Material: Replace .format with .alphaWrite

[donmccurdy]

Related issues:

• Fixes Material: transparent=true has no effect on glTF materials (since r132) #22598
• GLTFLoader: Texture alpha should be ignored on opaque materials #15483
• WebGLRenderer: Ignore map alpha in opaque materials #18631
• WebGLRenderer alpha GLTFLoader texture issue #21744
• ShaderChunk: Add opacity support to transmission shader #22196
• ShaderChunks: Ignore alpha in output_fragment when transparent = false #22414
• WebGLRenderer: Add support for GLTF opaque alpha_mode #22428

Implements the solution proposed in #22598 (comment). This fixes the problem of .transparent=true having no effect for materials loaded with THREE.GLTFLoader, because .alphaWrite is automatically flipped to true ignored when .transparent=true is set. I hope it's also a clearer name than .format=RGB[A]Format, and more comparable to .colorWrite.

Unfortunately I'm no longer sure how to test all of the issues discussed above. The test model provided with #15483 renders correctly regardless of whether alphaWrite is on or off:

[takahirox]

Thanks for making a PR.

This fixes the problem of .transparent=true having no effect for materials loaded with THREE.GLTFLoader, because .alphaWrite is automatically flipped to true when .transparent=true is set.

A concern of this is if .transparent is set back to false .alphaWrite won't be reset so the behavior can be different from the first one.

So I voted for the option B in #22598 (comment)

What do you think of it?

[donmccurdy]

Sounds good to me, switched to (B) here. 👍

[Mugen87]

Sidenote: The docs should also reflect this change (a separate PR for this is fine):

https://threejs.org/docs/#api/en/materials/Material.format

[mrdoob]

Thanks!

[WestLangley]

You should simply follow the glTF spec and have material.alphaMode (OPAQUE, BLEND, MASK) and material.alphaCutoff.

Remove material.alphaTest.

material.alphaWrite is misnamed. You can't "write alpha" into the destination if the destination has no alpha channel.

[mrdoob]

That may be the best solution indeed...

Not a fan of how GLTF is "designing" the Three.js API itself, but I guess it's inevitable.

[donmccurdy]

Hm, I can only really see .alphaMode working if it replaces .transparent. There are some benefits – .alphaMode is very similar to Blender's "Blend Mode" (although I don't like that name as much...) and provides a good place to offer alpha hash or other modes someday. But it certainly wasn't my intention to push a big change on the material API for what is currently a pretty subtle issue.

Blender's Material → Settings → Blend Mode UI:

[mrdoob]

@WestLangley

You should simply follow the glTF spec and have material.alphaMode (OPAQUE, BLEND, MASK) and material.alphaCutoff.

That sounds good to me.

We would then have to turn transparent into a getter/setter that basically enables/checks alphaMode = THREE.BlendAlphaMode, right?

And then WebGLRenderer would use alphaMode to build the render lists.

[WestLangley]

@mrdoob It is worth a try... but is it a bit complex... Here are a few random thoughts to get started:

1. transparent in three.js has two purposes. (1) to set the render list, and (2) to set the blending mode.

2. The current render lists are: background, opaque, transmissive, and transparent. Note, with "custom blending", the object could be in the opaque or transparent list depending on the value of transparent.

3. From WebGLState.js, transparent = false is only honored if blending is Normal Blending.

4. Setting material.transparent = false when material.blending = NormalBlending is the same as setting material.blending = THREE.NoBlending -- it just disables blending.

5. With opaque mode, we need to overwrite source alpha with 1 so that destination alpha is set to 1.

6. With mask mode, we need to discard fragments with alpha < alphaCufoff, and set source alpha to 1. (currently, we do not set source alpha to 1)

7. I think all the logic can go in a new alphaMode shader chunk, to replace the alphaTest chunk.

8. alphaMode should probably only apply in the case of Normal Blending.

[donmccurdy]

With mask mode, we need to discard fragments with alpha < alphaCufoff, and set source alpha to 1. (currently, we do not set source alpha to 1)

Would we also want a blend-and-mask mode? glTF doesn't offer that, but it's a reasonable combination I think.

alphaMode should probably only apply in the case of Normal Blending.

I think I would have expected alphaMode = BlendAlphaMode to be allowed with other blend modes. In Blender for example, "Blend Mode = Alpha Blend" appears to be required when using additive or multiply blending.