10.6.2 is out so lets rant about it

Wow, you're right (re:mesh template crash). I didn't even notice that yet (because I friggin' hate the new template setup). I don't even know what to say about that. Crap, I knew I should have waited to make the bootable copy of the last version, but I got impatient. That's really weird.

Missing mesh patches when re-opening comps? Is this the virtual patches having come to bite everyone in the butt like I knew they would? (Wild speculation at this point, and there is truly little satisfaction in being right on that).

This makes an error I had make a great deal more sense. I got an incomplete restoration error last night, but hadn't thought about it too much yet, because the composition was so simple and from the pre-release. I rebuilt it successfully, but this is still disturbing.

Thanks for reminding me not to update...

Am I imagining this, or did Apple release 10.6.2 a while back, then pull it again? Looks like they've re-released it with exactly the same bugs...

a|x

DO NOT UPGRADE TO 10.6.2

SERIOUSLY annoyed with the QC Apple team today 10.6.2 has broken iterators and we had a production application due for release this month!!

foward bugreport posted on the 19 Oct 2009!!

Problem ID: 7315086 Quartz Composer Iterator performance serious problem!

19-Oct-2009 03:58 PM Mark Rock: Please run example composition on attachment. Increase and iterations and watch performance DRAMATICALLY drop.

OpenGL driver monitor reports an a huge jump in resource on the 'CPU wait in User Code'

Quartz Composer version 4.0 (103.1)

ATIRadeonX1600

'bug.qtz' was successfully uploaded

19-Oct-2009 04:14 PM Mark Rock: Disabling colour correction on the images does seem to improve performance abit

Sorry guys, but I'm going to have to break rank and praise it. It's fixed the main bugs I was experiencing \o/

Not that they were QC related, but the horrible display corruption after a few hours (requiring regular reboots) seems to have stopped (hope I'm not speaking too soon here...) and our CIFS servers are now accessible again. Plus I hear that I'll finally get some QC openCL loving! :D

Wait.... the iterator has been breaking stuff all along, and has been slower. Now I'm confused :)

As a matter of fact, one bug I've never got around to filing; I can go try to open up some old comps of mine that iterate some projection environment stuff, like Ortho or FOV and they fail on restoration because of the iterator. That's not 10.6.2, that's another thing that has just been the "way it is". I think that would probably happen to me "New Land" composition that's in the repository. It's been awhile since I checked.

Nooooo. Come on. I'm way happy that some of your bugs have been fixed, but I'm going to say "just wait". Now I get to laugh my ass off when you start freaking because of all of the broken stuff, after your warm fuzzies wear off :) (I say this in good humor).

I might well be missing the point entirely, but if I iterate to select from an index and do so by asking for an iteration that asks for a non-existent item I will get a

Exception Type:  EXC_BAD_ACCESS (SIGSEGV)
Exception Codes: KERN_INVALID_ADDRESS at 0x0000000000000038
Crashed Thread:  0  Dispatch queue: com.apple.main-thread

because ...

function (__structure outputArray) main (__image a[9])
{
   var result = new Object();
   if (!a)
      return result;
 
   var number = [];
   for(var i=0; i<9; i++) {
      number.push(a[i]);
   }
   result.outputArray = number;
   return result;
}

I am asking for an iteration above and beyond the range specified in the JavaScript Patch and so QC does what it does when it cannot get what is being ostensibly asked for, it spits its dummy out :-).

Prior to getting a QC crash, I do find that performance diminishes somewhat.

I have also found that it is sometimes non-reproducible.

I have applied an Iterator 1 - 100 range to provide the Iteration count to your original bug.qtz and have not crashed the composition as yet. I think it might be the static count above the range specified in the JavaScript that is the real sticking point.

It was when I was just sliding the number range in your original composition [unmodified] and sticking at a non-available number that the crash occurred.

Might also be card specific - how nice if OS X could be more hardware agnostic.

Interesting to hear how well this update has affected the non nVidia Open CL blessed hardware users. I've run the item in Render, Debug and Profile mode with little except no image from structure to show that anything was wrong , namely blank empty space.

I'd think about changing the structure of the composition in any case, but that's just me.

Well, unless the Mouse was actually meant to be connected to any of the patches, nothing is wrong that I can tell from my research :-). No-one had commented upon the composition - unusually enough, it had kind of escaped my attention.

Oh, it works AOK - nice Visual Instrument too - another composition altogether :-)

Hmm, it's opening up ok for me now, whereas it has been having problems (even last night!). Uhm. I give up. For now. (smokris did fav it on Vimeo so it didn't go totally un-noted... it's one of my personal favs of mine). I guess I should just be glad it's opening up and not making an exception, at least some of the time.... but I would rather have a bug that happens "all of the time" than willy-nilly.

Well, the app I'm writing works fine (well, no more buggy and slow than it was yesterday at least ;) so I'm happy. I'll probably still be working on this when 6.3 comes out, so it's no issue as long as that works I reckon.

First job tonight: update my home mac to 10.6.2.

Maybe it's just me-

If it wasn't breaking, and now it is, that's a bug, no matter if it can be re-programmed to not crash.

At the same time, the organization of this qtz doesn't make sense (to me at least). What are you trying to accomplish by this? If you want to flip pictures, you can just use a multiplexer, and no javascript or iterator is needed. If you wanted to load all of the images at the same time, you would put the images+multiplexer inside of the iterator. I would avoid the javascript unless it was absolutely needed, and it doesn't look as if it is.

For flipping through an image at a time, this is an inefficient setup. Maybe this has been scaled back to show the bug, but in that, it begs the question of what exactly this is supposed to be doing, and why employ javascript and iteration to load a single image at a time (?)

At least SOMEBODY is stoked :)

Just rerun this in Leopard - fresh download of bug.qtz , get same down grading of performance when asking for a non existent member of an instantiated structure. Do not get a crash. FPS is lower on PPC G5 GeForce [6.7 low point] iMac Intel nVidia 9400 fps is about 15+ when both SL and Leo QC are running bug.qtz at 100 iterations. In fact for Iteration 7 PPC is almost 2 * fps - weird. SL fps approx 2/3rd of Leo PPC fps. Baffling.

Interesting how both SL and Leo QC give 0 Iteration as nothing at all, and structure member 0 is thus Iteration 1.

gtoledo3 wrote:

Maybe it's just me-

If it wasn't breaking, and now it is, that's a bug, no matter if it can be re-programmed to not crash.

No, it's not just you, it is a bug, so far as I can tell, but it isn't reliably reproducible as a crash and that is without re-programming the example.

The performance deterioration is most puzzling , although I suppose it could just be a bit misleading, nothing to display, so why add on frame rate wise to the workload, one blank square being much like any other. Just guessing.

That deterioration is not SL specific though.

In performance terms it makes some sense, doesn't it, for us to to not have QC spit its dummy out and crash whenever we haven't finally associated all essential assets to a called upon structure?

However, this is the sort of behaviour, crash wise, that we can find when .dae or .mp3 or .mov is not correctly connected, present or the correct file path is not used, so I don't find it too surprising. They usually just crash , whereas this example might or might not crash.

I dont think they ever released 10.6.2 previously, however some people do get early access to pre-release versions and may have been talking about them. I am generally too paranoid about the Apple NDA to do that, so Im not much good at talking about the history.

Thanks for the input as ever cybero but i'm not referring to a drop in performance from non existent members, but a MASSIVE drop in fps when running this composition in 10.6.1 and 10.6.2.

we've tested this on 3 separate computers (dare not on any more)

Macbook pro (with ATI card) 10.6.1 60 fps for 5 iterations

in 10.6.2 5 fps for 5 iterations

Macbook pro (brand new Unibody) 10.6.1 60 fps for 5 iterations

in 10.6.2 12 fps for 5 iterations

gtoledo3 wrote:
At the same time, the organization of this qtz doesn't make sense (to me at least). What are you trying to accomplish by this? )

gtoledo its not meant to make any sense its stripped down to its simplest form to reproduce the dramatic drop in fps from running in 10.6.1 and 10.6.2

OK, got it, you didn't mean going beyond the number range for which you have assets.

None the less, when further forcing the problem by raising the number of iterations I did get some odd results - for Iteration 7 PPC is almost 2 * fps of the SL 10.6.2 machine - baffling. SL fps approx 2/3rd of Leo PPC fps.

You're definitely on to something itsthejayj in regards of the Iteration increase resulting in a marked decrease of performance on 10.6.2 to 10.6.1 and also in relation to 10.5.x [60 fps] too :-) - you're getting 12 fps on a 9400M for Iteration 5, I'm getting 30 to 19 fps on that Iteration and asset on 10.6.2 9400 iMac, pretty weird?

I just changed that Iterator Input Splitter to a Number only item 1 - 19 and then fed that through to set the Iterations and some improvement in fact Iteration 5 gives 30 to 35 fps - not 60 fps admittedly, but a slight bump up from when it was set as an Index and gave 20 - 25 fps on Iteration 7

Gah I've ended up too tired after work again today to explore the mesh issues in more detail.

Still with free versions of unity3d and unreal development kit now available, I am seriously considering abandoning QC for the realtime stuff I tinker with.

Just wondering why 10.6.2 should fix QC4? I thought QC was only available as part of either the iPhone SDK or Mac SDK, in which case wouldn't any QC updates be included with an SDK/xcode update rather than an OSX update?

QC is actually 2 pieces: the editor App (part of dev tools) and the QC framework (inside /System/Library/Frameworks/Quartz.framework) -- OS X updates can and do update the QC framework (which sometimes fix bugs in the QC framework itself, such as when attaching an LFO to a Sphere's diameter would crash in early versions of Leopard.)

The editor app, however, isn't updated until you grab an updated SDK. That's how editor-app bugs get addressed.

Even further, there are other frameworks besides the Quartz.framework that various patches/functions may call upon as well, that are part of the regular OS.

Yeah, I was simplifying a bit: specifically, there's the editor app (QC.app) which is dev tools, and then a zillion frameworks, which are all updated with OS updates (QC.framework being a huge portion, but also JavaScriptCore (for JS - part of WebKit), Quartz (coreImage stuff), OpenGL, OpenCL, QuickTime/QTKit, IOKit, ImageIO (part of Quartz?), Accelerate, CoreVideo, Foundation, AppKit, and all their dependencies. All of those pieces are updated with OS updates, it's just the app portion (which is essentially a thin chunk of code that glues together all the other pieces together) that isn't.

The code that draws noodles and patches isn't in the QC app, the code that loads plugins isn't in the QC app, the code that renders anything isn't in the app... the only things in the app are the Patch Creator, the Template thing, and a lot of logic to tie everything else together.

UDK, yeah, so am i, seriously. However, i doubt there's ANY midi support in UDK, nor ArduinoSerial, nor DMX ... that just leaves an awesome 3D engine... which i doubt it would ever be as versatile as QC (art-installation wise)

Have you tryied UDK and UNITY in depth ?

The UDK supposedly has tcp / ip available to scripts, which would let you implement OSC yourself.

Not sure how the Unity Boo implementation likes custom py modules, but it certainly has some osrt of raw networking as well.

So, you might need to transate any serial stuff to OSC or similar in some other language (py / processing), but it should be doable.

Frankly, if I were to boot into Windows anyway, I'd just as soon play with vvvv as UDK, but that's just me.

Ok! thanks for that info.

Hi! Have you noticed of this? on math expresion, the caracter [] doesn't work anymore, may be you have min(Value[1],Value[2])

Also opening some old files, says: [NSPropertyListSerialization failed with error: "Conversion of string failed. The string is empty."]

and some others old files show repeated published splitters but there is only one splitter published.

Wired o_O

Someone made a Unity-max/pd bridge which I tried a while back and got working, which gives me a way to use midi or OSC with Unity. Someone also did some wonderful work on a 3D ocean simulator for Unity which I was extremely impressed with, and I liked the development UI and lighting. But for the stuff I wanted I really need the pro version and I didnt have the money so I put that on hold.

UDK I only started looking at yesterday, really love the engine, not so keen on the development UI or the documentation, and am concerned that I might need something like Maya to build suitable assets.

I'm hoping that the free versions of these technologies will make it a bit more likely that other people will write more scripts etc for being able to send midi etc to them.

Realistically I'll probably stick with QC for a lot of what I do, Im just in slight despair at how Apple have dropped the ball a bit in the last 6 months. Plus I'm struggling to retain hope that I'll ever get to see IK, skeletal animation etc which I really need to realise my virtual puppeteering dreams.

I'll probably wait and see how the new Blender turns out, as its Game Engine has potential, just need a saner UI. Derivative Touch on Windows is another option I toy with from time to time but its jargon & UI tends to melt my brain.

massive problems in terms of FPS here, under 10.6.2. i have several .qtz patches that show 2-3 FPS (!) in quartz composer 4, almost freezing the entire OS grafic environment if previewed in finder. they were running smooth under 10.6.1 ...

MBP with GeForce 8600M GT (256 Mb)

can i guess that you have images in an iterator.

so im just jumping in here the stuff thats broken seemed to have been all ready broken. despite the pre existing bugs with open cl crashes and what not is it safe to upgrade to 10.6.2 ? i didn't really see any discussion of this on the qc dev list. is there anyone actually using 10.6.2 without problems ? just wondering some problems are user specific some and some are apple fault. thanks i have yet to upgrade. waiting for confirmation.

Mesh Filter is broken, so if you loved that, don't bother with 10.6.2. Some other CL/Mesh stuff is broken too.

Otherwise, a few bugs were fixed internally, and some GPU driver bugs were fixed. I've been using it on my non-dev machine without any problems (other than meshes crashing).

yes, but its strange because i have other patches with images in iterators that still runs with zero problems. anyway i can confirm that all was ok before the upgrade to the 10.6.2.

Same problem here. Some compositions I used a few days ago before the 10.6.2 update don't work anymore. I see the 2-3 FPS in patches using Directory Scanner and Image Structures.. for instance this file:

http://kineme.net/composition/smokris/FileListImageStructure

(QC gets slow only trying to switch from the editor to the viewer)

However... I did restore one of my macminis from Zero... formatted disc, installed 10.5.6 from the Install DVD. Things were fine on the mini until I updated to 10.5.8 via Software update. I get the same problem now in a mac mini with 10.5.8. So I guess there was something else updated besides the System version.... or both system versions got the same problem.

ayalacortes wrote:
http://kineme.net/composition/smokris/FileListImageStructure (QC gets slow only trying to switch from the editor to the viewer)

This particular composition is an exception --- as soon as the Directory Scanner has results, the Image Importer loads all of the images during a single (reaallllly long) frame. After that, execution should be fast (it is, at least, on the 10.6.2 machines I've tested it on).

Thank you. I understand the exception.... however it takes about 3 minutes to start reacting and ~30 seconds to switch from editor to viewer. I mentioned this file I had open when I wrote the reply but some other files are also showing extremely slow fps... this one for instance:

http://www.memo.tv/files/memotv/Carousel.qtz

(this one loads only a few images)

I'm sure both files performed smoothly before the software update. I also wrote about it in the QC dev list and so far one person confirmed the problem.

Perhaps an isolated problem. But now I can confirm the same in my 10.5.8 system.

well.... connecting some hints here and there... plus reading a bit here: http://bit.ly/1x6hW8 I have solved my own problems under my own risk simply restoring an older Quartz framework I found via time machine. So my version of QC works just as it was before with a restored old framework as shown in the image. Perhaps it is a good idea to have a "Select Preferred Framework" option somewhere in the QC preferences... or as an option in KinemeCore :-)

Preview	Attachment	Size
	Screen shot 2009-11-19 at 08.40.12.png	28.04 KB

Smart idea about the QC preferences, ayalacortes :-)

Unfortunately, by the time KinemeCore is loaded the QC framework is already loaded (it's what loads kinemeCore, after all), so there's no way to unload the current QC framework, and then reload a different selected framework.

So could one have a pre-boot intervention to bespoke the ongoing environment without KinemeCore being involved ?

Physically replacing a framework is of course, one thing and it can work, but choosing a framework as if were an enabled plugin - rather different.

Perhaps it could be set as a next boot preference in QC?

BTW - did Aurora.qtz ever really work in iTunes like it did in QC? I found one needed to reduce the fog and light up the environment a little before I even got much noticeable action from the OpenCL Mesh in QC though. It's a little bit like too many options, speaking of which, compositions with options seem a little more likely to crash iTunes for some reason.

So just to confirm, has everyone filled out a bug report at bugreport.apple.com regarding all the issues come across?

You need an ID to login; btw it's free to join the Apple Developer Connection : http://developer.apple.com/mac/

Along with a bug report, it's also possible to send Apple the offending .qtz composition as an attachment with the report.

[we can't just rant here about the problems in 10.6.2; Apple should be made aware of the issues]

to do that, the tool would have to swap the framework in /System/Library/frameworks/Quartz.framework/... which is possible, but ugly. it'd also require distributing the old version of the framework, which is illegal (only apple is authorized to distribute apple code)

Yeah, for fun, read the agreement that you agree to when you install Xcode. Then, go poop yourself...lol...

I've hit massive slowdown too, and yep its 10.6.2's QuartzComposer.framework that is to blame: reverting to the 10.6.1 one fixes the performance issue for me.

To file a bug report, I need a sample composition / apple stock composition that demonstrates the issue. Has anybody been through this already? Openradar, copy'n'paste...

3D-Noise from the dev-folder is their own qtz, and it shows the bug well because of all of the levels of iteration.

I should clarify... it's from the 10.5 Developer examples.

just downgrade my system to leopard (10.5.8) and all seems perfect. i really don't want waste my time with this shit. anyway i had major FPS slowdown problems with several patches based on two zugakousaku compositions called "fishcrowd" and "kaboku" experiencing 2-3 FPS and a massive freezing of the entire 10.6.2 environment when the compositions were previewed in finder. they were run smooth as always under 10.6.1. you can download these patches from his website and test. http://www.zugakousaku.com this is my experience. hope this helps.

I downgraded my QC to 10.6.0 version, was getting no freeze up as you and I and many another will no doubt get with the QuickLook in Finder on .qtz files with iterations a plenty in 10.6.2.

Of course, this still doesn't address the OpenCL problems, but it does make something rather more serviceable, QC 4.

Until the next update •~

Have just downgraded my QuartzComposer.framework. Just tried running those zugokousaku examples and with the original Snow Leopard QuartzComposer framework on a 10.6.2 32 or 64 bit boot , get a good regular 60 fps.

OpenCL stuff lasts longer too, still crashes more often than not :-)

so basically 10.6.2 sucks with QC ? the only way to fix this is to use an older framework ?

dust wrote:
so basically 10.6.2 sucks with QC ? the only way to fix this is to use an older framework ?

Succinctly put - yes, or so it would seem.

I have found that downgrading the framework produces a much more workable QC 4 all around, but much at the expense of audio input dynamically generated OpenCL meshes within Tunes in particular.

AudioTools crashes and plays badly with OpenCL under the older framework. Ah well, it's not a perfect world by any means.

Till the next update, I guess •~

I have reported the crashing bore under the newest framework to Apple. That composition had two elements in particular that provoked problems, OpenCL and Iterations of these rendered assets.

I shall also be dovetailing my proof of observations concerning Iterations improving under the older framework [ 40 - 60 fps instead of 5 - 15 fps] in due course, along with those concerning .

Right now, I'm logging in from my PPC, as AHT is being run , just to double check that there isn't something wrong with CPU,GPU or RAM in particular.

I doubt that anything is wrong at all with the hardware, it just clears the ongoing line of enquiry .

so i have not yet gone to 10.6.2 thanks to the warnings so thank you guys for this. so maybe this has been discussed but what features is the updated framework giving to QC or QC hosts ? just trying to figure out why they would change it just so i can weigh the justification of the fps expenditures involved. maybe that didn;t come out right. what new features did they add and are they worth upgrading at the expense of loosing functionality. to me the mesh things and the iterator seem pretty important so im not sure what addition could persuade me to upgrade. maybe i will try on my macbook seeing the mesh doesn't work anyways. it would seem to me that the framework would be updated for a reason. or is it new things dependencies of the qc framework that are causing problems or the framework itself ? seems to me the framework because if it where things dependent on the framework then using the old one would cause the same problems. just trying to figure out what was fixed and why they would change something that relatively worked bug free ? i mean a kid in my class i have been helping with QC upgraded to 10.6.2 and he couldn't even open QC ?

Well the early alpha of the new Blender shows promise with the user interface, but I think it will take a while before it is fully functional.

Meanwhile I got an OSC script working with Unity and I also had cause to get MAX for Live, so I am taking a holiday from QC until hopefully Apple sort all the problems out.

I also like the cross platform nature of Unity, as well as the speed and lighting, especially that shadows dont slow it to a crawl and work as expected when more than one light is used. If I get on real well with unity then I may just end up using QC for effects via VDMX, although a future kineme3D plugin that supported animation in interesting ways would probably bring me rushing back :)