Issue is in the vertex shader

See some changes I made to make the drawing part working.

Code wise want to request some small changes most of them are better naming to improve the readability of the code.
Only suggestion I have is to encapsulate the cavity_factor/cavity_stroke_id in a struct for performance reasons.

Added Python API for devce type.
Documented usage of gpu_info.
Removed references to platform.

• Get GPU platform.
• Added GPU info as script to solve initialization errors.

Renamed to reference override dir

A similar fix can be done for MacOS workbench hair as a threshold of 0.05 would pass the test.
Adding support for other platforms would IMO require platform specific images.

Not sure about to reserve Mask and Mode suffix for enums only. Mask is something that already exists in the user domain.
I would allow Mode when we refer to edit/object/paint mode as it refers to something concrete and also present in the user domain.

image alpha ignore, image alpha ignore osl,image mapping udim tests are currently failing using (NVIDIA 1080 Ti/driver 515/Linux).

@Spundun Bhatt (spundun), yes wasn't aware that the python api was already added. Note that the python example files and documentation might need some examples using the new GPUShaderCreateInfo.

@Spundun Bhatt (spundun), yes we had to do the conversion already for the metal backend. Changes in eevee still needs to be done, but that will only be done for eevee-next. Will mark this task complete.

Interesting that AOV test case is working... will start a debugging session.
Ok it is working when the output is connected to an input node.
I thing that the data is read from a variable that isn't set, unless it is connected.

Issue is that the order of uniforms are reversed and an additional uniform overwrite the bloomAddBase to include the render result.

I reconfirm that the mentioned commit is the culprit. Will review the change perhaps it changes the gpu state in a way that wasn't meant to be.

Yes totally. I would say that all tools would follow the decision we make, including modifiers.

Not sure. I expect to use the color space during painting. But rendering to be linear. Might be confusing to some users. But making them more consistent might not solve any problems. Perhaps introduce several along the way. IMO we should mention what we decide in the manual.

Hi @Max Edge (bbbooo3) thanks for the patch. It seems that this is less work than I expected. The patch seems to work great. I will poke @Julian Eisel (Severin) to see if he has some feedback.

Was doing a code review and stumbled on this issue.
There is a shader compilation error that crashes Blender.

ERROR (gpu.shader): EEVEE_shaders_cryptomatte_sh_get VertShader: 
      | 
 5664 |   attrib_load();
      | 
      | Error: '' :   Too few args in Macro  attrib_load

Adding Julian as he is more familiar with other pickers and that this change doesn't break something elsewhere.

I did research this one a few weeks ago, but am still waiting to build a machine to reproduce this. What I found so far is faulty logic in the gpu shader descriptor when using conservative depth.

Should be done in master, add windows to the workaround.