Regression: Changes in curve normal behavior #107256

New Issue

mirh · 2023-04-23T03:10:09+02:00

mirh commented

2023-04-23 03:10:09 +02:00

Blender Version
Broken: 3.0
Worked: 2.92

Short description of error
Instead of a simple arrow, the scene is something very remotely in the neighbourhood of that shape plus random extruded edges.

Exact steps for others to reproduce the error
Open example file.

**Blender Version** Broken: 3.0 Worked: 2.92 **Short description of error** Instead of a simple arrow, the scene is something very remotely in the neighbourhood of that shape plus random extruded edges. **Exact steps for others to reproduce the error** Open example file.

default_arrow (1).blend

139 KiB

mirh added the

labels 2023-04-23 03:10:10 +02:00

Iliya Katushenock commented

2023-04-23 12:44:08 +02:00

Hello. I see strange curve behavior but I can't open it in earlier versions of blender, where would it work?

mirh commented

2023-04-23 15:07:36 +02:00

Uhm? Even with 2.78 I don't really have problems with opening it.
Then idk what kind of environment @gandalf3 couldn't have had.

Uhm? Even with 2.78 I don't really have problems with opening it. Then idk what kind of environment @gandalf3 couldn't have had.

gandalf3 commented

2023-04-23 23:21:29 +02:00

Looks like the distribution of the loop cuts is slightly different, which causes a slight overlap after the curve modifier. This causes the normals of those faces to be flipped, which messes up the solidify and it devolves from there.

On 2.72 (just a version i happened to have a copy of handy) this overlap did not occur, and the normals were consistent (and the final output of the modifier stack looked correct):

Looks like the distribution of the loop cuts is slightly different, which causes a slight overlap after the curve modifier. This causes the normals of those faces to be flipped, which messes up the solidify and it devolves from there. ![screenshot of the output of the curve modifier in blender 3.5, from a slightly tilted ortho side view, showing the flipped normals](https://projects.blender.org/attachments/573f48e2-554a-48bc-ac4d-b3d9f304072c) On 2.72 (just a version i happened to have a copy of handy) this overlap did not occur, and the normals were consistent (and the final output of the modifier stack looked correct): ![screenshot of the output of the curve modifier in blender 2.72, from a similar slightly tilted ortho side view, showing consistent normals](https://projects.blender.org/attachments/d6eb4a2e-8b1d-463e-a791-a027c747f257)

Selection_059.png

84 KiB

Selection_058.png

8.6 KiB

Iliya Katushenock commented

2023-04-23 23:30:11 +02:00

@gandalf3 It would be nice if you could share a file that can be opened in 2.72
Although I think this is expected behaviour. That is, you initially make a curve of an incorrect shape, which may cause an incorrect result.

@gandalf3 It would be nice if you could share a file that can be opened in 2.72 Although I think this is expected behaviour. That is, you initially make a curve of an incorrect shape, which may cause an incorrect result.

gandalf3 commented

2023-04-23 23:50:14 +02:00

Interestingly it looks like the subsurf modifier in 2.93+ creates twice as many loop cuts length wise (another screenshot attached for comparison).

I'm guessing this is just a quirk of opensubdiv (introduced in 2.80) vs blender's old subsurf algorithm? As such I think this is considered intended behavior, unfortunately, but I wouldn't be the one to say for certain.

Interestingly it looks like the subsurf modifier in 2.93+ creates twice as many loop cuts length wise (another screenshot attached for comparison). I'm guessing this is just a quirk of opensubdiv ([introduced in 2.80](https://wiki.blender.org/wiki/Reference/Release_Notes/2.80/Modeling)) vs blender's old subsurf algorithm? As such I think this is considered intended behavior, unfortunately, but I wouldn't be the one to say for certain.

Selection_060.png

165 KiB

gandalf3 commented

2023-04-23 23:53:29 +02:00

@mod_moder I actually just downloaded the file from the blend-exchange link, but I'll attach it anyway.
I started both versions of blender with --factory-settings as well.

@mod_moder I actually just downloaded the file from the blend-exchange link, but I'll attach it anyway. I started both versions of blender with `--factory-settings` as well.

default_arrow.blend

139 KiB

mirh commented

2023-04-24 02:49:04 +02:00

I'm guessing this is just a quirk of opensubdiv (introduced in 2.80) vs blender's old subsurf algorithm?

That still wouldn't explain why up to 2.92 everything was still seemingly fine, would it?

> I'm guessing this is just a quirk of opensubdiv (introduced in 2.80) vs blender's old subsurf algorithm? That still wouldn't explain why up to 2.92 everything was still seemingly fine, would it?

Iliya Katushenock commented

2023-04-24 06:49:47 +02:00

I can confirm the change in behavior between 2.9 and 3.0 .
But even within the same version, if I just move the point and bring it back, the normal will break. This doesn't look like a real result, but rather an abuse of a bug. And now this bug just looks different.

I can confirm the change in behavior between 2.9 and 3.0 . But even within the same version, if I just move the point and bring it back, the normal will break. This doesn't look like a real result, but rather an abuse of a bug. And now this bug just looks different.

Iliya Katushenock added

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Regression: Changes in curve normal behavior #107256