@Richard Antalik (ISS), thx for getting back.
Looks like this was indeed fixed, see T73770: Mantaflow is unable to bake fluid simulations on non-ASCII file paths, rB4453509d83dc: Fix T73770: Mantaflow is unable to bake fluid simulations on non-ASCII file…, rBb8af5e10620f: Fluid: Updated manta pp files.

Sorry, I haven't checked. I was quite certain, that this issue was dealt with recently.

@Richard Antalik (ISS): that commit is two years old (I just posted this as a reference)
Report is from a version two weeks old and on a 64bit windows... Are you sure this is fixed?

This should be resolved in 2.83 by rBb0af44fa4d7a: Fix T50630: Fluid fails on win32 w/ unicode paths

Just stumbled upon possibly related:

• rBb0af44fa4d7a: Fix T50630: Fluid fails on win32 w/ unicode paths
• T53159: OpenVDB cache doesn't support unicode characters in filename

In T74155#880378, @Mansur Şamil Güngör (msamilg) wrote:

I'm having the exact same problem. Other simulation caches are writing fine on the disk, but not Mantaflow.

Frames 88-89.

There should be similar report already, which I can't find. I will confirm this in case I don't remember correctly.

@Yuriy Thanks for simplified case. Can you also point out frames to look for? I am looking at cube render and don't see anything extraordinary

LGTM :)

Testing a bit more:

Blender seems to hang here:

I agree, that it is changing data, but this data is unrelated to simulation.
So I hope it will be checked later, what els is unrelated and should not interrupt caching.

Well, this is changing the data... More like a known issue, or even just a TODO imho. Definitively not a bug.

In D6911#164931, @Germano Cavalcante (mano-wii) wrote:

LGFM, but it might be good to wait another day to see what other reviewers think.

Simply adding straight_to_premul_v4(out); to the end of BKE_colorband_evaluate seems to make it behave as I'd expect. No idea of this is the correct fix. workbench_volume_frag.glsl seems to expect that rgb = 0 when alpha = 0.

LGFM, but it might be good to wait another day to see what other reviewers think.

Most likely, jitter occurs when the adaptive Domain begins to trim the excess voxels. When extra voxels are removed, the noise information becomes incorrect and the noise shows artifacts.

Updated with the requested changes.

I would suggest trying to strip this file from any unnecessary things such as camera animation.
Preferably also simplifying simulation. With file as is now it may be impossible to debug.

Can confirm.
For me, blender hangs
T66383: Boids particle system crash might be related, checking if this is a duplicate...

I found out that jitter only appears with Noise enabled in the Domain settings. Without noise, baked smoke does not tremble. In the video, the result is first with noise, then without noise.
I downloaded the cache. Only 180-190 frames.

Hi, I am having same problems.

I'm having the exact same problem. Other simulation caches are writing fine on the disk, but not Mantaflow.

In T74155#880084, @Ankit (ankitm) wrote:

Do you have protected folders turned on ? https://www.tenforums.com/tutorials/87858-add-protected-folders-controlled-folder-access-windows-10-a.html

Looks good. I'm not sure about the changes to the GHash code. Maybe @Campbell Barton (campbellbarton) could take a look at that part?

Use GHash with inthash_v2 for sewing edges test during self collision

Do you have protected folders turned on ? https://www.tenforums.com/tutorials/87858-add-protected-folders-controlled-folder-access-windows-10-a.html

@Yevgeny Makarov (jenkm) Do you have an example file for us that works in 2.79?