[Fromnowheree]

What

Matches 7 functions: the two level-26 functions, and the five other copies of one of them, which had the same catalogue defect.

func_L26_003D4590 (928 bytes; USA v1.01 and v2.00, not in PAL)

• Moves to L26_003D4590-g8.c with -O2 -G8 -ffunction-sections -frac2-zero-ti-por, the retained draft's profile. The draft was already the right size.
• What closes it: the two random-range upper bounds, D_L26_001B0464 and D_L26_001B045C, are declared volatile MACRO_ADDR, as D_L02_001B00FC is in the level-2 state machine (L02_003C43F0-state-g8.c). Each range reads its upper bound absolutely, before the lower bound, and each is read once. Either declaration alone leaves it unmatched.
• D_L26_001B1748 is the int frame word that L26_003D4930-g8.c and L26_004070F8.c declare, tested through its low byte (the draft read it as a byte).
• Its functional draft and nonmatching/functional.tsv row go too, as tests/check_repository.py requires.

func_L26_003547D8 and its five copies (272-byte bodies, all three releases)

Function	Placements	Pointer to the real start
func_L26_003547D8	26	26:00286D80
func_L04_00378838	4 and 10 (shared)	04:0029FE90, 10:0029F960
func_L05_00367340	5	05:0029D4F0
func_L12_0035A838	12	12:00288180
func_L14_0035F510	14	14:0028D180
func_L20_0036AE68	20	20:00292390

• The defect: each catalogue extent began with the previous body's 16-byte tail (sw $t3,0x40($v1); nop; addiu $sp,$sp,0x20; nop). docs/MATCHING-BLOCKERS.md recorded it for the level 5, 12 and 20 rows. The repair was left unapplied there only because the PAL and v2.00 catalogues had to be carried. In every placement a data word points to the real start, and nothing points to the four leading words.
• The repair: named FUNCTION_ENTRY_OFFSETS entries, (288, 16) each, move the entries and keep the names and ends.
  • Regenerating the USA catalogue and running tools/carry.py --entries for PAL and v2.00 changes exactly these six rows in each release.
  • The catalogues are stable on replay.
• The C: each body equals func_L00_00354DF8's (whose level-0 entry was corrected the same way) up to linker-filled addresses. So each is that function's C with its level's own tables, in a -g8 file of its own beside its old unit, at the default flags. The shared row names level 4's tables.
• Cleanup: the drafts for the level-26, level-4 and level-14 rows and their README rows are removed. docs/MATCHING-BLOCKERS.md records the repair.

Worth a look:

• The two volatile declarations in func_L26_003D4590. An independent review disassembled the object: retail's lui $at/lwc1 for each upper bound comes before the lower bound's $gp load in the call's delay slot, and each upper bound is read once, so the qualifier invents no access.
• The catalogue edits. They touch only these six rows, in all three releases, through the existing tools.

Checked

On 7c3ea917 (on fe9a7138), on the compiler rebuilt with tools/compiler/build.sh (no compiler changes since), on all three releases:

• python3 tools/build.py all --strict with --version us, pal and us2: 28 programs OK on each.
• python3 tools/check.py <func> --version us|pal|us2 for all seven: every place matches in each release that has the function (func_L04_00378838: 2 of 2 on USA; func_L26_003D4590 is not in PAL).
• python3 tools/levels.py check: all 28 dumps equal config/programs.tsv.
• python -m unittest discover -s tests: 339 tests OK.
• python tests/check_repository.py: passed.
• python tools/progress.py --check: passed; the reports of us, pal and us2 agree.

progress/ and the README table are not touched.

[bytematchdev]

ByteMatch progress report

SCES_516.07

Code matched	54.15% · 3.09 MB of 5.70 MB
Fuzzy match	54.15%
Functions	6,996 / 8,672
Data matched	0.00%
Units complete	958 / 1,196

No function changes.

SCUS_972.68_v2

Code matched	81.01% · 4.52 MB of 5.58 MB
Fuzzy match	81.01%
Functions	7,595 / 8,634
Data matched	0.00%
Units complete	973 / 1,156

No function changes.

SCUS_972.68

Code matched	84.74% · 4.68 MB of 5.53 MB
Fuzzy match	84.74%
Functions	7,760 / 8,579
Data matched	0.00%
Units complete	924 / 1,018

No function changes.

_{Commit 7c3ea91 · Ratchet & Clank: Going Commando on ByteMatch}