[proggeramlug]

Summary

Follow-up to #199, which moved the main pass's shadow receiver bias from a constant in NDC to one shadow texel of world distance. Every other reader of the directional cascades still subtracted a constant in NDC, and all of them scale the same way: the cascades pancake their depth range to the scene bounds, so a 600 m ground plane turns 0.002 into a metre and anything closer than that to a receiver stops casting on it (a deck no longer shadowing the hull under it, contact shadows floating).

This PR gives each remaining reader the same world-sized bias, measured from the cascade's own orthographic view-projection so no new uniform is needed: the x row of an ortho VP has length 2 / width in metres and the z row 1 / depth range, so one texel is 2 / (|x row| · map size) metres and |z row| times that in NDC.

Reader	Was	Now
GI mesh-card relight (CARD_LIGHT_WGSL)	0.002 NDC, position only	1.5-texel normal offset along the card normal + 1 texel depth (flags.x is now a texel count)
WSRC probe bake (WSRC_BAKE_WGSL)	0.002 NDC	2 texels depth; probes sit in free space, no normal to offset along
Path tracer hybrid sun (sun_vis_cascade)	0.002 NDC, caller added 2 cm	1.5-texel normal offset + 1 texel depth; takes the normal
Planar reflection pass (refl_sun_shadow)	0.001 NDC	1 texel depth (it already had the main pass's normal offset)
Transmitted shadow resolve	0.0015 NDC	2 transmitted-map texels
Transmitted shadow caster hidden-glass test	0.00075 NDC	2 opaque-map texels (the draw uniforms are now visible to the fragment stage for light_vp)
Virtual shadow maps, scene helper	0.001 NDC	1 virtual texel (receiver already normal-offset)
Virtual shadow maps, material helper	0.001 NDC	2 virtual texels

The hardware GI variants (CARD_LIGHT_HW_WGSL, WSRC_BAKE_HW_WGSL) trace their sun visibility and never read the bias, so they are unchanged. The SSGI probe-trace uniform still carries a 0.002 in shadow_params.x, but no shader reads it. Point and spot light VSM pages keep their perspective-depth bias; that one does not pancake.

Test plan

• cargo test --release --no-fail-fast in native/shared: 509 unit tests, the golden suite and the runtime suites pass with no golden image changed. The one failure, visibility_shading_matches_forward_reference (max delta 2 against a 1-LSB bound), fails identically on main at 49cf675 on this machine (148 changed channels in both).
• cargo fmt --check, ./scripts/ci-check.sh --quick --component lint and --component contracts (file-size ratchet included) pass.
• The Deep (600 m sea plane, SSGI on the hardware ray-query backend): hold, cargo deck and lots scenes rendered against this branch and against main are visually identical at 1600×900. The user-visible fix was shadows: the receiver bias is a shadow texel of world distance, not a constant in NDC #199; this one brings the bounce-light and reflection feeds in line with it.
• CI

Summary by CodeRabbit

• Bug Fixes

  • Shadow depth bias now scales with shadow-map texel size and cascade depth, improving consistency across shadow maps and scenes.
  • Shadow receivers receive a small normal-based offset to help reduce shadow artifacts.
  • Transmitted-shadow blocker and receiver tolerances now adapt to map resolution.

• Documentation

  • Clarified that shadow-bias settings are measured in shadow texels.

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

Shadow sampling paths replace fixed depth-bias values with calculations based on shadow texel size and projection scale. The changes also add receiver normal offsets to selected cascade sampling paths and derive transmitted-shadow blocker tolerance from map dimensions.

Changes: Shadow Bias Scaling

Layer / File(s)	Summary
Virtual shadow sampling native/shared/shaders/virtual_shadows/*	Virtual shadow helpers derive receiver depth bias from the cascade matrix and virtual-page dimensions instead of subtracting a fixed NDC value.
Environment and path-traced cascade sampling native/shared/src/renderer/shaders/env.rs, native/shared/src/renderer/shaders/pt.rs	Environment and path-traced cascade sampling derive receiver bias from projection scale and texel size. Path-traced sampling offsets the receiver along its normal; debug mode 18 checks visibility at the unoffset surface.
GI shadow sampling and bias parameters native/shared/src/renderer/shaders/gi.rs, native/shared/src/renderer/gi_bake.rs, native/shared/src/renderer/mod.rs, native/shared/src/renderer/types.rs	Card-light and WSRC shaders convert texel-count bias to NDC depth units. Card-light cascade candidates use a 1.5-texel normal offset. The configured bias values and parameter documentation use shadow texels.
Transmitted shadow tolerances native/shared/src/renderer/transmitted_shadows.rs	The caster derives blocker tolerance from the light matrix and opaque-map width. Visibility derives receiver bias from the cascade matrix and transmitted-map width. The draw uniform is available to both vertex and fragment stages.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Bug fix

Merge Risk: Minimal · up to 204ea

This change makes shadow receiver bias scale with world units across several shadow readers. No concrete merge-blocking defect was found, and the author reports tests, with no golden image changes.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title names the main change: shadow receiver bias is now measured in world units for every directional cascade reader. The changeset matches this. It replaces fixed NDC biases with texel-based wor…
Description check		The description covers the summary, the affected readers, and the exceptions. It includes a before-and-after table, the test results, and a note on the known failing test. It follows the template's Su…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title names the main change: shadow receiver bias is now measured in world units for every directional cascade reader. The changeset matches this. It replaces fixed NDC biases with texel-based world-sized biases across GI card relight, WSRC bake, path-tracer sun visibility, planar reflections, transmitted shadows, and virtual shadow maps. The title is specific and does not use vague terms. It does not mention that the change is a follow-up to #199 or that it leaves hardware GI and perspective-depth biases unchanged, but it does not need to cover every detail.

Full details: Description check

Explanation

The description covers the summary, the affected readers, and the exceptions. It includes a before-and-after table, the test results, and a note on the known failing test. It follows the template's Summary and Test plan sections. It has several gaps. The template's checklist items are not all addressed: it does not state whether the change was tested on a second platform, whether docs were updated for the public API, or whether the change was run through ./scripts/ci-check.sh, although the author ran a quick variant. The Notes for reviewers section is not present as a heading, though the failing test and the unchanged hardware variants are described in the body. The CI box is unchecked. These gaps are non-critical, so the description is mostly complete.

Finishing Touches

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn

Recent review info

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: d056e906-f652-43a4-8d49-f39d15d581df

Commits

Reviewing files that changed from the base of the PR and between 49cf675 and 204eac2.

Files selected for processing: 9

• native/shared/shaders/virtual_shadows/material_helper.wgsl
• native/shared/shaders/virtual_shadows/scene_helper.wgsl
• native/shared/src/renderer/gi_bake.rs
• native/shared/src/renderer/mod.rs
• native/shared/src/renderer/shaders/env.rs
• native/shared/src/renderer/shaders/gi.rs
• native/shared/src/renderer/shaders/pt.rs
• native/shared/src/renderer/transmitted_shadows.rs
• native/shared/src/renderer/types.rs

Included review availability: This review used your included allowance. Your plan provides up to 2 included reviews per hour; 1 remain after this review.