[gelatinous-bullion]

What

Four of USA v2.00's own functions, each the C of a matched USA v1.01 function with v2.00's differences, plus one measured catalogue entry correction.

Function	Bytes	Places	From	v2.00's differences
func_US2_002CCAA0	376	boot + 27 levels	func_002CBAE0	three callees renamed
func_US2_002957C0	1,452	boot + 27 levels	func_00294F10	allocator source-line arguments +8; func_0029D4D0 is func_US2_0029E308; same -G64 la-addend flags
func_US2_L05_0041F478	3,516	level 05	func_L05_0041DF58	HUD text ids 0x16A lower; two callees renamed
func_US2_0026EDE8	2,348	boot	func_0026EDE8	skips resource table 1; pad pressed word at 0x1A4 (as elsewhere in v2.00); sixteen callees renamed; same call-used flags

All four are reserved in my #152 comment. (func_US2_L04_004127A8, also in that reservation, landed meanwhile in 32707af; this branch is rebased on 045f54e and keeps master's version.)

Catalogue entry: func_US2_0026EDE8

The v2.00 boot row stopped at 0x0026F70C, before the function's own jr $ra; addiu $sp,$sp,0xD0; no row held those two words, and a padding nop follows at 0x0026F714. PAL's copy had the same defect (func_PAL_0026EC50). A measured FUNCTION_END_EXTRA entry, "func_US2_0026EDE8": (2340, 8), gives them back; python3 tools/carry.py --version us2 --entries changed only that row (2340 -> 2348 bytes). This also adds 8 bytes to v2.00's total code.

volatile in func_US2_002CCAA0, for a ruling

func_US2_002CCAA0 matches only with extern volatile int D_001AB910 __attribute__((sda));, the same declaration master already uses in its v1.01 twin func_002CBAE0 (src/game/002C7748-g8.c) and in func_002CBC58 (src/game/002C7748-panels-g8.c, 62919eb). Retail rereads the word for every test and keeps one load whose value is discarded (the empty D_0025A058/D_00259B48 arm); without the qualifier the function is 328 bytes, not 376.

AGENTS.md reserves volatile for memory a second context writes. I found no such writer: the boot never stores to -0x36E0($gp), and 0x001AB910 lies in the level programs' data (0 in the boot image, 1 in level 05's). So this follows master's precedent but not the rule's wording. If the maintainers rule against it, the same ruling applies to the two functions on master, and I will move this one to nonmatching/ with its marker restored.

Checks (on 045f54e + this commit)

• tools/check.py --version us2: func_US2_002CCAA0 28 of 28, func_US2_002957C0 28 of 28, func_US2_L05_0041F478 1 of 1, func_US2_0026EDE8 1 of 1 places match.
• python3 tools/build.py all --strict --version us2: 28 programs OK (e.g. us2 boot: OK: 2523640 loaded bytes equal; 2664 C functions (896900 bytes)). Every function is in placed.json of each program that has it; no C rejected in any program.
• python3 tools/build.py all --strict (USA v1.01, since tools/levels.py changed): 28 programs OK.
• python3 tests/check_repository.py: repository checks passed.
• python3 tools/progress.py --check: the reports of us, pal, us2 agree on every function they share.
• python3 -m unittest discover -s tests: 355 tests; the 11 compiler-output modules need the 32-bit compiler, so they were run inside the rac2-build container (85 tests, OK); the rest passed on the host.
• Not run: PAL (no PAL disc). The catalogue entry is v2.00's alone and the C is in src/us2/.

Progress (local tools/progress.py --version us2, report not committed)

• After: 8,529 of 8,579 functions, 5,480,864 of 5,577,956 bytes (98.26%).
• This PR: +4 functions, +7,692 bytes (98.1216% -> 98.2594%, +0.1378 points against the same master).

🤖 Generated with Claude Code