Nodes: Broken noodle display with reroutes #106929

New Issue

Sun Kim · 2023-04-14T08:01:09+02:00

Sun Kim commented

2023-04-14 08:01:09 +02:00

System Information
Operating system: Windows-10-10.0.19045-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 531.61

Blender Version
Broken: version: 3.6.0 Alpha, branch: main, commit date: 2023-04-12 22:05, hash: d48939f103c0
Worked: version: 3.6.0 Alpha, branch: main, commit date: 2023-03-14 01:53, hash: 84b66c5a9c47

Caused by something between 2.6 2023-03-14 01:53, hash: 84b66c5a9c47 and 3.6 2023-03-20 14:55 16fbadde363c.

Also can confirm in 3.5.0.

Short description of error
In some of hair nodes, node noodles look broken with reroutes.

The glitch simply goes away if I move the reroutes.

Exact steps for others to reproduce the error
I don't know how to replicate this from scratch, only found in hair node groups.

**System Information** Operating system: Windows-10-10.0.19045-SP0 64 Bits Graphics card: NVIDIA GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 531.61 **Blender Version** Broken: version: 3.6.0 Alpha, branch: main, commit date: 2023-04-12 22:05, hash: `d48939f103c0` Worked: version: 3.6.0 Alpha, branch: main, commit date: 2023-03-14 01:53, hash: `84b66c5a9c47` Caused by something between 2.6 2023-03-14 01:53, hash: `84b66c5a9c47` and 3.6 2023-03-20 14:55 `16fbadde363c`. Also can confirm in 3.5.0. **Short description of error** In some of hair nodes, node noodles look broken with reroutes. ![BrokenNoodles2.png](/attachments/756f2ec8-2476-4580-835f-b9b456f35e85) ![BrokenNoodles.gif](/attachments/6ca34974-918a-4274-834d-3866b6b62e91) The glitch simply goes away if I move the reroutes. **Exact steps for others to reproduce the error** I don't know how to replicate this from scratch, only found in hair node groups.

BrokenNoodles.gif

185 KiB

BrokenNoodles2.png

156 KiB

Sun Kim added the

labels 2023-04-14 08:01:09 +02:00

Pratik Borhade commented

2023-04-14 08:08:32 +02:00

Thanks for the report. Can confirm

Pratik Borhade added

and removed

labels 2023-04-14 08:08:54 +02:00

Pratik Borhade commented

2023-04-14 09:18:02 +02:00

Did bisecting two times and it pointed to: 740c9204cc
not sure if its correct responsible commit.

Did bisecting two times and it pointed to: 740c9204cc40565caf9026610145bc3a4ca5b75d not sure if its correct responsible commit.

Philipp Oeser commented

2023-04-14 10:46:32 +02:00

CC @bonj
CC @HooglyBoogly

CC @bonj CC @HooglyBoogly

Leon Schittek commented

2023-04-14 11:01:36 +02:00

@PratikPB2123, @lichtwerk I don't think that commit is at fault. :)

This has been a problem for quite a while when sockets align as reported in #89282. Originally this was only a problem for vertically aligning sockets but since 9a86255da8 it also happens to horizontally aligning sockets.

What happens is that for aligning sockets the handles of the bezier curve are coincident with the end points, and as a result these last segments of the curve can't be drawn. That is why this is only visible when the sockets are far enough apart that the last segment of the curve is longer than the size of the node socket.

It seemed like someone wanted to tackle this as their first issue, but it's been a while and I can make a fix over the weekend if it's urgent.

@PratikPB2123, @lichtwerk I don't think that commit is at fault. :) This has been a problem for quite a while when sockets align as reported in #89282. Originally this was only a problem for vertically aligning sockets but since 9a86255da8 it also happens to horizontally aligning sockets. What happens is that for aligning sockets the handles of the bezier curve are coincident with the end points, and as a result these last segments of the curve can't be drawn. That is why this is only visible when the sockets are far enough apart that the last segment of the curve is longer than the size of the node socket. It seemed like someone wanted to tackle this as their first issue, but it's been a while and I can make a fix over the weekend if it's urgent.

👍 1

Iliya Katushenock added

and removed

labels 2023-04-14 13:23:08 +02:00

Iliya Katushenock commented

2023-04-14 13:23:53 +02:00

This is definitely not a recent regression. And this can only be achieved by deliberately making such long links. So it would be nice if it gets fixed, but I don't think it's urgent.

Marino Toscano commented

2023-05-04 01:44:33 +02:00

Hi @lone_noel, @mod_moder

I found out what the problem is.
In the vertex shader the tangent is a zero vector for the endpoints.
When using normalize(tangent) the tangent becomes an invalid value.

When excluding the normalization for zero vectors the result is like in the image.
There is still an issue that the first vertex is not expanded and is rendered like a triangle, but I see that for vertical nodelinks it was already a problem also if the handles are not overlapping with the endpoints.

Should I create a PR with this first fix or should I try to fix the problem with the thickness of the endpoints?

Hi @lone_noel, @mod_moder I found out what the problem is. In the vertex shader the tangent is a zero vector for the endpoints. When using `normalize(tangent)` the tangent becomes an invalid value. When excluding the normalization for zero vectors the result is like in the image. There is still an issue that the first vertex is not expanded and is rendered like a triangle, but I see that for vertical nodelinks it was already a problem also if the handles are not overlapping with the endpoints. Should I create a PR with this first fix or should I try to fix the problem with the thickness of the endpoints?

nodelink.png

7.3 KiB

Leon Schittek commented

2023-05-04 07:37:11 +02:00

Hey, thanks for looking into it @merotosc!

Ideally the issue would be fixed completely. I can think of two ways to approach this:

Use a good heuristic to replace the invalid tangent in the shader.
The shape of node links is pretty constrained (symmetric, the „bezier handles“ are always horizontal) so you might be able to get away with replacing the tangent with something like normalize(P3 - P0) when the tangent is zero. I hope there isn't an edge case I'm missing here.
Avoid creating node links that cause issues.
The control polygon of the bezier curve is calculated in calculate_inner_link_bezier_points. Maybe you could make it so the inner points don’t overlap the end points exactly, which would sidestep the issue.

If there are no issues with it, I’d gravitate towards option 1. Not sure what would be preferred by the core team, though.
A PR would be very appreciated either way!

Hey, thanks for looking into it @merotosc! Ideally the issue would be fixed completely. I can think of two ways to approach this: 1. Use a good heuristic to replace the invalid tangent in the shader. The shape of node links is pretty constrained (symmetric, the „bezier handles“ are always horizontal) so you might be able to get away with replacing the tangent with something like `normalize(P3 - P0)` when the tangent is zero. I hope there isn't an edge case I'm missing here. 2. Avoid creating node links that cause issues. The control polygon of the bezier curve is calculated in `calculate_inner_link_bezier_points`. Maybe you could make it so the inner points don’t overlap the end points exactly, which would sidestep the issue. If there are no issues with it, I’d gravitate towards option 1. Not sure what would be preferred by the core team, though. A PR would be very appreciated either way!

Marino Toscano referenced a pull request that will close this issue

2023-05-05 01:03:58 +02:00

Fix #106929: adjust normalization of zero vector in nodelink shader #107636

Marino Toscano commented

2023-05-05 01:11:17 +02:00

Yes, you are right @lone_noel. Using normalize(P3 - P0) already solves it.

Now there is only the issue for the vertical nodelinks that don't overlap because the tangent is always (+/-1, 0) at endpoints since the handles are always horizontally aligned with the nodes.
But I can't think of an easy approximation for the tangent on the endpoints.

I now created a PR with this fixes: #107636

Yes, you are right @lone_noel. Using `normalize(P3 - P0)` already solves it. Now there is only the issue for the vertical nodelinks that don't overlap because the tangent is always (+/-1, 0) at endpoints since the handles are always horizontally aligned with the nodes. But I can't think of an easy approximation for the tangent on the endpoints. I now created a PR with this fixes: #107636

nodelink_2.png

15 KiB

Leon Schittek closed this issue

2023-05-12 20:35:32 +02:00

Blender Bot added

and removed

labels 2023-05-12 20:35:33 +02:00

Leon Schittek referenced this issue from a commit