Open Source Meeting/2010-08-10

From Second Life Wiki
Jump to: navigation, search

Agenda Tuesday, 10 August 2010

  1. Weekly Snowglobe update - (without Merov Linden, he's on vacation)
  2. Branch desync?
    • Some commits from 2009/trunk are missing from the RC (2009/1.4) (which is normal), but there are also commits in the RC not present in trunk:
  3. Porting http-textures fixes from 1.5 to 1.4 (see jiras linked from SNOW-782). This seem to work very well, except for:
    • SNOW-799 - An assert when going back from 1.5 to 1.4. Is this a bug or expected behavior? i.e. do we fix it, or just increment LLTextureCache::sHeaderCacheVersion to force a cache purge?

Jira Issues

  1. SNOW-619 (Robin Cornelius)
  2. SNOW-713 Make libllcommon shared for SLPlugin + plugins on 1.4
  3. SNOW-596 Ready to commit (Aleric Inglewood)
  4. SNOW-593 idem
  5. Unscheduled SNOW issues triage: would be good to go with the community through those and decide for some what we should take immediately (2.0 or 1.4) or later.


[13:57] Techwolf Lupindo: I have to go to dinnor now. So will be unable to responc to questions.
[13:59] WolfPup Lowenhar: ok
[13:59] WolfPup Lowenhar: hope im able to stick around with not net issues knock on wood
[14:00] Bronson Blackadder: knock on prims
[14:00] WolfPup Lowenhar: actualy i have a wood desk RL
[14:02] WolfPup Lowenhar: hey robin
[14:03] Robin Cornelius: Hi
[14:03] Boroondas Gupte: hi all
[14:03] Robin Cornelius: Are we expecting lindens?
[14:04] Bronson Blackadder: they're late
[14:04] Boroondas Gupte: Merov's still on vacation, I think.
[14:04] Boroondas Gupte: Dunno about Oz
[14:04] Robin Cornelius: kk, thats a pretty graph on the screen
[14:05] Boroondas Gupte: newest revisions at the bottom. confusing for a gitk user like me :-P
[14:06] Boroondas Gupte: let's get started, shall we?
[14:06] Aleric Inglewood: Still no Merov?
[14:06] Thickbrick Sleaford: no
[14:06] WolfPup Lowenhar: he still on vac
[14:06] Aleric Inglewood: next week?
[14:06] Boroondas Gupte: Hi's gone on vacation for two weeks, this here is the second.
[14:06] Boroondas Gupte: (If I counted correctly)
[14:07] Robin Cornelius: good time to be on vacation, i would have avoided this week as wekk ;-)
[14:07] Robin Cornelius: *well
[14:07] Aleric Inglewood: OMG a branch desync!!!
[14:07] Thickbrick Sleaford does a Merov impersonation: so let's get started?
[14:07] Aleric Inglewood: But it's not really desynced... the patch is in the queue somewhere.
[14:08] Ardy Lay: Aleric, sorry. I got confused.
[14:08] Thickbrick Sleaford: que queue?
[14:08] Aleric Inglewood: the queue in my head :p
[14:08] Thickbrick Sleaford: ah
[14:08] Boroondas Gupte: apply your head to trunk ASAP, then :-P
[14:09] Aleric Inglewood: Oh well... only SNOW-800 is in my queue ;) ... so maybe it's worse than I think :). Please tell
[14:09] JIRA-helper:
[#SNOW-800] Massively duplicated objects
[14:09] Bronson Blackadder: expected behavior NMI
[14:10] Boroondas Gupte: what?
[14:10] Bronson Blackadder: lol
[14:10] Boroondas Gupte: SNOW-710 and SNOW-196 both got emergency patches/workarounds on 1.4
[14:10] JIRA-helper:
[#SNOW-710] ERROR: onBakedTextureMasksLoaded: No auxiliary source data for onBakedTextureMasksLoaded
[14:10] JIRA-helper:
[#SNOW-196] Deadlock in LLTextureFetchWorker::lockWorkMutex / LLThread::lockData / LLTextureFetch::lockQueue
[14:10] Boroondas Gupte: I guess we don't want these in trunk, but the real fixes instead
[14:10] Aleric Inglewood: Yes, those the emergency patches were deliberatly not committed to trunk
[14:11] Thickbrick Sleaford: s/196/169/
[14:12] Aleric Inglewood: SNOW-169 yes, is wrong on agenda too?
[14:12] JIRA-helper:
[#SNOW-169] Viewer crash: LLViewerImage::doLoadedCallbacks: ASSERT (mRawImage.notNull())
[14:13] Thickbrick Sleaford: I mean the emergency fix was for SNOW-169, no -196
[14:13] Thickbrick Sleaford: or am I wrong?
[14:13] Aleric Inglewood: Yes, and 169 was committed to 1.4 and not 1.5. But I see a 196 on the agenda too
[14:13] Boroondas Gupte: yep, SNOW-169
[14:14] Boroondas Gupte: I got it wrong from the picture, it seems.
[14:14] Boroondas Gupte blames Ardy
[14:14] Ardy Lay: Sorry. Dyslexia strikes again.
[14:14] Boroondas Gupte: :-P
[14:14] Ardy Lay: I did edit the image.
[14:15] Aleric Inglewood: SNOW-796 is gone too now from the Agenda?
[14:15] JIRA-helper:
[#SNOW-796] Clicking 'Reset to default' in the Debug Settings floater doesn't update cached control values.
[14:15] Thickbrick Sleaford: anyway the commits in question are and
[14:15] Aleric Inglewood: Everything looks fine to me thus.
[14:16] Thickbrick Sleaford: -796 is both in 1.5 and 1.4, so that's good.
[14:16] Boroondas Gupte: SNOW-796 has a commit in both 1.4 and trunk
[14:16] Boroondas Gupte: ok, let's move on, then
[14:17] Thickbrick Sleaford: Porting http-textures fixes from 1.5 to 1.4 (see jiras linked from SNOW-782[c]). This seem to work very well, except for:
* SNOW-799[c] - An assert when going back from 1.5 to 1.4. Is this a bug or expected behavior? i.e. do we fix it, or just increment LLTextureCache::sHeaderCacheVersion to force a cache purge?
[14:17] JIRA-helper:
[#SNOW-782] Snowglobe 1.4 / 1.5 http texture related issues
[14:17] JIRA-helper:
[#SNOW-799] Assertion failure ASSERT (entry.mImageSize == 0 || entry.mImageSize == -1 || entry.mImageSize > entry.mBodySize)
[14:18] Thickbrick Sleaford: aleric, do you know which part of the assert was false?
[14:18] Aleric Inglewood: no, sorry
[14:18] Thickbrick Sleaford: I suspect (entry.mImageSize > entry.mBodySize)
[14:19] Robin Cornelius: yes
[14:19] Thickbrick Sleaford: but I haven't seen this crash.
[14:19] Robin Cornelius: i've seen it assert
[14:20] Thickbrick Sleaford: this is from 2.1: if(entry.mImageSize <= entry.mBodySize)//it happens on 64-bit systems, do not know why
[14:20] Ardy Lay: I didn't notice any deliberate changes to the texture cache structure. Did I miss that?
[14:21] Robin Cornelius: 1.5 should resemble 2.1 now
[14:21] Aleric Inglewood: I don't think that increasing the cache version can be wrong... if not for 1.4, then at least for 1.3, cause apparently the two cache formats are not compatible. My only question is: is what 1.5 (with HTTP textures) does to the cache really intended? Or can we just fix it so that 1.4 can read the cache too again?
[14:21] Thickbrick Sleaford: Robin, the difference being?
[14:22] Aleric Inglewood: Thickbrick: that "I don't know why" and only on 64-bit systems look like a BUG thus... Maybe we should fix 1.5 so it doesn't happen.
[14:22] Robin Cornelius: i've still got to add the mCanUseHTTP stuff in that fixes bakes 100%
[14:22] Robin Cornelius: and the are changes to mutex handling and i'm just not going near that
[14:23] Thickbrick Sleaford: yeah, that's not something to do in a RC
[14:23] Aleric Inglewood: Seems to me we should just disable HTTP textures completely in 1.4.
[14:23] Robin Cornelius: i'm working on 1,.5 now, but any other differences in texturefetch are probably worth checking
[14:23] Robin Cornelius: by default they are disabled
[14:24] Aleric Inglewood: Not if you just ran 1.5
[14:24] Robin Cornelius: ah good point
[14:24] Aleric Inglewood: It should FORCE it to off.
[14:24] Latif Khalifa: releasing 1.4 without http textures would be pretty useless IMHO
[14:24] Aleric Inglewood: And then when Oz is fired we willl.. Oh there he is!
[14:24] Aleric Inglewood: heheh
[14:24] Thickbrick Sleaford: lol
[14:25] Aleric Inglewood: You're late :)
[14:25] Thickbrick Sleaford: since this causes problem to other viewers, I think we need to either change the cache version (so other viewres would clear cache) or make sure it doesn't happen
[14:25] Oz Linden: sorry - prior meeting ran long
[14:25] Thickbrick Sleaford: Robin, do you know the specifics of what causes this?
[14:25] Aleric Inglewood: Latif Khalifa: releasing 1.4 without http textures would be pretty useless IMHO <--- why?
[14:26] Robin Cornelius: Thickbrick, no not really
[14:26] Boroondas Gupte: I think 1.4 is different enough from 1.3 to warrant a release, even without the recent stuff.
[14:26] Latif Khalifa: people want the performance improvement that http textures bring. they enable it in totally broken viewers like emerald. 1.4 without http textures will add very little to 1.3
[14:27] Latif Khalifa: gazillion of biild tweaks, but nothing really substantial for a regular user
[14:27] Aleric Inglewood: Excuse me, but 1.4 is the first with the water fixes! :p
[14:27] Latif Khalifa: you asked for my opinion, i gave it to you :P
[14:28] Thickbrick Sleaford: I think that people *will* enable http texture in 1.4, whatever we do. So we should make sure it work reasonably well.
[14:28] Latif Khalifa: so all i get when i upgrade from 1.23.5 to 1.4 are the freezes
[14:29] Boroondas Gupte: :-
[14:29] Aleric Inglewood: ok... if it doesn't cause crashes and seems to work fine then why not... still seems a big change for a RC.. so can we keep the default to off?
[14:30] Thickbrick Sleaford: I'm not advocating changing the deailt to on
[14:30] Thickbrick Sleaford: *the default
[14:30] Aleric Inglewood: We either have to apply Robin's fixes to 1.4, or make it impossible to enable HTTP textures in 1.4 imho (ie, remove it from the advanced menu)
[14:30] Aleric Inglewood: vote 1 2 3, assigned to Robin: apply patches to 1.4. Next!
[14:31] Thickbrick Sleaford: no argument there, but what do we do with SNOW-799?
[14:31] Boroondas Gupte: can we make it default off on 1.4 and not inherit the setting from previously run 1.5?
[14:32] Boroondas Gupte: (e.g. by naming the setting differently)
[14:32] Aleric Inglewood: I don't see a problem with purging the cache when going from 1.5 to 1.4.
[14:32] Thickbrick Sleaford: Boroondas, releases (and release candidate?) use a different setting file than "test builds"
[14:32] Aleric Inglewood: Boroondas: The setting for HTTP textures?
[14:33] Boroondas Gupte: so what's the issue, then?
[14:33] Boroondas Gupte: Aleric, yes.
[14:33] Thickbrick Sleaford: I don't see an issue
[14:34] Aleric Inglewood: Not sure what happens then... probably that it will turn off after running 1.5, even when it's persistent.
[14:34] Aleric Inglewood: Might as well share the setting name then (assuming we fix it, no?)
[14:34] Thickbrick Sleaford: ~/.secondlife/user_settings/settings_snowgloberelease.xml
[14:35] Thickbrick Sleaford: ~/.secondlife/user_settings/settings_snowglobetestbuild2.xml
[14:35] Thickbrick Sleaford: umm, ignore that "2" there, I added that by hand.
[14:36] Aleric Inglewood: ... I vote for you doing as you please :p
[14:36] Thickbrick Sleaford: heh
[14:37] Aleric Inglewood: It's much easier for me to complain afterwards :)
[14:37] Aleric Inglewood: heheh
[14:37] Latif Khalifa: why not change the assert (backport from 2.1)?
[14:38] Latif Khalifa: is it already part of http fixes+
[14:38] Latif Khalifa: ?
[14:38] Robin Cornelius: ok just done the bake/http avoidance backport SNOW-802
[14:38] JIRA-helper:
[#SNOW-802] Baked textures should not even be attempted to be downloaded by http
[14:38] Thickbrick Sleaford: because it includes a "we don't know why this happens" comment.
[14:39] Robin Cornelius: ok so we need to find out WTF is going on and why it asserts
[14:39] Robin Cornelius: i'm probably good to do that
[14:40] Thickbrick Sleaford: my question on the agenda was orginally more of a "should this block moving the other patches to 1.4 for now"?
[14:40] Boroondas Gupte: probably not
[14:40] Thickbrick Sleaford: well, I gues we did cover that.
[14:42] Aleric Inglewood spend the last full four days trying to find all the global objects and their type... :(
[14:42] Boroondas Gupte: ok, we're through with topics, then, aren't we? Shall we get to #Jira_Issues
[14:43] Thickbrick Sleaford: I removed the two jiras there that were remnants from the first attempt of merging 2.1 - they both seem to not happen in 2.1 now.
[14:43] Boroondas Gupte: yeah
[14:44] Latif Khalifa: now that many people use sg test builds i have heard many reports of or some varient there off
[14:44] JIRA-helper: [#SNOW-467] remove:region() can stall for long periods of time leading to a viewer disconnect, esp after teleport
[14:44] Latif Khalifa: and it has propagated to emerald and imprudence which both use 1.x sg as upsteam
[14:45] Aleric Inglewood: This is my result: all the different TYPES of the global/static variables in secondlife-bin:
[14:45] Techwolf Lupindo: I got that a couple times, but then I do a LOT of tps on gridhunts.
[14:45] Aleric Inglewood: (conclusion: a LLLOOOOOTTTTT... and no way to "clean up" before the end of main())
[14:45] Latif Khalifa: for me personally, it makes SG unusable
[14:46] Thickbrick Sleaford: Aleric, doesn't stuff like that belong in trunk, not the branch?
[14:46] Nicky Perian: I never got the SNOW-596 patch to apply on windows
[14:46] JIRA-helper:
[#SNOW-596] APR memory pool used in an error prone way
[14:47] Aleric Inglewood: Thickbrick: this is research for memory leaks, so 596
[14:47] Boroondas Gupte: can we focus on one issue at a time, please? (If yes, which one? ;-))
[14:47] Thickbrick Sleaford: (Sorry...)
[14:48] Aleric Inglewood: I stubornly wanted to release all libapr pools at the end of the program, but it appears impossible. We'll just have to exit the application without cleaning up. Too many global variables.
[14:49] Aleric Inglewood: I don't want to look at a new bug, until I have 596, 593 and 707 committed :/
[14:49] Aleric Inglewood: SNOW-707
[14:49] JIRA-helper:
[#SNOW-707] Start up crash of LLErrorThread due to corrupt map in CheckLevelMap
[14:49] Latif Khalifa: oh well
[14:50] Latif Khalifa: why release a viewer that people can actually use :P
[14:51] Aleric Inglewood: Did anyone besides Boroondas test SNOW-596 yet?
[14:51] Aleric Inglewood: NickyP tries, but he can't seem to get the patch to apply.
[14:51] Aleric Inglewood: I'd like someone to try it on windows though.
[14:52] Robin Cornelius: Aleric ping me tomorrow, i'll try it then when you are around
[14:52] Aleric Inglewood: okies
[14:52] Robin Cornelius: BRB
[14:52] Thickbrick Sleaford: Aleric, I built it, but haven't had a chance to really test running it yet.
[14:52] Aleric Inglewood: I see you finally put your clothes in the washing machine Robin?
[14:53] Aleric Inglewood: thickbrick: that is on linux right?
[14:53] Thickbrick Sleaford: yes, 32 bit, non-standalone build.
[14:53] Aleric Inglewood: ok kewl
[14:54] Aleric Inglewood: So you are using a viewer with the patch applied, constantly?
[14:54] Thickbrick Sleaford: no
[14:54] Aleric Inglewood: (I'm running 1.4.1 Release now... needed to have a feeling for how it performs)
[14:54] Robin Cornelius: SNOW-802 appears to be working
[14:54] Latif Khalifa: o/
[14:54] Aleric Inglewood: Thickbrick: why not?
[14:54] Aleric Inglewood: You don't use quilt?
[14:55] Thickbrick Sleaford: because my system is messy and and I have a ton of different viewer builds
[14:55] Aleric Inglewood: Thickbrick:
[14:55] Thickbrick Sleaford: I'll test it more seriously
[14:55] Nicky Perian: How can the viewer 1 then 2 then back to 1 best be handled
[14:56] Aleric Inglewood: Very clean! :)
[14:57] Nicky Perian: v2 royally screwed over my avatar setting
[14:57] WolfPup Lowenhar: i know one thing with a lot of difficult work was finaly able to just about to completely build useing vs 2008
[14:57] Aleric Inglewood: hikaru:/usr/src/secondlife/viewers/snowglobe>ls
diff-archive/ jira-archive/ snowglobe-1.4.1/ snowglobe-1.x-i686-svn/ snowglobe-1.x-svn/ snowglobe-2.x-i686-svn/ snowglobe-2.x-svn/
[14:57] Aleric Inglewood: every viewer: just 'cd theviewerdir; configure; make'
[14:58] Boroondas Gupte: Nicky, SNOW-595 and SNOW-790, I guess
[14:58] JIRA-helper:
[#SNOW-595] I tire of Viewer 2 and Snowglobe 2 chewing up Snowglobe 1.4 setting files so I propose making Snowglobe 1.4 use a different directory for settings.
[14:58] JIRA-helper:
[#SNOW-790] Snowglobe 1.x and 2.x installers should use different installation directories
[14:58] Nicky Perian: that i can handle it the operations
[14:58] Latif Khalifa: anybody have any clue why deletijg objects can be 10 orders of magnitute slower? was some common type destructor changes, something about heap that got "fixed"?
[14:58] Boroondas Gupte: hmm ... can it have to do with the fixed static object cache, maybe?
[14:59] Boroondas Gupte: (I don't think so, though)
[14:59] Aleric Inglewood: Latif: it can happen when a linear search is done in a list of to-be-freed objects, for every object... so a double loop over many objects and then doing something stupid in the center of the loops.
[14:59] WolfPup Lowenhar: robin you had a chance to get your head around the maze of python scripts used for packaging the viewer?
[14:59] Latif Khalifa: no, seems de-allocating stuff in sg.1x can get be 10^6 slower
[14:59] Thickbrick Sleaford: umm
[15:00] Latif Khalifa: Aleric, that doesn't explain why snowglobe has this, but not 1.23.5
[15:00] Aleric Inglewood: Ok.... help me to reprocude it. I'm using 1.4.1 now. What to do?
[15:00] Thickbrick Sleaford: deleting objects is inherently expensive
[15:00] Thickbrick Sleaford: it might be that some parenting safety check that got introduced upstream at some point is slowing it down, though
[15:01] Robin Cornelius: why not limit the number of deletes per frame?
[15:01] Latif Khalifa: what got introduced upstream? this problem *only* happens in SG
[15:01] Aleric Inglewood: I got a stable FPS of 32 now
[15:01] Thickbrick Sleaford: (there are one-way pointers to parents, so when removing an object you have to iterate over all other objects to make sure you're not breaking any links)
[15:02] Thickbrick Sleaford: Latif, I don't think thouched that part in SG.
[15:02] Boroondas Gupte: ew. Sounds like two way pointers could pay off here quickly?
[15:02] Thickbrick Sleaford: yes
[15:02] Thickbrick Sleaford: (possibly, I looked at it about 4 months ago, and I am clueless regardless)
[15:03] Latif Khalifa: Thickbrick, there is something wrong in sg 1.x that does not happen in 1.23.5 official, i'm asking if anyone has an idea about what my cause this, something about changing allocation / deallocaition of stuff. I have no idea what might be causing it
[15:03] Aleric Inglewood: Latif: reading SNOW-467 .. that is in the middle of a teleport, not after it(?)
[15:03] WolfPup Lowenhar: well tc folks im going to be diving in to some python scripts used for packaging the viewer :/
[15:03] Thickbrick Sleaford: I guess that calls for some diffing
[15:04] Latif Khalifa: Aleric, i can repro it during the teleport, or after it, or just teleporting from one side of the sim to the next
[15:04] Thickbrick Sleaford: tc WolfPup
[15:04] Boroondas Gupte: since when does snowglobe have that issue, Latif? maybe we can do some bisecting to see what changed
[15:04] Robin Cornelius: ok i'm running as well, but yea some diffs against 1.23 may be helpful
[15:04] Latif Khalifa: or sometimes even just rotating avatar around while stuff is loading after the tp
[15:04] Thickbrick Sleaford: Latif, how long is the freeze when you tp inside an island sim?
[15:05] Latif Khalifa: take care robin
[15:05] WolfPup Lowenhar: more like hope i survive the dive ;)
[15:05] Latif Khalifa: TB, i get 0.05 fps so like 3 frames per minute
[15:05] Latif Khalifa: eventually i get "too many messages to processs in a frame"
[15:05] Aleric Inglewood: TP me somewhere
[15:05] Aleric Inglewood: ask me to look is certain directions whatever
[15:05] Latif Khalifa: (i think it says 400 msgs max per frame)
[15:06] Aleric Inglewood: look in*
[15:06] Thickbrick Sleaford: the problems with SNOW-467 is that it's doing all the kills in a single frame
[15:06] Latif Khalifa: TBB no that is not the problem
[15:06] Aleric Inglewood: ........
[15:07] Aleric Inglewood: Are you going to show me how to reprocude it or not?
[15:07] Latif Khalifa: the problem is that the kills take like gazillion times more to execute in sg than in 1.23.5 or 2.x
[15:07] Latif Khalifa: Aleric, there is no 100% repro
[15:07] Latif Khalifa: it would have been eaiser if there was ;)
[15:07] Latif Khalifa: lol
[15:07] Aleric Inglewood: I NEVER saw it before
[15:07] Latif Khalifa: if i sneeze you will braek your leg
[15:07] Aleric Inglewood: so... lets do something
[15:07] Thickbrick Sleaford: Latif, did you see that in fast timers?
[15:07] Latif Khalifa: Aleric, you never run release build
[15:07] Aleric Inglewood: meeting is over anyway it's 3:07

Generated with SLog Wikifier