PassNode: Fix depthTexture creation when depthBuffer: false

[thelazylamaGit]

With PassNode a depth texture was always being created even when explicitly setting { depthBuffer: false }.

Additonally, Textures.updateRenderTarget() also handled depth before checking whether it was actually enabled. Because of that, a depth texture could end up being created even with depth disabled.

This pr ensures that when creating a pass such as

pass(this.scene, this.camera, { depthBuffer: false }))

no depth texture is created.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	360.01 85.46	360.01 85.46	+0 B +0 B
WebGPU	633.22 175.54	633.33 175.58	+113 B +32 B
WebGPU Nodes	631.34 175.25	631.45 175.28	+113 B +31 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	492.21 120.03	492.21 120.03	+0 B +0 B
WebGPU	705.25 190.44	705.36 190.46	+113 B +21 B
WebGPU Nodes	654.47 177.68	654.58 177.71	+113 B +22 B

[Mugen87]

This PR introduces breakage in the AO demo. The following WebGPU warning appears now in the browser console:

Copy origin ([Origin3D x:0, y:0, z:0]) and size ([Extent3D width:4400, height:1828, depthOrArrayLayers:1]) does not cover the entire subresource (origin: [x: 0, y: 0], size: [Extent3D width:1, height:1, depthOrArrayLayers:1]) of [Texture (unlabeled 1x1 px, TextureFormat::Depth24Plus)]. The entire subresource must be copied when the format (TextureFormat::Depth24Plus) is a depth/stencil format or the sample count (1) is > 1.
 - While validating destination [Texture (unlabeled 1x1 px, TextureFormat::Depth24Plus)].
 - While [Failed to format error message: "encoding %s.CopyTextureToTexture(%s, %s, %s)."].
 - While finishing [CommandEncoder "copyTextureToTexture_19_27"].

[thelazylamaGit]

Yep I see the problem, the AO example uses TRAANode which creates a renderTarget with depthBuffer: false but then manually supplies its own DepthTexture

this._historyRenderTarget = new RenderTarget( 1, 1, { depthBuffer: false, type: HalfFloatType, depthTexture: new DepthTexture() } );

I'll work on a fix making sure to handle this case

[Mugen87]

Thank you!

BTW: I know it can feel off-putting if the own PR is reverted but I just want to say that I don't intend to be mean here^^. It's more about having a functional dev branch that does not irritate or hamper the work of other developers. I'm thankful for everyones understanding in that regard.

[thelazylamaGit]

No worries, I understand. Just happy to contribute and glad the regression was caught so soon after being merged. New PR is up now which fixes the issue.