Entry 38

How the game draws a field: ground tiles, lighting, cover and object heights

2026-09-23worldrendertest

Filestools/field.pytools/test_field.pytools/field_tables.pytools/test_field_rs.pycrates/piney-data/src/fielddocs/engine/field.md

33 ported the field generator, but a viewer could not show a field yet. Two things were missing:

• how the height map becomes ground;
• where objects stand in z (the port estimated it bilinearly).

The user asked for fields in the viewer. I sent the field agent back to read the draw code instead of guessing, after the town placement lesson of 27. It extended tools/field.py and the Rust, and I re-ran its checks. What is true now is on the field page.

What it found

• The ground is a template.
  • Every chip draws one copy of the type's 24-vertex BaseMeshName model, translated only.
  • SetMESH2 rewrites only each vertex's z and colour, from the height and lit colour of the cell under it.
  • Which vertex goes to which cell was read by running SetMESH2. All ten types' ground models put their vertices exactly on those cells.
  • Chips under levelled objects and the lake draw nothing.
• The ground's colour is lighting, baked in.
  • The game builds per-cell face normals and averages six per vertex. One of the six is the wrong neighbour, quad[x+1][y-1]; the port keeps the bug, and a deliberate fix of it fails the comparison.
  • Each vertex is lit with the area's background light, from BGTBL and the bg_* file's frame-0 F_Ambient and F_DistantLight.
  • The same pass lights the object models in memory, so a clump that appears twice in a table is lit twice.
• Cover is a 6-vertex template given the absolute heights of its cells, 2.5 units above the ground.
• Object heights are a collision query. WORLD::GetHeight casts a segment into the cell's two triangles, made at the moment each object is placed. SetCover calls it for every cover, so the game's own Generate in eemu now spends most of its time there. The Python suite went from 95 to 167 seconds.
• Read from code, not run:
  • the torus and draw window: 12 × 12 chips around the player, fading from 6,700 to 7,200;
  • the lake's three water copies;
  • the background layers.

Checked

tools/test_field.py runs the game's functions in eemu. All match, most bit for bit:

• object z inside Generate: 5 Infection fields, plus 3 on each other volume;
• GetHeight at 178 points: on borders and on hidden chips;
• the whole map's normals and colours under one light: 6,400 of each;
• 20 ground tiles, 16 covers and 43 object vertex colours.

On integration:

• test_field.py: 8 tests OK (from the agent's run; its log is complete).
• test_field_rs.py: 7 tests, 0 mismatches. This covers object z over 220 random fields and 104 story areas, and 4 fields drawn in full: 6,000 tiles, 6,000 covers and 25,600 normals and colours.
• field_tables.py --check: current.
• cargo test -p piney-data --lib --test field: 20 and 9 passed.
• clippy: clean.

The agent's wait loops polled pgrep -f on a pattern that matched their own command line, so they could never finish. I stopped them; the test they waited for had passed.