WebGLRenderer: Fix feedback loop in transmission pass for WebGL 2.

[Mugen87]

Fixed #25990.

Description

#25502 introduced a feedback loop during the back-side transmission pass since the active render target is equal to the texture which is assigned to transmissionSamplerMap. The feedback loop produces a WebGL warning but also the result of the back-side pass is missing in the render target (so the transmissive surface looks like before #25502 was added). The warning looks like so:

GL_INVALID_OPERATION: Feedback loop formed between Framebuffer and active Texture.

This issue does not pop-up when antialias is set to true which is why the issue was overlooked initially. This happens because setting antialias to true means the transmissive render target is multisampled. Multisampled render targets do have more than one framebuffer so using it as the active render target and as a texture does not produce a feedback loop.

Solving this issue without multisampling is difficult because you need a separate code path that copies the contents of the transmissive render target via copyTexSubImage2D() into a texture and then uses it as the value for transmissionSamplerMap. When trying to get this to work, I have realized this will noticeably increase the code complexity with unclear performance implications (since you have to call copyTexSubImage2D() two times inside renderTransmissionPass()).

For simplicity reasons, I suggest to always use a multisampled render target for the transmission pass when using WebGL 2 which fixes the issue.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

Filesize dev	Filesize PR	Diff
644.1 kB (159.5 kB)	644.1 kB (159.5 kB)	-8 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

Filesize dev	Filesize PR	Diff
434.1 kB (105.1 kB)	434.1 kB (105.1 kB)	-8 B

[Mugen87]

@mrdoob Are you okay with merging this PR? Otherwise transmission is broken for all users who are not setting antialias to true.

Considering the complexity of a larger refactoring, it seems to me the proposed solution here is best.

[Mugen87]

I'm going ahead and merge this PR so we have a hotfix for r154.

[mrdoob]

@mrdoob Are you okay with merging this PR? Otherwise transmission is broken for all users who are not setting antialias to true.

Considering the complexity of a larger refactoring, it seems to me the proposed solution here is best.

Sounds good!