particles, background, etc... are moved or look different #43967

New Issue

mathieu menuet · 2015-03-11T23:09:25+01:00

mathieu menuet commented

2015-03-11 23:09:25 +01:00

System Information
All vendors

Blender Version
Broken: 2.74
Worked: 2.67 at least, maybe more recent too. Can try tomorow

Short description of error
Particles are impredictable. Setting a nice looking seed may be destroyed by random changes after changing the sample amount or visible layers, whatever. Couldn't reproduce a clean list of steps, doing twice the exact same steps doesn't lead to bug everytime. Scenes made previous 2.74 look completely different. It makes retouching old shots impossible without rerendering everything...

Exact steps for others to reproduce the error
Can't provide commercial scenes, but it works on many scenes found on the web. Fo rexample, the barcelona scene used for benchmark: http://www.emirage.org/2013/04/24/free-download-archviz-project-pabellon-barcelona-3d-scene-v1-2-updated/ . Render it on 2.67 and 2.74 it looks completely different (sky, trees, etc..)

**System Information** All vendors **Blender Version** Broken: 2.74 Worked: 2.67 at least, maybe more recent too. Can try tomorow **Short description of error** Particles are impredictable. Setting a nice looking seed may be destroyed by random changes after changing the sample amount or visible layers, whatever. Couldn't reproduce a clean list of steps, doing twice the exact same steps doesn't lead to bug everytime. Scenes made previous 2.74 look completely different. It makes retouching old shots impossible without rerendering everything... **Exact steps for others to reproduce the error** Can't provide commercial scenes, but it works on many scenes found on the web. Fo rexample, the barcelona scene used for benchmark: http://www.emirage.org/2013/04/24/free-download-archviz-project-pabellon-barcelona-3d-scene-v1-2-updated/ . Render it on 2.67 and 2.74 it looks completely different (sky, trees, etc..)

mathieu menuet commented

2015-03-11 23:09:25 +01:00

Changed status to: 'Open'

mathieu menuet commented

2015-03-11 23:09:25 +01:00

Added subscriber: @bliblubli

mathieu menuet commented

2015-03-11 23:10:01 +01:00

Some shots to show the differences:

Some shots to show the differences: ![274.jpg](https://archive.blender.org/developer/F149954/274.jpg) ![267.jpg](https://archive.blender.org/developer/F149955/267.jpg)

lopata commented

2015-03-12 00:12:02 +01:00

Added subscriber: @lopataasdf

ronan ducluzeau commented

2015-03-12 11:13:37 +01:00

Added subscriber: @zeauro

ronan ducluzeau commented

2015-03-12 11:13:37 +01:00

It is known that particles have several problems.
Rewriting them is todo at long term. Alembic caching made for gooseberry should help to solve this problems.

Currently, a workaround can be to replace object duplicated by a particle system by a plane in particle system, to make duplicates real, to make duplicated planes single users, to join them and use them as a duplifaces parent.

It is known that particles have several problems. Rewriting them is todo at long term. Alembic caching made for gooseberry should help to solve this problems. Currently, a workaround can be to replace object duplicated by a particle system by a plane in particle system, to make duplicates real, to make duplicated planes single users, to join them and use them as a duplifaces parent.

Sergey Sharybin self-assigned this 2015-03-12 13:16:31 +01:00

Sergey Sharybin commented

2015-03-12 13:16:31 +01:00

Will commit fix for the sky once the git is open for commits.

Particles issues we can't really totally avoid, this is a fragile area which is totally broken and fixing some aspects of it might cause regressions in how the scene looks. It is an on-going project to bring some better replacement for particles.

Will commit fix for the sky once the git is open for commits. Particles issues we can't really totally avoid, this is a fragile area which is totally broken and fixing some aspects of it might cause regressions in how the scene looks. It is an on-going project to bring some better replacement for particles.

Sergey Sharybin removed their assignment 2015-03-12 16:10:02 +01:00

Sergey Sharybin commented

2015-03-12 16:10:02 +01:00

Added subscribers: @LukasTonne, @Sergey

Sergey Sharybin commented

2015-03-12 16:10:02 +01:00

Fix for the sky is now in git now.

As was mentioned in previously, trees are likely caused by changes in particles. Not sure we'll be considering it a bug for until we've got rock-solid new particle system. @LukasTonne, mind giving feedback on this?

Fix for the sky is now in git now. As was mentioned in previously, trees are likely caused by changes in particles. Not sure we'll be considering it a bug for until we've got rock-solid new particle system. @LukasTonne, mind giving feedback on this?

Sergey Sharybin commented

2015-03-17 10:59:37 +01:00

Changed status from 'Open' to: 'Archived'

Sergey Sharybin closed this issue

2015-03-17 10:59:37 +01:00

Sergey Sharybin self-assigned this 2015-03-17 10:59:37 +01:00

Sergey Sharybin commented

2015-03-17 10:59:37 +01:00

Traced down the commit which caused the change, and it appear to be 0844451, which is a bug fix for threaded depsgraph. Old code indeed was incompatible with the threaded evaluation, new one changed random number distribution a bit, which is quite tricky to avoid.

Wouldn't consider this a bug, current particles are always fragile and there's work happening about upgrading them. Also changing the distribution again now would break files saved with blender >= 2.70.

So thanks for the report, but will consider it a TODO related to the particles recode project.

Traced down the commit which caused the change, and it appear to be 0844451, which is a bug fix for threaded depsgraph. Old code indeed was incompatible with the threaded evaluation, new one changed random number distribution a bit, which is quite tricky to avoid. Wouldn't consider this a bug, current particles are always fragile and there's work happening about upgrading them. Also changing the distribution again now would break files saved with blender >= 2.70. So thanks for the report, but will consider it a TODO related to the particles recode project.

Lukas Tönne commented

2015-03-17 11:16:40 +01:00

Should be fixed now by a0a6cb129e and subsequent bf8ea6b60e

Should be fixed now by a0a6cb129e9df46e80dec891b535992e773901af and subsequent bf8ea6b60e69e6dd77d94c66cbcbab2a8e0b73f8

Lukas Tönne commented

2015-03-17 11:16:49 +01:00

Changed status from 'Archived' to: 'Resolved'

Sergey Sharybin commented

2015-03-17 11:32:05 +01:00

Lukas, distribution changed comparing to pre-mentioned fixes, but it's still not the same as in original file from 2.69. So you fixed the jitter thing, but it doesn't mean render of the barcelona file will be the same now as it's original.

mathieu menuet commented

2015-03-20 22:16:50 +01:00

after compiling and testing buildbots, this is what I get on the same file, same blender version (hash 62b31a6), just start Blender, open it and render. Nearly every time a different render :

after compiling and testing buildbots, this is what I get on the same file, same blender version (hash 62b31a6), just start Blender, open it and render. Nearly every time a different render : ![4.21_min_master_test_1.png](https://archive.blender.org/developer/F153476/4.21_min_master_test_1.png) ![4.27_min_master_test2.png](https://archive.blender.org/developer/F153478/4.27_min_master_test2.png) ![4.36_more_random_master.png](https://archive.blender.org/developer/F153472/4.36_more_random_master.png) ![5.49_min_master_test3.png](https://archive.blender.org/developer/F153474/5.49_min_master_test3.png)

mathieu menuet commented

2015-03-21 20:21:50 +01:00

note also that the background is again without cloud in RC2

Sergey Sharybin commented

2015-03-21 20:31:28 +01:00

For the background you need to re-save file in previous release (to be more precise in 2.74.3 and lover), otherwise the versioning code does not have chance to run reliably.

For the particles issues, is it enough to simply hit F12 several times in order to see difference in distribution?

For the background you need to re-save file in previous release (to be more precise in 2.74.3 and lover), otherwise the versioning code does not have chance to run reliably. For the particles issues, is it enough to simply hit F12 several times in order to see difference in distribution?

mathieu menuet commented

2015-03-21 21:34:17 +01:00

ok
For the particles, it works really randomly. I restarted blender every time in my case.

ok For the particles, it works really randomly. I restarted blender every time in my case.

mathieu menuet commented

2015-03-21 21:42:53 +01:00

testes with several times hitting F12, no bug. I tried again with restarting blender, couldn't reproduce either after 8 renders. Yesterday it happened 3 times in a row. Weird. I guess we can use 2.73a for scenes with particles until the new particle rewrite comes. Don't loose to much time for that, better invest it on new solid code.

mathieu menuet commented

2015-03-21 21:49:28 +01:00

I guess it also happens faster at work because we have 200+ processors. The bug I reported last year that also appeared randomly ( https://developer.blender.org/T40313 )was easy to see at work and nearly impossible to reproduce at home. So just close it, we can render on 2.73a, it will do for the mean time.

Sergey Sharybin commented

2015-03-22 08:16:52 +01:00

You should be able to use 2.74 because particles changes were made after the branching and would not be backported to the release branch. At least i don't expect such a regressions in 2.74.

And it totally worth investigating such a randomness of behavior in current master, but seems for that we'll some investigation from you as well (since you're the only one so far who experienced the issue and could try isolating the issue ;)

You should be able to use 2.74 because particles changes were made after the branching and would not be backported to the release branch. At least i don't expect such a regressions in 2.74. And it totally worth investigating such a randomness of behavior in current master, but seems for that we'll some investigation from you as well (since you're the only one so far who experienced the issue and could try isolating the issue ;)

mathieu menuet commented

2015-03-22 10:43:24 +01:00

good for 2.74 not getting the changes.
I really would like to help, but I don't see how I can help further. The bug simply doesn't follow any easy-to-understand steps, it appears to be purely random. It surely isn't, maybe it's a thread concurrency problem, but I'm not good enough to see it. I also have limited time and I already use some of my own time because I've many other things to do at work.
As I said, most freelancers and small studios will most certainly never experience those problems unless Intel starts to sell affordable 16 cores CPU next year?
As you know it's also not the only problem with particles. We all know it has to be rewritten. You have limited resources, it would be more beneficial to use the limited money (even if it's many times higher than usual) given by Gooseberry on the new system, than loosing it on something that will be trashed. It looks like every new block of code in the current system makes 10 other parts unstable. The amount of time you programmers and we artists loose because of that code would be pretty high if we summed it up and it's more every day we stay on that old code. Please leave it as soon as possible.

good for 2.74 not getting the changes. I really would like to help, but I don't see how I can help further. The bug simply doesn't follow any easy-to-understand steps, it appears to be purely random. It surely isn't, maybe it's a thread concurrency problem, but I'm not good enough to see it. I also have limited time and I already use some of my own time because I've many other things to do at work. As I said, most freelancers and small studios will most certainly never experience those problems unless Intel starts to sell affordable 16 cores CPU next year? As you know it's also not the only problem with particles. We all know it has to be rewritten. You have limited resources, it would be more beneficial to use the limited money (even if it's many times higher than usual) given by Gooseberry on the new system, than loosing it on something that will be trashed. It looks like every new block of code in the current system makes 10 other parts unstable. The amount of time you programmers and we artists loose because of that code would be pretty high if we summed it up and it's more every day we stay on that old code. Please leave it as soon as possible.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

particles, background, etc... are moved or look different #43967