Geometry nodes attribute output depends on order of outputs #93715

New Issue

Nathan Vasil · 2021-12-05T19:49:59+01:00

Nathan Vasil commented

2021-12-05 19:49:59 +01:00

System Information
Operating system: Windows-10-10.0.18363-SP0 64 Bits
Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.71

Blender Version
Broken: version: 3.0.0, branch: master, commit date: 2021-12-02 18:35, hash: f1cca30557
Broken: version: 3.1.0 Alpha, branch: master, commit date: 2021-12-04 01:26, hash: 5ef5a9fc24

Short description of error
GN uses the full geometry as outputted by each output, apparently in order of creation, to determine the inputs to fields involved in later outputs.

Exact steps for others to reproduce the error
Attached file is handy for demonstrating the error:

You can see I've got stuff labelled here. "Modifier" row uses live geometry nodes modifiers; "Applied" row are the same objects, but modifiers applied (via convert to mesh) in order to examine the output. "Base" column has geometry nodes as shown; "Switched Outputs" 2 columns have the outputs reorganized to get GN to evaluate in a different order. First and second columns have inputs of "StealFrom" and "StealTo" of 1.0 each; third column has inputs of "StealFrom" of 0.5 and "StealTo" of 1.0. All vertices in the objects are assigned to these groups with the weights mentioned.

Anticipated output from first two columns is StealTo 1.0, StealFrom 0.0; in third column, anticipated output is StealTo 0.5, StealFrom 0.0. Actual output, as shown on the Applied row, differs: 0,0 in first row; 0,1 in second row; 0.25, 0.5 in third row.

It appears that the fields reading the groups are acquiring their values, in order, from each preceding output. So in the first column, we're first creating geometry (which is identical to input geometry); then running the StealFrom field on that geometry, giving us a StealFrom of 0 and outputting new geometry; then running the StealTo field on that output, giving us a StealTo of 0. And of course when we switch the outputs, these fields are evaluated in a different order, leading to different final output.

It's not 100% clear to me that this is a bug; order of operations at the final output is only implied, and I don't think is specified anywhere in the manual at https:*docs.blender.org/manual/en/latest/modeling/modifiers/generate/geometry_nodes.html# . I'm making this as at the request of @JacquesLucke , who asked for this report in a post at https:*devtalk.blender.org/t/fields-and-anonymous-attributes-proposal/19450/651 .

There is a workaround, which is to use a capture attribute node to guarantee that the initial values of the attributes are used for both chains:

GNbug.blend

**System Information** Operating system: Windows-10-10.0.18363-SP0 64 Bits Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.71 **Blender Version** Broken: version: 3.0.0, branch: master, commit date: 2021-12-02 18:35, hash: `f1cca30557` Broken: version: 3.1.0 Alpha, branch: master, commit date: 2021-12-04 01:26, hash: `5ef5a9fc24` **Short description of error** GN uses the full geometry as outputted by each output, apparently in order of creation, to determine the inputs to fields involved in later outputs. **Exact steps for others to reproduce the error** Attached file is handy for demonstrating the error: ![image.png](https://archive.blender.org/developer/F12692577/image.png) You can see I've got stuff labelled here. "Modifier" row uses live geometry nodes modifiers; "Applied" row are the same objects, but modifiers applied (via convert to mesh) in order to examine the output. "Base" column has geometry nodes as shown; "Switched Outputs" 2 columns have the outputs reorganized to get GN to evaluate in a different order. First and second columns have inputs of "StealFrom" and "StealTo" of 1.0 each; third column has inputs of "StealFrom" of 0.5 and "StealTo" of 1.0. All vertices in the objects are assigned to these groups with the weights mentioned. Anticipated output from first two columns is StealTo 1.0, StealFrom 0.0; in third column, anticipated output is StealTo 0.5, StealFrom 0.0. Actual output, as shown on the Applied row, differs: 0,0 in first row; 0,1 in second row; 0.25, 0.5 in third row. It appears that the fields reading the groups are acquiring their values, in order, from each preceding output. So in the first column, we're first creating geometry (which is identical to input geometry); then running the StealFrom field on that geometry, giving us a StealFrom of 0 and outputting new geometry; then running the StealTo field on that output, giving us a StealTo of 0. And of course when we switch the outputs, these fields are evaluated in a different order, leading to different final output. It's not 100% clear to me that this is a bug; order of operations at the final output is only implied, and I don't think is specified anywhere in the manual at https:*docs.blender.org/manual/en/latest/modeling/modifiers/generate/geometry_nodes.html# . I'm making this as at the request of @JacquesLucke , who asked for this report in a post at https:*devtalk.blender.org/t/fields-and-anonymous-attributes-proposal/19450/651 . There is a workaround, which is to use a capture attribute node to guarantee that the initial values of the attributes are used for both chains: ![image.png](https://archive.blender.org/developer/F12692752/image.png) [GNbug.blend](https://archive.blender.org/developer/F12692572/GNbug.blend)

Nathan Vasil commented

2021-12-05 19:49:59 +01:00

Added subscribers: @JacquesLucke, @vasiln

blender-admin commented

2021-12-05 19:49:59 +01:00

#93572 was marked as duplicate of this issue

Hans Goudey commented

2021-12-06 05:34:07 +01:00

Added subscribers: @christopher.burns, @HooglyBoogly, @null

Hans Goudey commented

2021-12-06 05:34:26 +01:00

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

Hans Goudey changed title from ~~Geometry nodes output depends on order of outputs~~ to Geometry nodes attribute output depends on order of outputs

2021-12-06 05:34:51 +01:00

Hans Goudey commented

2022-04-28 18:13:02 +02:00

I think this should be resolved after a985f558a6

@vasiln, could you check that? I tried looking through the file but I'm finding it a bit confusing and it might be better to check that it matches your expectations anyway. Thanks.

I think this should be resolved after a985f558a6 @vasiln, could you check that? I tried looking through the file but I'm finding it a bit confusing and it might be better to check that it matches your expectations anyway. Thanks.

Hans Goudey commented

2022-04-28 18:31:33 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Hans Goudey closed this issue

2022-04-28 18:31:33 +02:00

Hans Goudey self-assigned this 2022-04-28 18:31:33 +02:00

Hans Goudey commented

2022-04-28 18:31:33 +02:00

I'll close this as resolved since I'm pretty confident the issue was solved from looking at the code. We can easily reopen the report if I'm wrong though. Thanks.

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

Geometry nodes attribute output depends on order of outputs #93715