Examples: Remove EquirectangularToCubeGenerator.

[Mugen87]

...and use WebGLRenderTargetCube.fromEquirectangularTexture() instead.

[WestLangley]

I'm not sure that makes sense. Here, or in the lines above. Probably should be Linear.

You will have to step through the code to see if it matters.

Also, you will almost surely have to change the loader output data type to HalfFloat to duplicate the previous results.

.setDataType( THREE.HalfFloatType )

[Mugen87]

I'm not sure that makes sense. Here, or in the lines above. Probably should be Linear.

Why do you think are the previous texture filtering settings wrong? It does actually look good.

Also, you will almost surely have to change the loader output data type to HalfFloat to duplicate the previous results.

That's true. Let me fix this!

[Mugen87]

Links for direct comparison:

https://threejs.org/examples/webgl_materials_envmaps_exr
https://raw.githack.com/Mugen87/three.js/dev39/examples/webgl_materials_envmaps_exr.html

[WestLangley]

Your example is throwing errors on my machine.

I have always used Linear interpolation when the encoding supports it. Some encodings like RGBE are not compatible with linear, so nearest must be used. I have never seen Linear and Nearest mixed in the same texture. If there is a use case for that, I am not aware of it.

[Mugen87]

Your example is throwing errors on my machine.

Um, can't reproduce this on my iMac with Chrome. What errors are logged?

[WestLangley]

Failure is on macOS Safari.

WebGL: INVALID_OPERATION: texImage2D: type HALF_FLOAT_OES but ArrayBufferView is not NULL

[Mugen87]

Okay, I can reproduce with Safari. It seems WebGLRenderTargetCube.fromEquirectangularTexture() throws this error when the input texture is of THREE.HalfFloatType. Using THREE.FloatType with EXRLoader as before solves the issue. However, since WebGLRenderTargetCube.fromEquirectangularTexture() strictly derives the type of the source texture (unlike EquirectangularToCubeGenerator which does type: options.type || this.sourceTexture.type), I can't quickly fix this issue without changing previous values (see #17708 (comment)).

[Mugen87]

see #17727

[WestLangley]

@Mugen87

1. I forked your branch and tested webgl_loader_gltf_extensions.html, and it appears to be OK.

2. webgl_materials_envmaps_exr.html has the issues I mentioned above.

3. I did not study webgl_loader_gltf.html. You can do that one. ;-)

---

As I did, you need to make sure encoding, generateMipmaps, magFilter, minFilter, and type are unchanged from the previous examples -- for both the background and the environment map. If the settings are unchanged, but do not make sense, then this is an opportunity to fix them.

As I said above, PMREM changes the input texture, so you need to check the background texture after PMREM is finished.

---

Personally, I think setting a background cube map with mipmapping disabled is probably not what we want to be recommending, but I guess that can be a separate issue.

[Mugen87]

I've checked all examples again and made sure the properties of the background's texture are the same as before. Let me list those values in the following table:

Background:

Example	encoding	generateMipmaps	minFilter	magFilter	type
gltf	RGBEEncoding	false	NearestFilter	NearestFilter	UnsignedByteType
gltf_ext	RGBEEncoding	false	NearestFilter	NearestFilter	UnsignedByteType
exr (exr)	LinearEncoding	false	LinearFilter	LinearFilter	HalfFloatType
exr (png)	sRGBEncoding	true	LinearFilter	LinearFilter	UnsignedByteType

EnvMap:

Example	encoding	generateMipmaps	minFilter	magFilter	type
gltf	RGBM16Encoding	false	LinearFilter	LinearFilter	UnsignedByteType
gltf_ext	RGBM16Encoding	false	LinearFilter	LinearFilter	UnsignedByteType
exr (exr)	LinearEncoding	false	LinearFilter	LinearFilter	HalfFloatType
exr (png)	sRGBEncoding	true	LinearFilter	LinearFilter	UnsignedByteType

Measured after PMREM has finished.

[Mugen87]

If the settings are unchanged, but do not make sense, then this is an opportunity to fix them.

The PNG background/env maps in the EXR example have set generateMipmaps to true although only LinearFilter is used. I think this can be changed, right?

[WestLangley]

Right. Fix it if you can. Otherwise, wait until the PMREM rewrite is finalized.

[Mugen87]

Merging for now. The settings in webgl_materials_envmaps_exr can be optimized with a different PR.

Sidenote: Let's not forget webgl_materials_envmaps_exr does not work in FF (see #16778).