Weight paint fails the normalization tests in suite #41916

New Issue

Antonis Ryakiotakis · 2014-09-23T13:41:17+02:00

Antonis Ryakiotakis commented

2014-09-23 13:41:17 +02:00

System Information
Ubuntu 14.04
Intel Xeon E31280 @ 3.50GHz x 4
5,8 GB RAM
NVIDIA Quadro 2000

Blender Version
Broken: (example: 2.69.7 4b206af, see splash screen)
2ed1b67

Worked: (optional)

Short description of error
Test files with normalization in test suite fail

Exact steps for others to reproduce the error
Open autoNormalizeTest.blend and mutlipaintTest.blend in tests/paint and follow the instructions on the relevant .txt files. Results will be different. If results are the expected ones we should really change the files in the suite.
Also autoNormalizeTest.blend has some duplicate vertices and the description of the test is not very good/clear either.

**System Information** Ubuntu 14.04 Intel Xeon E31280 @ 3.50GHz x 4 5,8 GB RAM NVIDIA Quadro 2000 **Blender Version** Broken: (example: 2.69.7 4b206af, see splash screen) 2ed1b67 Worked: (optional) **Short description of error** Test files with normalization in test suite fail **Exact steps for others to reproduce the error** Open autoNormalizeTest.blend and mutlipaintTest.blend in tests/paint and follow the instructions on the relevant .txt files. Results will be different. If results are the expected ones we should really change the files in the suite. Also autoNormalizeTest.blend has some duplicate vertices and the description of the test is not very good/clear either.

Antonis Ryakiotakis commented

2014-09-23 13:41:17 +02:00

Changed status to: 'Open'

Jason Hays was assigned by Antonis Ryakiotakis

2014-09-23 13:41:17 +02:00

Antonis Ryakiotakis commented

2014-09-23 13:41:17 +02:00

Added subscriber: @Psy-Fi

Antonis Ryakiotakis commented

2014-09-23 13:41:30 +02:00

Added subscriber: @ideasman42

Jason Hays was unassigned by Antonis Ryakiotakis

2014-09-23 13:51:02 +02:00

Campbell Barton was assigned by Antonis Ryakiotakis

2014-09-23 13:51:02 +02:00

Antonis Ryakiotakis commented

2014-09-23 13:51:02 +02:00

Added subscriber: @jason_hays22

Campbell Barton commented

2014-09-23 18:48:40 +02:00

This is a regression, but its intentional and I rather not change this right before release.

The problem is limited to painting when normalize is enabled, but the current weights aren't normalized.

If all groups other groups besides active are 0.0 options are:

Set the active group to 1.0, leave other groups zero (old behavior)
Keep the active group weight as-is, and adjust other weights to compensate for that. (whats happening now).

The problem with the old behavior, is - if you enabled normalize - and all other groups were zero, painting 0.5 weight (for example), would always re-normalize to 1.0.

The way it works now, you always get the weight you paint on the active group and enabling the normalize option adjusts the others.

We could change how it works...

Detect if the weight is nearing the target weight or not (to make it possible to paint 0.5 weight onto a 1.0 weight), but think this may be over complicating things, it would need to switch behavior
Detect if the weights are already normalized, since this whole issue is mainly caused by having to handle the case where you start out non-normalized (think this could work quite well), since it means on first use - you have the active weight group initialized to a useful unit value, but afterwards - other groups compensate for changes to that active group.

Will have to update the regression instructions.

This is a regression, but its intentional and I rather not change this right before release. The problem is limited to painting when normalize is enabled, but the current weights aren't normalized. If all groups other groups besides active are 0.0 options are: - Set the active group to 1.0, leave other groups zero (old behavior) - Keep the active group weight as-is, and adjust other weights to compensate for that. (whats happening now). The problem with the old behavior, is - if you enabled normalize - and all other groups were zero, painting 0.5 weight (for example), would always re-normalize to 1.0. The way it works now, you always get the weight you paint on the active group and enabling the normalize option adjusts the others. We could change how it works... - Detect if the weight is nearing the target weight or not (to make it possible to paint 0.5 weight onto a 1.0 weight), *but think this may be over complicating things, it would need to switch behavior* - Detect if the weights are already normalized, since this whole issue is mainly caused by having to handle the case where you start out non-normalized (think this could work quite well), since it means on first use - you have the active weight group initialized to a useful unit value, but afterwards - other groups compensate for changes to that active group. Will have to update the regression instructions.

Antonis Ryakiotakis commented

2014-09-23 20:52:36 +02:00

Agree, when I looked into the first file I also thought it worked correctly. I also think that it would be good to be able to paint what you have as a weight, not get something else as a result of normalization. This is what the test file suggested and I was slightly mystified by that.