User:Andrew Linden/Office Hours/2010 05 18

Transcript

[10:57]	Ardy Lay:	Hi Simon
[10:58]	Simon Linden:	Hello
[10:59]	Bronson Blackadder:	meow
[10:59]	Sindra Aquila:	hello simon and all others
[10:59]	Ovaltine Constantine:	Hai guize
[10:59]	Ovaltine Constantine:	Teehee
[11:00]	Ardy Lay:	Hi Andrew
[11:00]	Sindra Aquila:	hi Andrew
[11:00]	Welcome to	Linden office hours
[11:00]	Andrew Linden:	hello
[11:00]	SHAMUS Carter:	hi andrwe
[11:02]	Sindra Aquila:	i am still trying to solve anti bounce but all i do makes it much worse instead better
[11:02]	Andrew Linden:	I don't have any new ideas on that. Have you asked any star SL resident scriptors?
[11:02]	SHAMUS Carter:	your office hours are open to the pub right?
[11:03]	Andrew Linden:	Yes SHAMUS. Welcome.
[11:03]	Sindra Aquila:	i asked several except the weaponmakers yet, i am bit weary to do that
[11:03]	Ovaltine Constantine:	I'm the best scripter on the grid guize. Whats the problem?
[11:03]	Ovaltine Constantine:	Teehee
[11:04]	SHAMUS Carter:	hi yah xstorm
[11:04]	Sindra Aquila:	i tried mass, i tried the line i showed all makes it lot worse
[11:04]	Sindra Aquila:	anti bounce, BUT i no longer believe it is in a scripot, but tempering weith prim especially frictioon
[11:05]	Andrew Linden:	Ovaltine, Sindra is trying to make a bow+arrow where the arrow actually stops when it collides with something
[11:05]	Sindra Aquila:	problem is the makers have done it, but it is a secret
[11:05]	Bronson Blackadder:	lol
[11:05]	Bronson Blackadder:	LLtemp_link
[11:06]	Andrew Linden:	since physics and scripts run asnynchronously, and scripts only get a little runtime window, sometimes there is lag between the collision and its scripted callback
[11:06]	xstorm Radek:	i think i have found a little problem with the server software and havok 7 but im still running crash tests till i find what the trigger is
[11:06]	Opensource Obscure:	I think I made a ball that stops x seconds after hitting something. can this help?
[11:06]	Andrew Linden:	which causes the arrow to bounce a bit away before it can stop itself
[11:06]	Simon Linden:	So to refresh my faulty memory, if you stop the arrow on the collision event, it doesn't work because of the delay in getting the event, right?
[11:07]	Sindra Aquila:	yes, but needs then tested
[11:07]	xstorm Radek:	hi Shamus i think you seen the sim crasher ?
[11:07]	Simon Linden:	Does the collision event have enough information so you can place the arrow (after you stop it) in the right spot?
[11:07]	Andrew Linden:	xstorm, you think you found a crash mode in Havok7?
[11:07]	SHAMUS Carter:	yah
[11:07]	xstorm Radek:	yes
[11:07]	SHAMUS Carter:	it was nasty
[11:07]	xstorm Radek:	Shamus has seen it too
[11:08]	Sindra Aquila:	the line that sets the arrow to non phys comes to late in l;aggy sims like ENLAVED and then the arrow bounces a lot
[11:08]	Sindra Aquila:	that ball just might work
[11:08]	Andrew Linden:	xstorm, what region were you able to crash? What is the region name?
[11:08]	xstorm Radek:	it seems to kill the physics my making like a malform shock wave in physics and temp timing
[11:09]	xstorm Radek:	All of them
[11:09]	xstorm Radek:	seems even Agni can crash do to this
[11:09]	SHAMUS Carter:	there was a noticable ripple like aftershock after allthe phasyals were gone
[11:10]	SHAMUS Carter:	in the lag
[11:10]	Andrew Linden:	I was wondering what one specific region was crashed so I could look it up in the crash reporter tool.
[11:10]	xstorm Radek:	look at Trendone
[11:10]	xstorm Radek:	i crashed it more then Oatmeal 7
[11:11]	SHAMUS Carter:	how meny times did that sim crash it was prity carzy what the phsycs as doign to that pore sim
[11:11]	xstorm Radek:	i even was working on a film to show what leeds up to a crash
[11:11]	xstorm Radek:	one problem i crash film stops befor crash lol
[11:12]	SHAMUS Carter:	fraps to the resque xstorm
[11:12]	Andrew Linden:	yes, Trendone was definitely crashed in the physics system
[11:12]	Andrew Linden:	Falcon will be interested.
[11:12]	xstorm Radek:	but i did get the ball effect that happens
[11:13]	SHAMUS Carter:	yah that ball like efect is very new
[11:13]	Andrew Linden:	I wonder if he has already looked into it. He's been on top of new crash modes in Havok7.
[11:13]	xstorm Radek:	i think the trigger is not so much the physics
[11:13]	SHAMUS Carter:	prims self intersecign and coliding in a small space fdoming a huge ball
[11:13]	xstorm Radek:	but how the server handles the data coming in to it
[11:14]	SHAMUS Carter:	a beep load of colisions in a shot amount of timre
[11:14]	Andrew Linden:	well, the crash was in the LLhkPerformanceManager class, which kicks in special code when physics lag gets very bad.
[11:14]	xstorm Radek:	that is why i think its making a ball
[11:14]	Simon Linden:	That's the classic physics / collision overload problem. We have code that tries to deal with it, but it definitely should be exercised
[11:15]	xstorm Radek:	the next step it is to take is not taking place do to the large number of prims
[11:15]	xstorm Radek:	so it forms a ball
[11:15]	xstorm Radek:	till the software has time to take the next step
[11:16]	Andrew Linden:	a "ball"? You mean a sphere? or the pile turns into a ball shape?
[11:16]	SHAMUS Carter:	the pilee forms a huge sphere
[11:16]	SHAMUS Carter:	all self intersecting
[11:16]	xstorm Radek:	the wave spike is the software trying to cap the data coming in to it
[11:17]	SHAMUS Carter:	forim insteifn art like paterns
[11:17]	Andrew Linden:	that is possible. The Havok "penetration resolver" algorithm might let the pile form a ball while it tries to figure out how to move them apart.
[11:18]	Andrew Linden:	The ball shape is probably an artifact of the algorithm that is trying to extract the overlapping objects.
[11:18]	xstorm Radek:	yes and thats whats triggering the wave type spike in lag
[11:19]	Andrew Linden:	Well, I'm sure Falcon will try to figure out how to make the performance management code more stable.
[11:19]	Andrew Linden:	Dunno if he's given up on this particular crash mode yet or not. I'll ask him about it.
[11:19]	xstorm Radek:	is there not a better way to handle a large spike of data like that ?
[11:19]	Ardy Lay:	He is rezzing large numbers of identicle prims at one coordinate and setting them all physical at the same time?
[11:20]	Ardy Lay:	xstorm, is that what you are doing?
[11:21]	xstorm Radek:	yes Ardy it seems a group of griefers have been setting up Invise mines you step on to triggers a sim crash
[11:21]	Andrew Linden:	The spike of lag includes a large amount of "collision points", which burst the memory consumption very high.
[11:21]	Andrew Linden:	It is possible, perhaps, to cause the system to run out of memory... unless we watch it and cap.
[11:21]	Andrew Linden:	Which I think we do.
[11:21]	Andrew Linden:	But perhaps that cap isnt' working as intended.
[11:22]	xstorm Radek:	so if we can stiop this it will help end the griefers from doing this later on
[11:22]	Andrew Linden:	yes
[11:22]	Ardy Lay:	That is also going to generate a large volume of object update traffic to the viewers witnessing the event. Another workload to consider.
[11:22]	Andrew Linden:	however, if it is a current crash mode of Havok4 then its existence won't block the havok7 project's deploy schedule.
[11:23]	xstorm Radek:	i only found out about this by error when i lost my Invise prim and seen some one step on one and the sim crashed
[11:23]	Ardy Lay:	"Invise"?
[11:24]	xstorm Radek:	yes the prim we use on our avatars
[11:24]	SHAMUS Carter:	inviso
[11:24]	xstorm Radek:	they use it to hide there griefer tool till some one steps on it and it gets trigger
[11:25]	xstorm Radek:	so the griefer do not need to be in the sim
[11:25]	Andrew Linden:	Simon, I saw a rumor that some of the multithreading work you did for avatar region crossings will be delayed (not in server-1.40)
[11:25]	Andrew Linden:	What's up with that?
[11:25]	Simon Linden:	Yeah, we just decided to turn it off yesterday
[11:26]	Simon Linden:	During code reviews we looked deeper into our LLSD class (the same one that's in the viewer) and it's really not thread safe
[11:26]	SHAMUS Carter:	sim corsing on aditi were funcky yesterday
[11:26]	Simon Linden:	Not just in the 'multiple threads use the same data' case,
[11:26]	SHAMUS Carter:	i had an out of av esperence
[11:26]	Simon Linden:	but multiple threads using LLSD at all
[11:26]	xstorm Radek:	yes all broken
[11:26]	SHAMUS Carter:	to mew it looked like i was in one spot but to somone eses i wazs in another
[11:26]	Slee Mayo:	sorry
[11:26]	Simon Linden:	So we need to fix that at a basic level before pushing a lot of LLSD procesing into other threads
[11:27]	xstorm Radek:	i was in a bad spot on Aditi with Shamus lol
[11:27]	Andrew Linden:	Ok, so not in server-1.40, but maybe 1.42.
[11:28]	Simon Linden:	Yes, I'm working to get the fix in and the threads turned back on for 1.42
[11:28]	xstorm Radek:	we did a lot of crossings and our avatars was not going with our cams
[11:28]	SHAMUS Carter:	or the av went but did not go client side
[11:28]	Simon Linden:	Did you see a lot of the rubber-banding?
[11:28]	SHAMUS Carter:	yah a bit of rubber banding
[11:29]	xstorm Radek:	it was like back in 2004 when there was that out of body bug lol
[11:29]	xstorm Radek:	body was still moving but cam stops
[11:30]	office hours	is half over
[11:30]	xstorm Radek:	if i remember right it was a timing problem back in 2008 that did it
[11:31]	SHAMUS Carter:	YEEEAAAAHH!!!
[11:31]	SHAMUS Carter:	but was fun trying to pilot blind not. had to be told where my av was as to mee it looked as if i was not moving
[11:31]	xstorm Radek:	lol
[11:32]	Simon Linden:	The basic rubber banding is caused by the viewer's prediction code ... it blissfully extrapolates your velocity into the new region, without consideration for any lag on the crossing or obstacles there
[11:32]	Simon Linden:	When you actually enter the new region and get an update from the server, you snap back to that position
[11:32]	xstorm Radek:	was no rubber banding just cam was not moving with avatar body
[11:33]	Simon Linden:	I haven't heard of that one before
[11:33]	SHAMUS Carter:	and the funy thing is my av was in the same sim just two difren points
[11:33]	Susie Chaffe:	that is a wonderfully clear explanation Simon....do you mind if you I use it my ferry notecards
[11:33]	xstorm Radek:	well you have now ;-)
[11:33]	reddot99 Republic:	andrew what the heck? :[11:33] Twisted Radar HUD V 7.1.11worn: Andrew Linden <212,43,33> Age: 2903 2002-06-04 No payment info Flags: flying, sitting on object, online, wtf this
[11:33]	Simon Linden:	Andrew, is the cam tied to physics other than the ray-tracing to set it's position without objects in between?
[11:34]	Andrew Linden:	No, think the camera is specified by the viewer.
[11:34]	Simon Linden:	Susie - sure, feel free to use and modify :)
[11:34]	Susie Chaffe:	ty :)
[11:34]	Andrew Linden:	The simulator doesn't really have much limitation as to where the camera is.
[11:34]	SHAMUS Carter:	to me it looked as if i was not moving at all just waling in place but to the other users i was walkign about walkign into stuf as i was pilotin g blind
[11:34]	Simon Linden:	fwiw there is some work to make that prediction code better on the viewer, but I'm not sure when we'll see it
[11:35]	Andrew Linden:	Oh heh. My language setting is messed up.
[11:35]	SHAMUS Carter:	had to be told where my av was and gided by voice to the sim crossing lol
[11:35]	Andrew Linden:	I hacked it when testing a fix for multi-line language settings.
[11:36]	SHAMUS Carter:	and on cosing i sudenly snapped to the posting where the other persion saw me
[11:36]	xstorm Radek:	Shamus and me will have some other tests to do
[11:36]	xstorm Radek:	we found out a lot of items are broken
[11:36]	reddot99 Republic:	speaking of cameras, are there any plans to allow a camera to base its position relative to a given key? as to make things that follow the camera work better?
[11:37]	reddot99 Republic:	erm, that rely on the camera following them, rather
[11:37]	Andrew Linden:	No plans on that reddot99, that I know of.
[11:37]	reddot99 Republic:	such as rc vehicles
[11:37]	Simon Linden:	So you'd fix the camera to a position relative to an object, not an AV?
[11:37]	Andrew Linden:	xstorm, you found a lot of content brokein in server-1.40?
[11:37]	reddot99 Republic:	as it is, camera control is very jumpy
[11:37]	xstorm Radek:	yes we did
[11:38]	SHAMUS Carter:	well the child prim bug
[11:38]	SHAMUS Carter:	that one should be showstopper
[11:38]	Andrew Linden:	child prim collision callbacks?
[11:38]	xstorm Radek:	yes its all do to the shild prim not sending data to the root prim
[11:38]	xstorm Radek:	child
[11:38]	SHAMUS Carter:	lookign up the jirra
[11:39]	reddot99 Republic:	turn passcollisions() on to fix that xstorm
[11:39]	SHAMUS Carter:	http://jira.secondlife.com/browse/SVC-5805
[11:39]	Touching an	unscripted child prim does not trigger a touch event in the object's root prim
[11:39]	Andrew Linden:	I think Falcon already fixed that child-prim bug. Dunno if it has been deployed though.
[11:39]	reddot99 Republic:	and pass touchs
[11:39]	xstorm Radek:	did not work reddot
[11:39]	reddot99 Republic:	it works everytime i use it
[11:39]	SHAMUS Carter:	that one bug breaks almost everyting in sl
[11:39]	xstorm Radek:	scrips did not transfer using mono
[11:40]	SHAMUS Carter:	as people have moved to more eficent scripting and stoped puting scripts in child prims on huds ect
[11:40]	xstorm Radek:	unless every prim has a script in it then they do not work on 1.40
[11:41]	xstorm Radek:	even all of hippo was not working
[11:41]	reddot99 Republic:	i normally use touchst and put the child prim behind an alpha layer to catch the touch events
[11:41]	Andrew Linden:	yeah that one sounds like a blocker. I'll do some legwork after this meeting to see if anyone is working on it yet.
[11:41]	Andrew Linden:	If not then I'll look into it.
[11:41]	SHAMUS Carter:	♥ Thank You!! ♥
[11:41]	xstorm Radek:	im finding more bugs every day
[11:42]	xstorm Radek:	:-S
[11:42]	SHAMUS Carter:	well thats what the beta grid is there for hehe
[11:42]	SHAMUS Carter:	i just hope they get stomped befor they hit angi
[11:43]	Andrew Linden:	I'll ask Oskar if he needs help. Maybe I'll pile on bug fixing in 1.40 for a while.
[11:43]	xstorm Radek:	oh yes if you can have Falcon look at Nyx linden office hour meeting last week the sim crashed i think its link to the same bug
[11:43]	xstorm Radek:	the crash happen on Agni
[11:44]	xstorm Radek:	it seems i found the same loop wave type in Agni server so its not new
[11:44]	reddot99 Republic:	is it possible to add a camera velocity setting to the camera parameters?
[11:45]	xstorm Radek:	but on agni prims go off map then return and start all over again
[11:45]	Andrew Linden:	reddot99, possible in theory. It would mostly be a viewer-side project.
[11:45]	reddot99 Republic:	it'd still need a bit of lsl code additions, wouldnt it?
[11:46]	Andrew Linden:	yeah, if you wanted to control the camera with LSL
[11:46]	Ardy Lay:	I saw that. Temporary physical prims just topped and hung or drifted slowly as if in microgravity. The did not time out but vanished from the region when touched by anyone.
[11:46]	xstorm Radek:	was not flea bussy working on a cam script one time ?
[11:46]	Simon Linden:	That looks like it would be a new constant as opposed to a new function, so it's a bit easier to do
[11:46]	reddot99 Republic:	theres a bunch of camera scripts out there
[11:47]	Skills Hak:	that sounds like it was selected ardy
[11:47]	Skills Hak:	by someone
[11:47]	Simon Linden:	There might be some interesting side effects however ... what kind of velocity limits, if you should stop at the edge of a region,n etc
[11:48]	Simon Linden:	But I can see that would be a nice low level feature for making videos
[11:48]	SHAMUS Carter:	seams like temp prims are lasting loger then normal and not temping out as fast
[11:48]	Skills Hak:	well the drift would be aclientside effect
[11:48]	Andrew Linden:	when an object slowly drifts East it is a sign that there was a NaN (Not a Number) injected into the physics engine
[11:48]	Skills Hak:	if you unselect it it would snap back into the original position
[11:48]	Ardy Lay:	Skills, if so, they had a large number selected and kept them selected for over half an hour.
[11:48]	Skills Hak:	bypasses autoreturn/temp too yes
[11:49]	Andrew Linden:	hrm... that "snap back on selection" sounds like dropped interestlist packets
[11:49]	Skills Hak:	yes there are clientside functions for that called physics interceptors / mass select
[11:49]	Skills Hak:	in certain viewers
[11:49]	xstorm Radek:	that can make problems
[11:49]	Skills Hak:	you can block phsyics in an n meter radius by mass selecting
[11:49]	reddot99 Republic:	yeah. those are mainly in selfcompiled veiwers. but it is very useful if your an estate manager]
[11:50]	Ardy Lay:	Only Bronson, xstorm Nyx and me were there at the start of the event.
[11:50]	Andrew Linden:	Hrm... the temp object reaper code probably needs some design tweaks.
[11:50]	Sindra Aquila:	this is used by griefers
[11:50]	xstorm Radek:	true
[11:50]	Sindra Aquila:	to stop arrows in flight
[11:50]	Skills Hak:	hmm odd
[11:51]	xstorm Radek:	yes i have seen that
[11:51]	reddot99 Republic:	it'd be hard to setup a filter to catch that rather than dragselection
[11:51]	SHAMUS Carter:	Yus !
[11:51]	reddot99 Republic:	as i doubt the veiwer makes any distinction how things are selected
[11:51]	xstorm Radek:	i did a test close to that in this office last week
[11:52]	reddot99 Republic:	or at least, doesnt send said distinction to server
[11:52]	xstorm Radek:	the temp removed them but rex seen it and trigger a event all over again to loop
[11:52]	Skills Hak:	see them drifting?
[11:53]	xstorm Radek:	yes
[11:53]	Skills Hak:	not an issue IMO though
[11:53]	reddot99 Republic:	yeah, turn off velocity interpolation and it fixes that issue
[11:53]	xstorm Radek:	but that can crash the server
[11:53]	SHAMUS Carter:	prims drift a lot in h7 dung high phsytcs load
[11:53]	SHAMUS Carter:	almost more so then h4
[11:53]	reddot99 Republic:	have you tested with velocity interpolation turned off?
[11:53]	xstorm Radek:	its like one of the key factors
[11:54]	xstorm Radek:	yes
[11:54]	xstorm Radek:	same problem
[11:54]	reddot99 Republic:	ok, so then its not the same bug as over here when they do the snapback
[11:54]	xstorm Radek:	even if the viewer can not see it the server is still running the steps
[11:55]	office hours	is almost over
[11:56]	Andrew Linden:	Ok, I guess there is some work to do in server-1.40.
[11:57]	Andrew Linden:	It isnt' clear to me how much of this discussion was about new bugs, or longtime bugs in SL.
[11:57]	Andrew Linden:	I'll hunt jira for new bugs that you created xstorm to see what you've found recently.
[11:57]	xstorm Radek:	ok thank you
[11:57]	xstorm Radek:	:-)
[11:58]	Andrew Linden:	I'm supposed to do some bug triaging in the next hour, and I suppose that work counts as such.
[11:58]	xstorm Radek:	seems i found some cars that hate sim crossings
[11:58]	reddot99 Republic:	most cars do, i thought?
[11:58]	xstorm Radek:	yes
[11:59]	SHAMUS Carter:	vechiles pluse sim crosign == messy always has
[11:59]	xstorm Radek:	the car i was in lifted off the ground and got up to 2000 m lol
[11:59]	Andrew Linden:	Well, with the rollback of the multithreaded region crossing optimization code I wonder how those vehicle crossings will be affected.
[11:59]	Ovaltine Constantine:	Wait hold the phone
[11:59]	reddot99 Republic:	was that intended for havok 4 or 7?
[11:59]	Ovaltine Constantine:	"rollback of the multithreaded region crossing optimization code"?
[12:00]	Simon Linden:	That new code has been turned off in 1.40 due to some problems we discovered
[12:00]	Andrew Linden:	Hrm... sounds unrelated actually... sounds like a physcis bug, perhaps penetration resolution kicking the vehicle far.
[12:00]	Ovaltine Constantine:	You haz plans to fix, yes?
[12:00]	reddot99 Republic:	that the replacement for the 1.38 monofix?
[12:00]	Simon Linden:	yar
[12:00]	Thank you	for coming to Linden office hours
[12:00]	Andrew Linden:	Well, lots of work to do in any case.
[12:00]	Andrew Linden:	I think I'll get started.
[12:01]	SHAMUS Carter:	♥ Thank You!! ♥
[12:01]	Simon Linden:	None of this has anything to do with the mono slowness or 1.38 fix
[12:01]	Ardy Lay:	Thanks Andrew
[12:01]	Sindra Aquila:	thanks and have nice time all
[12:01]	Simon Linden:	Thanks everyone for coming, I have to run
[12:01]	xstorm Radek:	thank you from all of us :-)
[12:01]	Andrew Linden:	See you all later. Thanks for coming.
[12:01]	reddot99 Republic:	whats the status on that mono loading bug anyhow?
[12:01]	Ardy Lay:	Thanks Simon
[12:01]	Simon Linden:	See you next time

Generated with SLog Wikifier