Quite old revision. The idea could still be useful, but needs to be done as an object level flags.

Recently I came across this, too, quoting @Brecht Van Lommel (brecht) here:

Just tested it. The issue is still there.

I can only say it may be a numerical issue, pin pointed the issue and seems to happen when switching from a 0.05 to a 0.06 Sun Size.
Also, this may not be relevant to the issue but, since this commit the sun size is clamped internally to 0.0005 https://developer.blender.org/rCb40cde55cb6ee375eb4e514bec7bb294d6beda04

Sure. Will do.

This is fixed in 3.0 and will be fixed in the next 2.93 LTS as well. Note that we no longer fix bugs in 2.92.

Thx checking, @Brecht Van Lommel (brecht)

Possibly fixed by rB283481643c56: Fix T89736: Cycles error with persistent data, displacement and motion blur, which was ported to 2.93.2 just today.

The analysis by @Philipp Oeser (lichtwerk) is correct. More control over the weighting would be a good feature to add, but that's outside the scope of the bug tracker.

Checked. 2.93.2 also suffers. 3.0 doing fine.

After some testing, I have verified that the definition of each property added in the .osl file is processed and added in metadata format inside the .oso file

I can repro in 2.93 (it is a bit hard/inconsistent though -- sometimes more than 3 render attemps were neccessary, also a framechange in between), but havent been able to get it to freeze in 3.0.
@Yegor (Yegor): could you check if the situation improves for you in a frash 3.0 build from https://builder.blender.org/download/daily/?

Still happens on the latest 3.0 alpha (Mac intel version: 3.0.0 Alpha, branch: master, commit date: 2021-07-24 20:25, hash: rBaa3212117471), often not the first time, but consistently after switching input passes a few times.

I checked, and it still doesn't work, a clue i spotted to when it's malfunctioning is that when separating by materials, some objects disapear or change their materials to none, like in this mercedez where the door seems to be with no materials, and it looks like it only happens on the objects that are related to tohers, like the ones that are mirrerored (mirrors, doors, parts of the chassi, etc.)

Here is an example for Manta generated VDB (the fire) overlapping with some high commercial VDB. I do not know the source app of that VDB.

It happens with Manta flow VBDs as well. I am actually experiencing this issue with my project with larger simulations. I just created this scene for the bug testing purpose.

Yes, we've the same problem.
I made a water drops seamless,

Just checked, and P2243 does not really make a difference in this particular case, so there is something else going on as well.

This seems to be partially solved by the combination of D11841 and D11597, but not entirely yet..

Can confirm. Also with Mantaflow but i think its also vdb?

I can reproduce a problem (not totally sure yet if this is the only problem here).
In a Debug build, I first isolated this by removing stuff one by one ended up with a single object and culprit material called r light0.
Then I noticed this material had luxcore Custom properties (including an ID pointer to a Nodetree). If I removed the Custom property, everything was fine again.
Then I noticed luxcore properties on the scene as well. If I removed the Custom properties on the scene (as opposed to doing it on materials), everything was fine again as well.
So this is your scene with just that one property removed (and it now renders fine for me)

Probably the same as T89596: Hair density texture map, low render performance in cycles -- which ends up being the same as T89598: Hair density maps increases hair children amount on reopening scene file (or rendering which will get slow)

Got the same error with 2.93.1 today.
"Cycles" just stopped rendering in overview.
GTX 1650 Sup

I don't think we should do the same trick for material preview as it is too obvious it is fake.

• Renamed get_preview_floor_material_name to preview_floor_material_name.

Some of the logic has been tweaked in the rB289a173d938f, which helped a lot to avoid huge speed regreesion in the Pabellon scene when fast GPU is used together with slow CPU.

Confirming it's still an issue on daily build "3.0.0 Alpha, branch: master, commit date: 2021-07-18 23:24, hash: 83883ae66ad3, type: release". It should be noted it only happens when using CUDA device + Optix viewport denoiser. CUDA device + Optix render denoiser works fine until you break everything through the viewport.

Thanks for your tips, Brecht.
Vector input: the idea is save vertical space in the node, but yes, in the 'normal' node sizes the horizontal space maybe is not enough. But the user can extend the node size easily.
Image datablock: currently I use the typicall image file interface for my nodes, that is not perfect but is very clossely to the use in the Blender texture node.