Follow-up: Port facing_ok culling to world-space

[deevus]

This was generated by AI during triage.

Parent

#116 — PRD: Deepen game_render interface for GPU-ready world-space rendering

Status

This is now a non-blocking follow-up cleanup/deepening task, not a blocker for completing #116.

The original issue assumed that facing_ok was the last consumer of the DrawTrack3 screen-space projection pass. That is no longer accurate: DrawTrack3 still uses projected TrackScreenXYZ / GroundScreenXYZ data for clipping and render-depth bookkeeping, and func3.c still has repo-wide consumers of TrackScreenXYZ.

What to build

Port the five DrawTrack3 ground backface-culling call sites from legacy projected/view-space facing_ok(...) inputs to a world/camera-space helper such as facing_ok_world(...).

The helper should compute the same front/back-facing decision from world-space vertices, using the current camera position to make the determinant equivalent to the legacy camera-relative projected coordinates.

Acceptance criteria

• DrawTrack3 ground culling uses world-space vertices or tVec3 / GameRenderVertex inputs for the five current facing_ok decisions.
• Legacy facing_ok(...) is either replaced by facing_ok_world(...) in DrawTrack3 or clearly scoped as legacy-only.
• Culling results are visually unchanged: no new pop-in/pop-out at grazing angles.
• The change is covered by tests/scene_render_seam_check.py or equivalent static seam checks that prevent reintroducing projected TrackScreenXYZ/GroundScreenXYZ inputs to DrawTrack3 facing decisions.
• zig build and zig build test-snapshots pass.

Explicitly out of scope

• Removing the entire DrawTrack3 screen-space projection/depth bookkeeping pass.
• Removing TrackScreenXYZ[] / GroundScreenXYZ[] globals repo-wide.
• Refactoring func3.c legacy screen-space track rendering.
• Completing PRD: Deepen game_render interface for GPU-ready world-space rendering #116’s core game_render seam work.

Follow-up candidate

After #116 is complete, a later cleanup can replace DrawTrack3’s projected-depth and clip-count bookkeeping with explicit camera-space/world-space helpers, then remove the remaining DrawTrack3 projection pass if no longer needed.

[deevus]

This was generated by AI during triage.

Agent Brief

Category: enhancement
State: ready-for-agent

Summary: Port facing_ok / track backface culling from projected view-space data to world-space vertices so the remaining TrackScreenXYZ projection pass can be removed.

Current behavior:
Current feat/game-render still computes TrackScreenXYZ[] / GroundScreenXYZ[] projected coordinates in DrawTrack3, and facing_ok() still consumes projected/view-space values. Rendering itself now submits world-space quads, so this screen-space projection pass is retained mainly for culling and related legacy bookkeeping.

Desired behavior:

• Replace or update facing_ok() with a world-space equivalent using track point/world vertices and camera/view-forward data.
• Preserve existing culling decisions, especially at grazing angles.
• Remove the now-unused screen-space projection pass and globals once no consumers remain.

Verification focus:
Walls, ground, roof, road markings, and shoulders should render without pop-in/pop-out regressions. Run zig build and zig build test-snapshots.

Blocked by: None — #121 has landed.

[deevus]

This was generated by AI during triage.

Triage Update

#128 is no longer considered a blocker for #116.

What changed:

• facing_ok is not the final consumer of projected TrackScreenXYZ / GroundScreenXYZ data in DrawTrack3.
• DrawTrack3 still uses projected data for clip/depth bookkeeping.
• func3.c still has repo-wide TrackScreenXYZ consumers.
• The core PRD: Deepen game_render interface for GPU-ready world-space rendering #116 interface goal can proceed because scene geometry is already submitted through world-space game_render / scene_render seams.

Updated scope: keep this issue as a follow-up cleanup to port the five DrawTrack3 facing/culling decisions to world-space, without requiring projection-pass or global-array removal.