jqueuniet/dm2-tools
1
0
Fork 0
Code Issues 2 Pull requests Projects Releases Packages Wiki Activity Actions

Palette pairing session 2: validate the lighting-stripe hypothesis via emulator captures #214

New issue
Closed
opened 2026-07-07 17:23:09 +02:00 by jqueuniet · 0 comments
jqueuniet commented 2026-07-07 17:23:09 +02:00 (Migrated from codeberg.org)
Copy link

Continuation of the static research merged in #213 (stripe-aware palette_oracle_match + cross-port docs). This issue carries the handoff so the next session can run on any machine with the fixtures staged — the working scratch directory from session 1 is machine-local and its artifacts are all re-derivable (commands below).

Goal

Confirm or refute, against fresh Sega CD CRAM captures:

CRAM line 1 == conv(bank[map_graphics_style(current map)][light_stripe])

where bank = the six 96-entry records at cls=(0x08, 0x00..0x05, 0x09, 0x00), map_graphics_style = dungeon.dat per-map w14 bits 7:4 (static; see docs/formats/dungeon-dat.md), and light_stripe = the engine's 6-state light level. Background and evidence: docs/formats/palette-lut.md § "The lighting-stripe hypothesis".

Protocol (Sega CD / BlastEm — harness in tooling/palette-oracle/, Phase-0 gate already passed)

  1. Build BlastEm via tooling/palette-oracle/blastem.nix; debugger commands are written up in tooling/palette-oracle/blastem-debugger.md.
  2. Capture 3–4 varied dungeon scenes with save/w <path>/cram_<scene>.bin vdp:cram 64 — and note per capture: the party's current map and the torch/light state. Map → environment via dm2 dump fixtures/dos-en/dungeon.dat (mapGraphicsStyle field; the console dungeons differ but carry the same field — dump the Sega CD dungeon for exact values).
  3. Run the stripe-aware matcher (landed in #213):
    cargo run -p dm2-graphics --example palette_oracle_match -- fixtures/sega-cd-us/graphics.dat <capture-dir>
    Expect line-1 hits labeled #idx stripe s with s tracking the light state and idx tracking map_graphics_style.
  4. If confirmed: console viewport colour is fully explained by static data + (map, light) — file the decoder-wiring work and skip the live-trace fallback.
  5. If refuted: fall back to the live ADDR_PALSEL trace — resolver 0x1e78e, policy sites 0x147d4 / 0xf6aa (CODE-hunk-relative; jump-table anchored), runtime entry previously seen at 0x375a0 (bias-relative, re-verify per boot). All documented in docs/formats/palette-lut.md.

Secondary target (either outcome): the per-object LUT selector

Static RE showed the cls3=0x0E LUT cls4 selector is computed per drawn object: cls4 = helper(record) via A5 jump-table slot 186 (Amiga offsets; see docs/formats/palette-lut.md § "Amiga cross-port corroboration" — the Amiga and Sega CD SKUL.FTL are the same C codebase, so either port works). Options:

  • Static: decode each module's BSS jump table (dm2-ftl parses BSS) to reconstruct the merged A5 table (merge order unconfirmed; slot 186 / the resolver at slot 337 likely land in GRF1.FTL), then read the helper's body.
  • Dynamic: breakpoint the policy site, log (object record, cls4, chosen LUT) triples; a handful of samples should pin the function.

On Amiga (single CPU — easiest correlation): breakpoint the VDEO palette loader (CODE 0x4ed6 + hunk bias) or watch the staging global -0xc3c(a5) in WinUAE/fs-uae.

FM-Towns: SKULL.EXP is a "P3" flat-386 RUN386 executable (decompilable with rz-ghidra x86); cold byte-scans hit only data tables, so start from MAME watchpoints on the Towns palette I/O and walk back.

Re-deriving the session-1 scratch artifacts

  • FTL code hunks (all ports): dm2 extract fixtures/amiga-eu/DM2/<MOD>.FTL -o <dir> (also tooling/palette-oracle/extract_code_hunk.sh for Sega CD SKUL).
  • DOS Beta joined archive: cat fixtures/dos-en-beta/DATA/GRAPHICS.DAT fixtures/dos-en-beta/DATA/GRAPHIC2.DAT > beta.dat; then dm2 extract beta.dat -o <dir> --all-bins reproduces the six banks (08_0?_09_00__*.gpl, 384 B) and the 3,357 cls3=0x0D Palette16 population analysed in docs/sources/dos-en-beta.md § "Palette architecture".
  • Amiga disassembly: rizin -N -a m68k -b 32 -e cfg.bigendian=true <hunk> — key offsets in docs/formats/palette-lut.md.

References

  • #213 — stripe-aware matcher + all findings docs (merged).
  • docs/formats/palette-lut.md — RE hub: hypothesis, offsets, capture architecture, Amiga/FM-Towns sections.
  • docs/formats/dungeon-dat.md — map_graphics_style semantics + engine citations.
  • docs/sources/dos-en-beta.md § "Palette architecture" — the static-pairing Rosetta stone.
  • ROADMAP § Phase 7 "Sega CD / Amiga / FM-Towns colour resolution" — the tracked residue: (a) this validation, (b) the cls4 helper, (c) decoder wiring.
Continuation of the static research merged in #213 (stripe-aware `palette_oracle_match` + cross-port docs). This issue carries the handoff so the next session can run on any machine with the fixtures staged — the working scratch directory from session 1 is machine-local and its artifacts are all re-derivable (commands below). ## Goal Confirm or refute, against fresh Sega CD CRAM captures: CRAM line 1 == conv(bank[map_graphics_style(current map)][light_stripe]) where `bank` = the six 96-entry records at `cls=(0x08, 0x00..0x05, 0x09, 0x00)`, `map_graphics_style` = `dungeon.dat` per-map `w14` bits 7:4 (static; see `docs/formats/dungeon-dat.md`), and `light_stripe` = the engine's 6-state light level. Background and evidence: `docs/formats/palette-lut.md` § "The lighting-stripe hypothesis". ## Protocol (Sega CD / BlastEm — harness in `tooling/palette-oracle/`, Phase-0 gate already passed) 1. Build BlastEm via `tooling/palette-oracle/blastem.nix`; debugger commands are written up in `tooling/palette-oracle/blastem-debugger.md`. 2. Capture 3–4 varied dungeon scenes with `save/w <path>/cram_<scene>.bin vdp:cram 64` — **and note per capture: the party's current map and the torch/light state.** Map → environment via `dm2 dump fixtures/dos-en/dungeon.dat` (`mapGraphicsStyle` field; the console dungeons differ but carry the same field — dump the Sega CD dungeon for exact values). 3. Run the stripe-aware matcher (landed in #213): `cargo run -p dm2-graphics --example palette_oracle_match -- fixtures/sega-cd-us/graphics.dat <capture-dir>` Expect line-1 hits labeled `#idx stripe s` with `s` tracking the light state and `idx` tracking `map_graphics_style`. 4. **If confirmed**: console viewport colour is fully explained by static data + (map, light) — file the decoder-wiring work and skip the live-trace fallback. 5. **If refuted**: fall back to the live `ADDR_PALSEL` trace — resolver `0x1e78e`, policy sites `0x147d4` / `0xf6aa` (CODE-hunk-relative; jump-table anchored), runtime entry previously seen at `0x375a0` (bias-relative, re-verify per boot). All documented in `docs/formats/palette-lut.md`. ## Secondary target (either outcome): the per-object LUT selector Static RE showed the `cls3=0x0E` LUT `cls4` selector is computed per drawn object: `cls4 = helper(record)` via A5 jump-table slot 186 (Amiga offsets; see `docs/formats/palette-lut.md` § "Amiga cross-port corroboration" — the Amiga and Sega CD `SKUL.FTL` are the same C codebase, so either port works). Options: - Static: decode each module's BSS jump table (`dm2-ftl` parses BSS) to reconstruct the merged A5 table (merge order unconfirmed; slot 186 / the resolver at slot 337 likely land in `GRF1.FTL`), then read the helper's body. - Dynamic: breakpoint the policy site, log `(object record, cls4, chosen LUT)` triples; a handful of samples should pin the function. On Amiga (single CPU — easiest correlation): breakpoint the `VDEO` palette loader (CODE `0x4ed6` + hunk bias) or watch the staging global `-0xc3c(a5)` in WinUAE/fs-uae. FM-Towns: `SKULL.EXP` is a "P3" flat-386 RUN386 executable (decompilable with rz-ghidra x86); cold byte-scans hit only data tables, so start from MAME watchpoints on the Towns palette I/O and walk back. ## Re-deriving the session-1 scratch artifacts - FTL code hunks (all ports): `dm2 extract fixtures/amiga-eu/DM2/<MOD>.FTL -o <dir>` (also `tooling/palette-oracle/extract_code_hunk.sh` for Sega CD `SKUL`). - DOS Beta joined archive: `cat fixtures/dos-en-beta/DATA/GRAPHICS.DAT fixtures/dos-en-beta/DATA/GRAPHIC2.DAT > beta.dat`; then `dm2 extract beta.dat -o <dir> --all-bins` reproduces the six banks (`08_0?_09_00__*.gpl`, 384 B) and the 3,357 `cls3=0x0D` Palette16 population analysed in `docs/sources/dos-en-beta.md` § "Palette architecture". - Amiga disassembly: `rizin -N -a m68k -b 32 -e cfg.bigendian=true <hunk>` — key offsets in `docs/formats/palette-lut.md`. ## References - #213 — stripe-aware matcher + all findings docs (merged). - `docs/formats/palette-lut.md` — RE hub: hypothesis, offsets, capture architecture, Amiga/FM-Towns sections. - `docs/formats/dungeon-dat.md` — `map_graphics_style` semantics + engine citations. - `docs/sources/dos-en-beta.md` § "Palette architecture" — the static-pairing Rosetta stone. - ROADMAP § Phase 7 "Sega CD / Amiga / FM-Towns colour resolution" — the tracked residue: (a) this validation, (b) the cls4 helper, (c) decoder wiring.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
pages
No results found.
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jqueuniet/dm2-tools#214
No description provided.