Class ShadowUniforms

One shaded pipeline's shadow uniforms: the six locations, the sampler-unit upload, and the CPU mirror that keeps a frame with no casting light emitting no GL call at all (§69, R-18).

const shadows = ShadowUniforms.resolve(gl, program, "lit");
shadows.uploadView(lights);                            // once per viewport
shadows.setReceiving(active && item.receiveShadow);    // once per draw
Copy

Shared by LitProgram and gl-standard.ts's StandardProgram for PunctualLightUniforms' reason: the two stages consume the identical SceneLights fields through the identical GLSL chunk, and a class each would be two places for the skip rule to drift apart.

The skip rule, and why it is byte-identity rather than an optimisation

useShadow is a bool uniform whose CPU mirror starts at false, which is what GL initializes a bool uniform to. A frame in which no light casts therefore uploads nothing at all — not the matrix, not the biases, not the sampler, not the switch — so a scene lit the way every scene was lit before §69 shipped issues byte-for-byte the GL sequence it always did. The mirror-at-GL-initial-0 technique on its fourth recorded use (R-15's bool encode switch, R-38's permissive layer mask, R-17's light count).

The pixel half of that claim lives in the two shaded fragment stages, and is kept the way R-17 kept it: the shadow term multiplies the existing directional expression inside an if (useShadow), so with the switch off the arithmetic is the arithmetic those stages performed before, operation for operation. Nothing is re-associated.

The split between the two methods is the split between per-view state (the matrix and the biases, which belong to the light) and per-draw state (§49's receiveShadow, which belongs to the node).