Geometry nodes Volume Cube crashes when volume is zero #99004

New Issue

Rygel · 2022-06-20T04:32:42+02:00

Rygel commented

2022-06-20 04:32:42 +02:00

System Information
Operating system: Linux 5.10.117-1-MANJARO
Graphics card: AMD ATI Radeon RX 580

Blender Version
Broken: 3.3.0 alpha (since 838c4a97f1)

Short description of error
Sometimes when dragging either of the limits of the node Volume Cube, it will crash when they are equal to each other. Other times, it will correctly report that the volume is zero, avoiding the crash.

Exact steps for others to reproduce the error

Open .blend file
Select Volume Cube node
select all three min input (click and drag down)
Type 1.0001 (Will crash)
Will also crash on dragging all the values of the "min" input until they are at 1.0 (The input must be dragged all at once, or it won't crash)

volume crash.blend

**System Information** Operating system: Linux 5.10.117-1-MANJARO Graphics card: AMD ATI Radeon RX 580 **Blender Version** Broken: 3.3.0 alpha (since 838c4a97f1) **Short description of error** Sometimes when dragging either of the limits of the node Volume Cube, it will crash when they are equal to each other. Other times, it will correctly report that the volume is zero, avoiding the crash. **Exact steps for others to reproduce the error** - Open .blend file - Select `Volume Cube` node - select all three `min` input (click and drag down) - Type 1.0001 (Will crash) Will also crash on dragging all the values of the "min" input until they are at 1.0 (The input must be dragged all at once, or it won't crash) [volume crash.blend](https://archive.blender.org/developer/F13205314/volume_crash.blend)

Rygel commented

2022-06-20 04:32:42 +02:00

Added subscriber: @rygelXVI

blender-admin commented

2022-06-20 04:32:42 +02:00

#100087 was marked as duplicate of this issue

Pratik Borhade commented

2022-06-20 04:48:25 +02:00

Added subscriber: @PratikPB2123

Pratik Borhade commented

2022-06-20 05:25:40 +02:00

Changed status from 'Needs Triage' to: 'Confirmed'

Pratik Borhade commented

2022-06-20 05:25:40 +02:00

Added subscriber: @lateasusual

Dalai Felinto commented

2022-06-24 16:47:56 +02:00

Closed as duplicate of #94969

Dalai Felinto closed this issue

2022-06-24 16:47:56 +02:00

Dalai Felinto commented

2022-06-24 16:50:46 +02:00

Added subscriber: @dfelinto

Dalai Felinto commented

2022-06-24 16:50:46 +02:00

Thanks for the report. I'm assuming this was what was fixed on 522dcc54af.

In the future remember to include the hash of the Blender build you used for test (if you report the bug from within Blender it is filled automatically). Otherwise it is hard to tell if it was fixed already.

Thanks for the report. I'm assuming this was what was fixed on 522dcc54af. In the future remember to include the hash of the Blender build you used for test (if you report the bug from within Blender it is filled automatically). Otherwise it is hard to tell if it was fixed already.

Pratik Borhade commented

2022-06-24 17:59:55 +02:00

Changed status from 'Duplicate' to: 'Confirmed'

Pratik Borhade reopened this issue

2022-06-24 17:59:55 +02:00

Pratik Borhade commented

2022-06-24 17:59:55 +02:00

Can still repro crash in recent buildbot build. So reopening the ticket :)
(Also I've made some edits in task description)

Can still repro crash in recent buildbot build. So reopening the ticket :) (Also I've made some edits in task description)

Jacques Lucke commented

2022-07-26 13:39:11 +02:00

Added subscribers: @brecht, @JacquesLucke

Jacques Lucke commented

2022-07-26 13:39:11 +02:00

Hm, that's annoying, there is an assert in openvdb:

The geometry nodes side feels quite reasonable, not sure if there is really an error in the logic. We could clamp the values on geometry nodes, which would help solve the crash in that specific case (at the cost of a wrong result), but not the more general case. For example, using the Transform node to scale the volume down a lot would still result in this crash. The only real solution I see right now is that we could store a transformation matrix on top of the transform stored in an opengdb grid. @brecht what do you think about that?

Hm, that's annoying, there is an assert in openvdb: ![image.png](https://archive.blender.org/developer/F13318460/image.png) The geometry nodes side feels quite reasonable, not sure if there is really an error in the logic. We could clamp the values on geometry nodes, which would help solve the crash in that specific case (at the cost of a wrong result), but not the more general case. For example, using the Transform node to scale the volume down a lot would still result in this crash. The only real solution I see right now is that we could store a transformation matrix on top of the transform stored in an opengdb grid. @brecht what do you think about that?

Brecht Van Lommel commented

2022-07-26 20:07:40 +02:00

An additional transformation matrix I guess will make the API and I/O more complicated? I'm not sure how that would map for renderers, exporters or saving OpenVDB files.

Would it be bad to generate an empty grid with unit scale (maybe keep translation and rotation)? The Transform node could do the same. It's somewhat inconsistent with other geometry types, but if you scale a mesh down to zero with the Transform node it is also effectively destroyed, so I'm not sure it matters much.

An additional transformation matrix I guess will make the API and I/O more complicated? I'm not sure how that would map for renderers, exporters or saving OpenVDB files. Would it be bad to generate an empty grid with unit scale (maybe keep translation and rotation)? The Transform node could do the same. It's somewhat inconsistent with other geometry types, but if you scale a mesh down to zero with the Transform node it is also effectively destroyed, so I'm not sure it matters much.

Hans Goudey commented

2022-07-26 20:12:13 +02:00

Added subscriber: @HooglyBoogly

Hans Goudey commented

2022-07-26 20:12:13 +02:00

I agree with Brecht that supporting this case is not worth the complication in other areas. The additional matrix would also have a performance cost when sampling volumes.
I don't have much of an opinion on how to solve this otherwise though. Generating an empty grid seems fine to me.

I agree with Brecht that supporting this case is not worth the complication in other areas. The additional matrix would also have a performance cost when sampling volumes. I don't have much of an opinion on how to solve this otherwise though. Generating an empty grid seems fine to me.

Chris Clyne (Lateasusual) commented

2022-07-26 20:15:21 +02:00

I think an empty grid would be sensible.