Entry 136

"Stream 15's opening: the black is the game's, and the stream lights' order"

2026-09-26rendertest

Filescrates/piney-stream/src/scene.rscrates/piney-stream/tests/stream.rscrates/piney-stream/tests/stream_fixture.txtcrates/piney-stream/examples/stream_probe.rstools/test_stream_rs.pydocs/engine/stream.mdBUGS.md

BUGS.md had "first few scenes of story:31 the camera clips the ground culling it out". The ending's first scene, str0580, looks like this in the port:

• the world goes black but for Kite and the stone figure;
• the sky comes back tilted across the top of the screen;
• the lower half stays black for about 200 frames.

133 showed that neither effect task draws before frame 1225, so the black had to come from the scene draw itself. This entry checks that draw against the game's, frame by frame. The black turns out to be the game's. The check also found a real bug, in the order of the scene's lights.

The game's draw

• The scene check. tools/test_stream_rs.py fixture runs the game's DecodeSetup, InitScene, DecodeFrameSection, SetView and layer draw in eemu. It covered streams 0, 106 and 2. It now also takes stream 15's str0580 to frame 420, and stream 6's first 60 frames.
• What it took.
  • str0580e uploads textures: sceGsSetDefLoadImage uses an MMI por the machine lacks, and the uploads touch the DMA registers. Those calls are stubbed. The draw list does not depend on them.
  • Kite's bones have spaces in their names (EXT_t0 r belt3). The fixture's draw and light lines now escape them as %20.
  • The file area runs from DATA up to SYS. str0580 to frame 300 already ran past SYS and overwrote ccSys's screen size: the first camera came out with its x and y rows equal. The guard now checks SYS, not HEAP0, and a scene file that does not fit goes in the heap.
• What it shows. The table is in docs/engine/stream.md, "Stream 15's opening".
  • At frames 61-70 the scene's own tracks fade the whole se1_5 set to 0: the sky, the clouds, the ground and the towers.
  • From frame 121 the sky and clouds fade back in. The ground and towers never come back.
  • At frames 316-404 the next set, se1_6, fades in with its own floor.
  • The scene's fifteen smoke effect objects keep pattern 0xffff and draw nothing, and neither file holds particles.
• Why the lower half is dark. The camera looks down at the figures from above, past the horizon.
  • The sky dome, MDL_se1_5bac1, is a closed shell down to a pole at z -4928.
  • Its texture runs from brown at the horizon to black at the bottom pole.
  • Every sky triangle on screen at frame 145 is wholly inside VU1's space, so none is dropped.
  • The port's frame 145 falls from an average (50, 90, 55) on its top row to (3, 2, 1) near the bottom, a gradient and not an edge.
• The divZ. Func_str0580 sets it to 2000, and the port clips at 1000. In frames 1-65 only one or two ground triangles on screen are not wholly inside, and their last vertex is nearer than 1000 either way. So divZ is not the cause either.

The lights

• The failure. With stream 15 in the check, the draw list, every object's transparency and matrix, and the camera matched. The lights did not. At frame 1 the game's slot 2 held omni light LGT_se1_5omn01 (709) and slot 1 the distant LGT_se1_5lig1 (728). The port had them the other way round.
• The cause.
  • ccCreateLight (0x001389b0) gives a distant light priority -1 and every other kind 0.
  • ccLightGrp::AddGrp (0x00139060) keeps the group in descending priority, equal priorities in the order added.
  • So the distant lights come after the omni ones. The port had kept only the file's reversed order.
  • docs/engine/stream.md also named ccCreateLight wrongly (0x00138960 is ccBbox_SetBox) and gave every light priority -1.
• The fix. Scene::new sorts the lights stably by that priority.
• What it changes. The order only matters when more than three lights answer, so stream 15's pictures do not change.
  • A survey over the streams' lit models: streams 3, 5, 6, 7, 8, 10, 11 and 16 pick different lights now. In 6, 7, 11 and 16 that happens in nearly every frame.
  • Stream 6 is in the check to prove it. With the old order its first step fails: the port lit the first object drawn with a white distant light where the game uses an omni at 0.22.

Checked

• scenes_play_as_the_game_plays_them now covers five scenes, 1347 frames. It compares the draw list, each object drawn (name, transparency, lwMatrix bit for bit), the camera (worst 2.8e-5) and the lights (worst 5e-7). Streams 0, 106 and 2's lines are unchanged.
• Shots. piney-game --mode story:31 --shot --every 30 after the fix is pixel-identical to 133's 180 shots. stream_shot before and after the light fix differs in streams 6, 7 and 11: Kite and Blackrose (stream 7) and Balmung (stream 11) are lit by other lights.
• stream_probe --all lists the nodes a frame does not draw too, with their disp and transparency.