overlays: list hand-written level functions as original assembly

[xx255q]

Level code has functions written as assembly in the original, like the executable's config/handwritten_asm.txt. Until now it had no way to say so: they sit as INCLUDE_ASM stubs that can never be matched in C and are counted as unfinished. This adds the level-code counterpart, built the same way as the level linker remnants (config/overlays/linker_remnants.txt / LINKER_REMNANT).

docs/ASM_CLASSIFICATION.md notes that changing the published progress policy needs upstream review, so this is a proposal. Happy to narrow the list or change the approach.

What changes

• config/overlays/handwritten_asm.txt: 67 entries, covering 94 catalogue entries / 51,748 bytes once joined pieces are counted.
• src/overlays/: each entry's INCLUDE_ASM("asm/overlays", …) becomes ASM_FUNC("asm/overlays", …). ASM_FUNC is INCLUDE_ASM, so the image is unchanged.
• tools/gen_progress_report.py:
  • counts listed functions as finished, in both the build and --check;
  • checks the list against the ASM_FUNC lines and the catalogue, sharing the remnants' check (overlay_listed());
  • finishes later pieces of a listed joined function with it, as for C.
• tools/overlay_variants.py: its stub pattern accepts ASM_FUNC, as it already does LINKER_REMNANT.
• docs/ASM_CLASSIFICATION.md: a "Hand-written level functions" section with the rule and evidence.

Which functions
The executable's rule: every function the disassembler marks "Handwritten function" (87 entries, 67 after folding joined pieces into their first names). They hold instructions ee-gcc 2.9 doesn't emit: a trapping addi/teq, VU0 macro instructions, MMI, $at as a working register.

• Two joined functions are listed under first pieces the disassembler didn't mark. Their other pieces are marked, and the first pieces hold VU0/MMI themselves: func_L00_001FD1D8 (lqc2) and func_L00_00257F4C (prot3w).
• Two are listed by hand:
  • func_L00_00251328: a nop before jr $31 with the slot filled;
  • func_L00_00251358: $at as an ordinary register, and the jr slot left empty.

Cross-check against the NTSC-U decompilation (Lombyte, MIT)

• 87 of the 89 are assembly there too: INCLUDE_ASM stubs, or copies of executable functions it keeps under assembly/.
• It documents the two hand-listed ones as "hand-written in retail" for the same reasons.
• The other two have no NTSC-U counterpart: func_L00_0024EC20 and func_L01_00262BB0 are teq $0, $0, 0 trap stubs.

Every listed function already works in the native port
None of these are marked finished without working code behind them. Checked against OpenRAC's port/game/rac1-pal (no_c.tsv, hand/hand.tsv, libraries.tsv) and port/game/rac1/host:

• 48 of the 67 entries have port C here, in nonmatching/functional.tsv (the entry or one of its joined pieces).
• 13 are implemented by the port:
  • host/collision.c: func_L00_001EFFF0, func_L00_001F10E0, func_L00_001F1D20, func_L00_001F2BE8, func_L00_001F34F0;
  • host/collision.c + host/render_native.c: func_L00_00254218;
  • host/render_native.c: func_L00_001FE688, func_L00_002963D8;
  • host/anim.c: func_L00_00252D20;
  • host/effect_quad.c: func_L00_001FD1D8;
  • hand/: func_L02_00250A58, func_L00_002539C8;
  • behind a host function: func_L00_001EEBD8.
• 6 need no C of their own (no_c.tsv classes a/b): pieces of a function the port replaces whole, or unreachable. They are func_L00_001EF944, func_L00_001F080C, func_L00_001FEFE0, func_L00_002535C8, func_L00_00254A70 and func_L01_00262BB0.
  Left out: func_L00_002569F8 + func_L00_00257848, the level programs' copy of MobyProc (5,152 bytes, as func_00212658 is). The disassembler marks both, but the port does not run this copy yet (no_c.tsv: needs C). It replaces the executable's MobyProc in host/render_native.c, so they stay INCLUDE_ASM until the port does the same for the level copy.

Functions already done in C are untouched.

Checks

• tools/overlay_file_check.py on the 17 changed level files: 0 problems (ASM_FUNC lines assemble to the same bytes).
• progress/report.json regenerated: 4,904 → 4,998 of 5,121 functions finished (+94: the 67 entries and their joined pieces), code finished 95.14% → 96.53% (level code 95.39% → 96.98%).
• python tools/gen_progress_report.py --check passes.

🤖 Generated with Claude Code

[bytematchdev]

ByteMatch progress report

Code matched	96.53% (+1.39%) · 3.58 MB of 3.71 MB
Fuzzy match	97.17% (+1.39%)
Functions	4,998 (+94) / 5,121
Data matched	0.52%
Units complete	112 (+1) / 141

✅ Newly matched (94)

• func_L00_001F1D20 (3,784 bytes)
• func_L00_001EEBD8 (3,436 bytes)
• func_L00_00252018 (3,284 bytes)
• func_L00_001F10E0 (3,136 bytes)
• func_L00_00255B90 (2,664 bytes)
• func_L00_001F2BE8 (2,248 bytes)
• func_L00_00252D20 (2,216 bytes)
• func_L00_001F080C (2,212 bytes)
• func_L00_00254218 (2,136 bytes)
• func_L00_001EFFF0 (1,976 bytes)
• func_L00_002539C8 (1,896 bytes)
• func_L00_001EF944 (1,548 bytes)
• func_L00_001F34F0 (1,128 bytes)
• func_L00_00254A70 (1,040 bytes)
• func_L00_002535C8 (1,024 bytes)
• func_L06_00216B38 (940 bytes)
• func_L01_00262FE8 (848 bytes)
• func_L02_00250C78 (836 bytes)
• func_L02_00250720 (824 bytes)
• func_L01_0021F9C8 (784 bytes)
• func_L00_001FE688 (696 bytes)
• func_L01_00263338 (656 bytes)
• func_L00_001FD428 (640 bytes)
• func_L00_001FD1D8 (592 bytes)
• func_L00_002963D8 (584 bytes)
• func_L01_00262DA8 (576 bytes)
• func_L02_00250A58 (544 bytes)
• func_L00_001FDE48 (500 bytes)
• func_L00_0025805C (496 bytes)
• func_L00_00251B58 (448 bytes)
• …and 64 more

_{Commit 1619a1e · Ratchet & Clank (PAL) on ByteMatch}