Files
goldeneye-pc-port/docs/dev
James Dansereau 50d73c519c D217 (M-90): align texpool arena bases to 8 bytes on PC — fixes solid wrong-colour weapon/model CI tiles (#47)
* D217 (M-90): align texpool arena bases to 8 bytes on PC — fixes solid wrong-colour weapon/model CI tiles

Root cause: texAlignIndices pads CI index rows to ABSOLUTE 8-byte
boundaries, and model-DL G_LOADTLUT palette offsets are a formula that
only matches the real layout when the texpool base is 8-aligned. On N64
the memp bump allocations feeding these pools happened to be 8-multiples;
PC pointer bloat (D36/D37-class struct growth) made some STAGE-bank sizes
not, so weapon-pool bases land %8==4 (e.g. FACILITY). Every palette in a
misaligned pool then sits 4 bytes early inside its allocation while the
draw reads the N64 offset: the PP7 grip's count=1 TLUT lands on the next
tenant's 8-byte header slot (texnum=26 -> LE s16 -> 0x1A00 = the green),
and larger loads read a 2-entry zero pad then the real palette (the M-86
'leading zero pad' tell). BUNKER1 was clean only because its bases
happened to be aligned.

Fix: texInitPool rounds the arena base up to 8 under #ifdef PORT; end
unchanged (pool loses <=7 bytes). No-op on N64 where bases were already
aligned. ABI/layout-class edit per AGENTS.md rule 2 narrow exception —
no game logic touched.

Verified: FACILITY 188/188 decodes %8==0, draw reads pal0=0001 (was
1a00), 0 green pixels in frames 250-450 (pre-fix ~1300); BUNKER1 + DEPOT
still clean; full verify.sh sweep 21/21 PASS. Pending: human playtest
eyeball of the M-87 weapon matrix (closes defect 1). Defect 2 (CI cache
content-hash key, fix/d217-model-texture-palette) remains separate/parked.

findings.md: new section H 'D217 - M-90' + D217 row update; GRAPHICS-BACKLOG
D217 status.

* docs(D217): M-92 - human playtest of PR #47 confirms PP7/PPK/watch fixed; AK47 + NPC-held world-model weapons are a distinct, still-open white-texture residual

Claude-Session: https://claude.ai/code/session_016TaKy6AAsVUTfjkhrX6CjG

* docs(D217/D228): M-97 human playtest - green FIXED, AK47 white patches split out as D228

Second independent playtest (after M-92) of PR #47 on FACILITY, this
time with the branch merged up to date with main. Confirms the same
split the M-90 write-up predicted: PP7/PPK grip+hammer, watch face and
world-model green are all fixed by the texpool alignment; NPC-held
AK47 white missing-texture patches are untouched.

White is a different signature from wrong-colour -- a wrong palette
binds a real TLUT with wrong contents, white means the texture was
never decoded/uploaded at all -- so this is a separate mechanism, not
an incomplete fix. Split out as D228 so D217 can close on the green
mechanism it actually describes instead of carrying an unrelated open
failure mode indefinitely.
2026-09-11 18:06:03 -04:00
..
2026-09-01 19:28:59 -04:00