memory growing at constant rate over the lifetime of the program.
More leak finds followed from Nicholaz, each without further discussion on-list:
- fix a leak in llmessageconfig
- fix a leak in llcallingcard
- fix a leak in llviewerwindow
- fix a leak in llcurl
- fix a leak in lltoolmgr.cpp
- fix a leak in message.cpp
- fix a leak in llviewermenu.cpp
- fix a leak in llview.cpp
- fix a leak in llviewerpartsource.cpp, which paralleled the problem at
Most of these appear to be minor, i.e. mostly just memory not given up correctly at the end of the program.
System Memory Use Spiked by High End Graphic Boards
Nicholaz also discovered that the viewer allocates system memory matching the size of the video card's texture memory, plus a small multiple. This is problematic for low memory systems with high end graphics cards. Nicholaz posted a proposed patch, attached toand Steve Linden took the patch up for examination. After a revision and discussion, a fix may go into the next viewer.
Nicholaz posts a related patch at, which tackles the problem in the opposite direction, where texture buffers are too small on low-end cards. This one didn't receive attention. Callum Lerwick noted that this was related to , and Nicholaz updated the patch to address this issue as well.
Texture Debug Console Fixes
Nicholaz again! Nicholaz posted changes that unclutter the texture-mem part of the Shift-Ctrl-3 texture debug console. The change also fixes a couple uninitialized variables. This, at
Possible Particle Patch
Nicholaz (time for a macro!) also posted a patch, attached to, which attempts to solve the problem of new particles not showing up when lots of infinite-life (lifespan of zero) particles are emitted. As part of the patch, Nicholaz also fixes a particle memory leak associated with region crossings/changes. There is some ambiguous discussion on the list, and it's not clear whether this patch overlaps the patch on , also from Nicholaz.
Nicholaz is one busy Second Lifer!
Tateru Nino offered Nicholaz a kiss. :)
SL Viewer Stats Requested
LL loves keeping stats. Jason Giglio inquired as to whether SLDev could learn the breakdown on platforms in use, video cards, CPUs, average framerates, or any other information that might help the list.
To date there hasn't been a response.
Ben Byer notes that there seems to be a lot of dead code in the SL Viewer. Callum Lerwick points out that some of the code is common to the viewer and to server code, with Jason Giglio pointing out that shared code should exist in llcommon... the rest is fair game.
Ben made a first attempt at identifying dead code, however he was bitten by inlining making some functions appear to be unreferenced when linking. Nobody has yet tried again with inlining disabled. Meanwhile, Jacek Antonelli points to llanimalcontrols.cpp, llcape.cpp, llgenepool.cpp and llhippo.cpp as likely candidates for removal.
Bridie Holds Bug Triage
Rob Lanphier (Linden) was out this week. Bridie Saccocio (Linden) stepped in to take over Monday's session. Notes are at Bug_triage/2007-05-21. Thanks, Bridie!
A.C.Vandergraaf posted a survey link to the list, stating that there was a L$ prize for answering the questions. I'm not linking the survey directly; instead read the message and judge whether you think it's appropriate to the list. (I honestly couldn't decide, so I'm erring on the side of reproducing this -- Soft) Vandergraaf also posts a link, promising the results of the survey will appear there.
Texture Cache Update
Steve Linden posts a heads up that the texture cache is changing dramatically, specifically in the code pertaining to the entry lists. Check with Steve if you've got open patches against this system and want to know if they'll still apply. Steve indicates that the interface to LLTextureCache hasn't changed, so anyone working on a complete replacement - something Steve encourages - will be unaffected.
Sculpted Prims Patch
Dirk Moerenhout posts a patch to the list, stating that it speeds up patch generation and fixes problems with 32x32 textures. There was no JIRA created, and no discussion followed. The patch is attached to the original message.