Entry 25

Two-sided model faces, and hunting the trash can

2026-10-1513mvideotooling

Filescrates/rugrats-render/src/models.rscrates/rugrats/examples/opening_shot.rscine_faces.rscrates/rugrats-hle/examples/cine_frames.rsdocs/formats/mod.md

The user reported that Chuckie vanishes for a frame when he crashes into the trash can. That search led to two fixes and one dead end.

Tools:

• The port's opening shot example clashed with the oracle's cine_shot: both build to the same target directory, so whichever built last ran. The port's is now opening_shot. It also takes a range (F DIR TO STEP) for contact sheets.
• cine_frames (oracle) writes the picture for every cinematic frame counter in a range in one run.

The search: I compared the port to the oracle frame by frame over records 170-240 and 280-700, plus every 20 records across the whole opening.

• The steady difference is about 12 (mean absolute per channel), from the shading the port does not draw.
• Every spike traced back to oracle drops. After a drop, the oracle's picture for a counter shows a record two or three behind. At 338 the oracle matches the port's 335.
• Chuckie's run, his fall, the can and the robot (records 500-545) match the oracle. Nothing vanishes in the records themselves, so the report has to come from live playback or from a moment I have not located.

Two things in the hand-written model renderer's face test (0x80010174) were missing in the port:

• Flag bit 0 skips NCLIP, so the face is drawn from both sides. The port culled back faces for every model, which loses the milk spray (model 953, flags 1) and hair strands seen from behind. These faces now go in both windings.
• Bits 8-9 pick the ordering-table depth: average (and dropped at 16000 or more), farthest, or nearest vertex, shifted right 2. This is documented for the draw-order work and not used yet.

Also, a template swap in the actor reader (1149 to 1130's model) does not apply to the opening: DB22 places neither template.