WebGLUniforms: Assign compareFunction lazily to retain tree-shaking.

[CodyJasonBennett]

Related issue: #24199

Description

Did a quick test for tree-shaking in core (importing a constant), and found the usual expected culprits (mutations from setting static properties in Texture, Euler, Object3D, and devtools), but also a seemingly low hanging fruit in WebGL source. Everything else is both annoying to fix and expected to be present in user-land code for rendering.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

Filesize dev	Filesize PR	Diff
679.4 kB (168.3 kB)	679.4 kB (168.3 kB)	+3 B

🌳 Bundle size after tree-shaking

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

Filesize dev	Filesize PR	Diff
457.4 kB (110.4 kB)	457.4 kB (110.4 kB)	+10 B

[Mugen87]

Would it also fix the tree shaking issue if we do the following?

let emptyTexture2D;

if ( this.type === gl.SAMPLER_2D_SHADOW ) {

    emptyShadowTexture.compareFunction = LessEqualCompare;
    emptyTexture2D = emptyShadowTexture;

} else {

    emptyTexture2D = emptyTexture;

}

I suspect tree-shaking breaks since the code accesses compareFunction right away. If we just move this bit to setValueT1(), we don't need an additional if check.

[CodyJasonBennett]

Yes, anything were we don't do a mutation in the top-level scope.

[mrdoob]

So this PR allows DepthTexture to be tree-shaked?

[CodyJasonBennett]

Yes, effectively. I'm seeing unreasonable effectiveness in Vite (importing from three) with the changes which tree-shakes down to REVISION, __THREE_DEVTOOLS__ and __THREE__. Before, it included Object3D etc. which is more expected. Trying to understand this.

[mrdoob]

🌳 Bundle size after tree-shaking

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

Filesize dev Filesize PR Diff
457.4 kB (110.4 kB) 457.4 kB (110.4 kB) +10 B

How come this PR made the tree shaking number bigger and not smaller? 🤔

[CodyJasonBennett]

That doesn't measure tree-shaking actually, but the output of one configuration of one tool. The untreeshaken size increases since we're adding a branch to the source.

It's like saying something affected device performance, but only accounting for one device/driver at one version.

[mrdoob]

Oh, what should we do to measure tree shaking then...?

[CodyJasonBennett]

Either a matrix of tools and configuration, or be okay with measuring the lowest common denominator. That might mean testing tree-shaking in three as a whole rather than the practical case of rendering a scene with a WebGLRenderer, where this has no effect.

I test a bunch here for instance. https://github.com/CodyJasonBennett/treeshaking

[mrdoob]

@marcofugaro 👀