mirror of
https://github.com/Druthulu/BFM-decomp
synced 2026-09-27 22:45:39 -04:00
chore(phase-35): CLOSE — Gen3's dedup phase: one source per unique function (3,175 shared headers, 3,173 registry groups incl. the h_text tier, the macro header gone, the five twins from one directory, S1 a tools-health gate with a second oracle, 14 macro-era tools frozen and 4 retired behind a guard, the published counts corrected) — v2.1.0
- PhaseEnd_Phase35.md written; CURRENT_PHASE.md archived to phase-ends/logs/Phase35.md (R19); DIGEST §0/§2 appended (step 3b) - T0–T8 + T5b across S94–S96 (S94 died at 91 % without a checkpoint; S95 recovered it from the transcript; S96 finished) - close: R22 clean fleet 218/218; tools-health OK (S1 10,180/10,180, 0 violations; dedup-check 3173/0; the guard 0) - rules: R96–R99 ratified at gate 1; candidates (a)–(g) proposed for Phase 36 gate 1 - v2.0.0 -> v2.1.0
This commit is contained in:
Drew T
@@ -0,0 +1,5 @@
|
||||
.venv/bin/python tools/tool_census.py --all
|
||||
tool_census: 295 tool files (find == git ls-files) + 38 retired; classes {'LIVE': 236, 'ORPHAN': 30, 'REFERENCED': 29, 'RETIRED': 38}; dictionary rows 333
|
||||
tool_census: wrote docs/tool-index.md
|
||||
tool_census: wrote decomp-architect/tools/MANIFEST.md
|
||||
tool_census: corpus materialised — 362 verbatim copies, 30 pointers, INDEX.md
|
||||
@@ -0,0 +1,5 @@
|
||||
.venv/bin/python tools/tool_census.py --all
|
||||
tool_census: 295 tool files (find == git ls-files) + 38 retired; classes {'LIVE': 236, 'ORPHAN': 30, 'REFERENCED': 29, 'RETIRED': 38}; dictionary rows 333
|
||||
tool_census: wrote docs/tool-index.md
|
||||
tool_census: wrote decomp-architect/tools/MANIFEST.md
|
||||
tool_census: corpus materialised — 362 verbatim copies, 30 pointers, INDEX.md
|
||||
@@ -0,0 +1,476 @@
|
||||
clean: removed build/, expected/, and the regenerated splat tree (asm/, assets/, undefined_*_auto.txt).
|
||||
disc-extract: up to date — extracted/retail/ matches the committed manifest (sha1 e5c6019249dec3c71bb2cf40658a78439dddb037)
|
||||
splat 0.41.0 (powered by spimdisasm 1.41.0)
|
||||
|
||||
0%| | 0/1096 [00:00<?, ?it/s]
|
||||
Loading symbols (symbols.us): 0%| | 0/1096 [00:00<?, ?it/s]
|
||||
Loading symbols (symbols.us): 100%|██████████| 1096/1096 [00:00<00:00, 214200.51it/s]
|
||||
|
||||
0%| | 0/47 [00:00<?, ?it/s]
|
||||
Loading symbols (symbols.us.ram): 0%| | 0/47 [00:00<?, ?it/s]
|
||||
Loading symbols (symbols.us.ram): 100%|██████████| 47/47 [00:00<00:00, 612212.07it/s]
|
||||
|
||||
0%| | 0/2 [00:00<?, ?it/s]
|
||||
Scanning header: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Scanning main: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Rodata segment '800' may belong to the text segment '800'
|
||||
Based on the usage from the function LzssDecodeSector to the symbol jtbl_80072A38
|
||||
|
||||
The rodata segment '800' has jumptables that are not aligned properly file-wise, indicating one or more likely file split.
|
||||
File split suggestions for this segment will follow in config yaml format:
|
||||
- [0x6324C, .rodata]
|
||||
- [0x63388, .rodata]
|
||||
- [0x633FC, .rodata]
|
||||
|
||||
|
||||
Rodata segment '800_b' may belong to the text segment '800_b'
|
||||
Based on the usage from the function func_8002B0B4 to the symbol jtbl_80072E44
|
||||
|
||||
Rodata segment '800_b_2' may belong to the text segment '800_b_2'
|
||||
Based on the usage from the function func_8002DC68 to the symbol jtbl_80072F3C
|
||||
|
||||
The rodata segment '800_b_2' has jumptables that are not aligned properly file-wise, indicating one or more likely file split.
|
||||
File split suggestions for this segment will follow in config yaml format:
|
||||
- [0x63920, .rodata]
|
||||
|
||||
|
||||
Rodata segment '800_c' may belong to the text segment '800_c'
|
||||
Based on the usage from the function func_80035270 to the symbol jtbl_800732A0
|
||||
|
||||
Scanning main: 100%|██████████| 2/2 [00:01<00:00, 1.62it/s]
|
||||
Scanning main: 100%|██████████| 2/2 [00:01<00:00, 1.62it/s]
|
||||
|
||||
0%| | 0/2 [00:00<?, ?it/s]
|
||||
Splitting header: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Splitting main: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Splitting main: 100%|██████████| 2/2 [00:00<00:00, 2.23it/s]
|
||||
Splitting main: 100%|██████████| 2/2 [00:00<00:00, 2.23it/s]
|
||||
|
||||
0%| | 0/2 [00:00<?, ?it/s]
|
||||
Linker script header: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Linker script main: 0%| | 0/2 [00:00<?, ?it/s]
|
||||
Linker script main: 100%|██████████| 2/2 [00:00<00:00, 91.31it/s]
|
||||
Split 340 KB (82.28%) in defined segments
|
||||
header: 2 KB (0.50%) 1 split, 0 cached
|
||||
c: 338 KB (81.78%) 85 split, 0 cached
|
||||
unknown: 0 B (0.00%) from unknown bin files
|
||||
ld_interleave --order: .main island = 8 pieces [53198.data.o, 800.o, 63470.data.o, 800_b.o, 800_b_2.o, 63940.data.o, 800_c.o, 63C4C.data.o]; pre=0 text=85 empties=166 bss=85
|
||||
[ OK ] resident
|
||||
[ OK ] ov_SC05_000
|
||||
[ OK ] ov_SC01_000
|
||||
[ OK ] ov_SC04_000
|
||||
[ OK ] ov_SC06_000
|
||||
[ OK ] ov_SC01_006
|
||||
[ OK ] ov_SC01_005
|
||||
[ OK ] ov_SC03_002
|
||||
[ OK ] ov_SC01_077
|
||||
[ OK ] ov_SC02_000
|
||||
[ OK ] ov_SC04_018
|
||||
[ OK ] ov_SC02_003
|
||||
[ OK ] ov_SC04_019
|
||||
[ OK ] ov_SC03_006
|
||||
[ OK ] ov_SC01_001
|
||||
[ OK ] ov_SC03_001
|
||||
[ OK ] ov_SC01_004
|
||||
[ OK ] ov_SC01_074
|
||||
[ OK ] ov_SC01_008
|
||||
[ OK ] ov_SC07_000
|
||||
[ OK ] ov_SC02_004
|
||||
[ OK ] ov_SC02_015
|
||||
[ OK ] ov_SC02_021
|
||||
[ OK ] ov_SC01_080
|
||||
[ OK ] ov_SC01_009
|
||||
[ OK ] ov_SC01_084
|
||||
[ OK ] ov_SC02_016
|
||||
[ OK ] ov_SC02_026
|
||||
[ OK ] ov_SC02_005
|
||||
[ OK ] ov_SC02_031
|
||||
[ OK ] ov_SC02_011
|
||||
[ OK ] ov_SC02_039
|
||||
[ OK ] ov_SC02_028
|
||||
[ OK ] ov_SC02_027
|
||||
[ OK ] ov_SC02_035
|
||||
[ OK ] ov_SC03_003
|
||||
[ OK ] ov_SC02_017
|
||||
[ OK ] ov_SC02_041
|
||||
[ OK ] ov_SC03_010
|
||||
[ OK ] ov_SC03_012
|
||||
[ OK ] ov_SC03_011
|
||||
[ OK ] ov_SC03_013
|
||||
[ OK ] ov_SC03_023
|
||||
[ OK ] ov_SC03_007
|
||||
[ OK ] ov_SC03_024
|
||||
[ OK ] ov_SC03_031
|
||||
[ OK ] ov_SC03_089
|
||||
[ OK ] ov_SC03_092
|
||||
[ OK ] ov_SC03_029
|
||||
[ OK ] ov_SC03_014
|
||||
[ OK ] ov_SC03_090
|
||||
[ OK ] ov_SC03_015
|
||||
[ OK ] ov_SC03_030
|
||||
[ OK ] ov_SC03_028
|
||||
[ OK ] ov_SC03_095
|
||||
[ OK ] ov_SC03_091
|
||||
[ OK ] ov_SC03_096
|
||||
[ OK ] ov_SC03_093
|
||||
[ OK ] ov_SC03_094
|
||||
[ OK ] ov_SC03_097
|
||||
[ OK ] ov_SC03_101
|
||||
[ OK ] ov_SC03_098
|
||||
[ OK ] ov_SC03_099
|
||||
[ OK ] ov_SC03_109
|
||||
[ OK ] ov_SC03_108
|
||||
[ OK ] ov_SC03_100
|
||||
[ OK ] ov_SC03_110
|
||||
[ OK ] ov_SC03_102
|
||||
[ OK ] ov_SC03_105
|
||||
[ OK ] ov_SC03_103
|
||||
[ OK ] ov_SC03_112
|
||||
[ OK ] ov_SC03_104
|
||||
[ OK ] ov_SC03_113
|
||||
[ OK ] ov_SC03_114
|
||||
[ OK ] ov_SC03_111
|
||||
[ OK ] ov_SC03_115
|
||||
[ OK ] ov_SC03_116
|
||||
[ OK ] ov_SC03_126
|
||||
[ OK ] ov_SC03_121
|
||||
[ OK ] ov_SC04_003
|
||||
[ OK ] ov_SC03_117
|
||||
[ OK ] ov_SC03_118
|
||||
[ OK ] ov_SC04_008
|
||||
[ OK ] ov_SC04_006
|
||||
[ OK ] ov_SC03_119
|
||||
[ OK ] ov_SC04_009
|
||||
[ OK ] ov_SC03_125
|
||||
[ OK ] ov_SC04_004
|
||||
[ OK ] ov_SC04_007
|
||||
[ OK ] ov_SC04_002
|
||||
[ OK ] ov_SC04_005
|
||||
[ OK ] ov_SC04_010
|
||||
[ OK ] ov_SC03_124
|
||||
[ OK ] ov_SC04_012
|
||||
[ OK ] ov_SC04_021
|
||||
[ OK ] ov_SC04_016
|
||||
[ OK ] ov_SC05_006
|
||||
[ OK ] ov_SC05_002
|
||||
[ OK ] ov_SC05_004
|
||||
[ OK ] ov_SC04_015
|
||||
[ OK ] ov_SC05_009
|
||||
[ OK ] ov_SC04_020
|
||||
[ OK ] ov_SC05_003
|
||||
[ OK ] ov_SC05_001
|
||||
[ OK ] ov_SC05_007
|
||||
[ OK ] ov_SC05_005
|
||||
[ OK ] ov_SC05_008
|
||||
[ OK ] ov_SC04_011
|
||||
[ OK ] ov_SC05_011
|
||||
[ OK ] ov_SC05_010
|
||||
[ OK ] ov_SC05_019
|
||||
[ OK ] ov_SC06_015
|
||||
[ OK ] ov_SC06_013
|
||||
[ OK ] ov_SC06_014
|
||||
[ OK ] ov_SC06_008
|
||||
[ OK ] ov_SC06_010
|
||||
[ OK ] ov_SC06_011
|
||||
[ OK ] ov_SC06_016
|
||||
[ OK ] ov_SC05_017
|
||||
[ OK ] ov_SC05_018
|
||||
[ OK ] ov_SC06_020
|
||||
[ OK ] ov_SC06_018
|
||||
[ OK ] ov_SC06_027
|
||||
[ OK ] ov_SC06_025
|
||||
[ OK ] ov_SC06_006
|
||||
[ OK ] ov_SC06_022
|
||||
[ OK ] ov_SC06_024
|
||||
[ OK ] md_MAIN_013
|
||||
[ OK ] ov_SC07_008
|
||||
[ OK ] md_MAIN_014
|
||||
[ OK ] md_MAIN_016
|
||||
[ OK ] md_MAIN_015
|
||||
[ OK ] ov_SC07_002
|
||||
[ OK ] ov_SC07_001
|
||||
[ OK ] md_MAIN_018
|
||||
[ OK ] md_MAIN_017
|
||||
[ OK ] ov_SC06_030
|
||||
[ OK ] md_MAIN_019
|
||||
[ OK ] md_MAIN_020
|
||||
[ OK ] md_MAIN_021
|
||||
[ OK ] md_MAIN_022
|
||||
[ OK ] md_MAIN_023
|
||||
[ OK ] md_MAIN_024
|
||||
[ OK ] md_MAIN_025
|
||||
[ OK ] md_MAIN_026
|
||||
[ OK ] md_MAIN_028
|
||||
[ OK ] md_MAIN_027
|
||||
[ OK ] md_MAIN_029
|
||||
[ OK ] md_MAIN_030
|
||||
[ OK ] ov_SC07_009
|
||||
[ OK ] md_MAIN_031
|
||||
[ OK ] ov_MAIN_012
|
||||
[ OK ] md_MAIN_033
|
||||
[ OK ] md_MAIN_032
|
||||
[ OK ] md_MAIN_035
|
||||
[ OK ] md_MAIN_034
|
||||
[ OK ] md_MAIN_036
|
||||
[ OK ] ov_SC06_033
|
||||
[ OK ] ov_SC07_011
|
||||
[ OK ] ov_SC06_032
|
||||
[ OK ] md_MAIN_037
|
||||
[ OK ] md_MAIN_039
|
||||
[ OK ] md_MAIN_040
|
||||
[ OK ] md_MAIN_038
|
||||
[ OK ] md_MAIN_041
|
||||
[ OK ] md_MAIN_042
|
||||
[ OK ] md_MAIN_043
|
||||
[ OK ] md_MAIN_044
|
||||
[ OK ] ov_SC06_029
|
||||
[ OK ] md_MAIN_045
|
||||
[ OK ] md_MAIN_046
|
||||
[ OK ] md_MAIN_047
|
||||
[ OK ] md_MAIN_008
|
||||
[ OK ] md_MAIN_011
|
||||
[ OK ] ov_SC07_010
|
||||
[ OK ] md_MAIN_001
|
||||
[ OK ] md_SC03_074
|
||||
[ OK ] md_SC03_073
|
||||
[ OK ] md_SC03_078
|
||||
[ OK ] md_SC03_075
|
||||
[ OK ] md_SC03_077
|
||||
[ OK ] ov_SC07_007
|
||||
[ OK ] md_SC03_076
|
||||
[ OK ] md_SC03_133
|
||||
[ OK ] md_SC03_132
|
||||
[ OK ] md_SC03_079
|
||||
[ OK ] md_SC03_137
|
||||
[ OK ] md_SC03_134
|
||||
[ OK ] md_SC03_136
|
||||
[ OK ] ov_SC03_107
|
||||
[ OK ] md_SC03_135
|
||||
[ OK ] md_SC04_025
|
||||
[ OK ] md_SC04_024
|
||||
[ OK ] md_SC03_138
|
||||
[ OK ] md_SC04_026
|
||||
[ OK ] md_MAIN_003
|
||||
[ OK ] md_SC04_029
|
||||
[ OK ] md_SC04_028
|
||||
[ OK ] md_SC04_027
|
||||
[ OK ] md_SC04_030
|
||||
[ OK ] md_SC05_023
|
||||
[ OK ] md_SC05_024
|
||||
[ OK ] md_SC05_025
|
||||
[ OK ] md_SC05_028
|
||||
[ OK ] md_MAIN_009
|
||||
[ OK ] md_MAIN_007
|
||||
[ OK ] md_SC05_027
|
||||
[ OK ] md_SC03_053
|
||||
[ OK ] ov_SC02_037
|
||||
[ OK ] md_SC05_026
|
||||
[ OK ] md_SC05_029
|
||||
[ OK ] md_SC03_054
|
||||
[ OK ] md_SC03_056
|
||||
[ OK ] md_SC02_009
|
||||
[ OK ] ov_SC07_006
|
||||
[ OK ] md_SC07_004
|
||||
[ OK ] md_SC07_003
|
||||
extract-all: 217 extracted, 0 failed of 217 (+ main, serial)
|
||||
[ OK ] resident
|
||||
[ OK ] ov_SC01_000
|
||||
[ OK ] ov_SC05_000
|
||||
[ OK ] ov_SC01_006
|
||||
[ OK ] ov_SC01_005
|
||||
[ OK ] ov_SC01_001
|
||||
[ OK ] ov_SC04_000
|
||||
[ OK ] ov_SC03_002
|
||||
[ OK ] ov_SC03_006
|
||||
[ OK ] ov_SC06_000
|
||||
[ OK ] ov_SC04_019
|
||||
[ OK ] ov_SC01_077
|
||||
[ OK ] ov_SC04_018
|
||||
[ OK ] ov_SC03_001
|
||||
[ OK ] ov_SC02_003
|
||||
[ OK ] ov_SC02_000
|
||||
[ OK ] ov_SC01_004
|
||||
[ OK ] ov_SC01_074
|
||||
[ OK ] ov_SC07_000
|
||||
[ OK ] ov_SC02_004
|
||||
[ OK ] ov_SC01_008
|
||||
[ OK ] ov_SC01_080
|
||||
[ OK ] ov_SC01_009
|
||||
[ OK ] ov_SC02_015
|
||||
[ OK ] ov_SC01_084
|
||||
[ OK ] ov_SC02_016
|
||||
[ OK ] ov_SC02_021
|
||||
[ OK ] ov_SC02_026
|
||||
[ OK ] ov_SC02_011
|
||||
[ OK ] ov_SC02_017
|
||||
[ OK ] ov_SC02_005
|
||||
[ OK ] ov_SC02_039
|
||||
[ OK ] ov_SC02_031
|
||||
[ OK ] ov_SC02_027
|
||||
[ OK ] ov_SC02_041
|
||||
[ OK ] ov_SC03_003
|
||||
[ OK ] ov_SC02_035
|
||||
[ OK ] ov_SC02_028
|
||||
[ OK ] ov_SC03_012
|
||||
[ OK ] ov_SC03_010
|
||||
[ OK ] ov_SC03_007
|
||||
[ OK ] ov_SC03_011
|
||||
[ OK ] ov_SC03_013
|
||||
[ OK ] ov_SC03_023
|
||||
[ OK ] ov_SC03_014
|
||||
[ OK ] ov_SC03_015
|
||||
[ OK ] ov_SC03_030
|
||||
[ OK ] ov_SC03_031
|
||||
[ OK ] ov_SC03_024
|
||||
[ OK ] ov_SC03_029
|
||||
[ OK ] ov_SC03_089
|
||||
[ OK ] ov_SC03_092
|
||||
[ OK ] ov_SC03_028
|
||||
[ OK ] ov_SC03_090
|
||||
[ OK ] ov_SC03_093
|
||||
[ OK ] ov_SC03_094
|
||||
[ OK ] ov_SC03_096
|
||||
[ OK ] ov_SC03_095
|
||||
[ OK ] ov_SC03_091
|
||||
[ OK ] main
|
||||
[ OK ] ov_SC03_098
|
||||
[ OK ] ov_SC03_097
|
||||
[ OK ] ov_SC03_100
|
||||
[ OK ] ov_SC03_101
|
||||
[ OK ] ov_SC03_099
|
||||
[ OK ] ov_SC03_102
|
||||
[ OK ] ov_SC03_109
|
||||
[ OK ] ov_SC03_103
|
||||
[ OK ] ov_SC03_108
|
||||
[ OK ] ov_SC03_110
|
||||
[ OK ] ov_SC03_104
|
||||
[ OK ] ov_SC03_111
|
||||
[ OK ] ov_SC03_113
|
||||
[ OK ] ov_SC03_105
|
||||
[ OK ] ov_SC03_114
|
||||
[ OK ] ov_SC03_112
|
||||
[ OK ] ov_SC03_115
|
||||
[ OK ] ov_SC03_116
|
||||
[ OK ] ov_SC03_117
|
||||
[ OK ] ov_SC03_118
|
||||
[ OK ] ov_SC03_126
|
||||
[ OK ] ov_SC03_119
|
||||
[ OK ] ov_SC03_121
|
||||
[ OK ] ov_SC03_125
|
||||
[ OK ] ov_SC04_003
|
||||
[ OK ] ov_SC04_006
|
||||
[ OK ] ov_SC04_008
|
||||
[ OK ] ov_SC04_004
|
||||
[ OK ] ov_SC03_124
|
||||
[ OK ] ov_SC04_002
|
||||
[ OK ] ov_SC04_007
|
||||
[ OK ] ov_SC04_005
|
||||
[ OK ] ov_SC04_009
|
||||
[ OK ] ov_SC04_010
|
||||
[ OK ] ov_SC04_012
|
||||
[ OK ] ov_SC04_021
|
||||
[ OK ] ov_SC04_016
|
||||
[ OK ] ov_SC04_015
|
||||
[ OK ] ov_SC04_020
|
||||
[ OK ] ov_SC04_011
|
||||
[ OK ] ov_SC05_003
|
||||
[ OK ] ov_SC05_006
|
||||
[ OK ] ov_SC05_002
|
||||
[ OK ] ov_SC05_001
|
||||
[ OK ] ov_SC05_005
|
||||
[ OK ] ov_SC05_004
|
||||
[ OK ] ov_SC05_008
|
||||
[ OK ] ov_SC05_007
|
||||
[ OK ] ov_SC05_009
|
||||
[ OK ] ov_SC05_010
|
||||
[ OK ] ov_SC05_011
|
||||
[ OK ] ov_SC05_019
|
||||
[ OK ] ov_SC05_018
|
||||
[ OK ] ov_SC06_008
|
||||
[ OK ] ov_SC06_015
|
||||
[ OK ] ov_SC06_006
|
||||
[ OK ] ov_SC06_013
|
||||
[ OK ] ov_SC06_011
|
||||
[ OK ] ov_SC06_014
|
||||
[ OK ] ov_SC06_010
|
||||
[ OK ] ov_SC06_016
|
||||
[ OK ] ov_SC06_020
|
||||
[ OK ] ov_SC05_017
|
||||
[ OK ] ov_SC06_018
|
||||
[ OK ] ov_SC06_022
|
||||
[ OK ] ov_SC06_024
|
||||
[ OK ] ov_SC06_025
|
||||
[ OK ] ov_SC06_027
|
||||
[ OK ] md_MAIN_013
|
||||
[ OK ] md_MAIN_014
|
||||
[ OK ] ov_SC07_008
|
||||
[ OK ] md_MAIN_015
|
||||
[ OK ] ov_SC07_009
|
||||
[ OK ] ov_SC07_001
|
||||
[ OK ] ov_SC06_030
|
||||
[ OK ] md_MAIN_017
|
||||
[ OK ] md_MAIN_016
|
||||
[ OK ] md_MAIN_018
|
||||
[ OK ] md_MAIN_019
|
||||
[ OK ] ov_SC07_002
|
||||
[ OK ] md_MAIN_020
|
||||
[ OK ] md_MAIN_021
|
||||
[ OK ] md_MAIN_022
|
||||
[ OK ] md_MAIN_023
|
||||
[ OK ] md_MAIN_024
|
||||
[ OK ] md_MAIN_025
|
||||
[ OK ] md_MAIN_026
|
||||
[ OK ] md_MAIN_028
|
||||
[ OK ] md_MAIN_027
|
||||
[ OK ] md_MAIN_029
|
||||
[ OK ] ov_SC07_010
|
||||
[ OK ] md_MAIN_030
|
||||
[ OK ] md_MAIN_031
|
||||
[ OK ] md_MAIN_032
|
||||
[ OK ] ov_SC06_029
|
||||
[ OK ] md_MAIN_033
|
||||
[ OK ] md_MAIN_034
|
||||
[ OK ] md_MAIN_035
|
||||
[ OK ] md_MAIN_036
|
||||
[ OK ] ov_SC07_007
|
||||
[ OK ] md_MAIN_037
|
||||
[ OK ] md_MAIN_039
|
||||
[ OK ] md_MAIN_038
|
||||
[ OK ] md_MAIN_040
|
||||
[ OK ] md_MAIN_042
|
||||
[ OK ] md_MAIN_041
|
||||
[ OK ] md_MAIN_043
|
||||
[ OK ] md_MAIN_045
|
||||
[ OK ] md_MAIN_044
|
||||
[ OK ] md_MAIN_046
|
||||
[ OK ] md_MAIN_001
|
||||
[ OK ] md_MAIN_047
|
||||
[ OK ] md_MAIN_008
|
||||
[ OK ] md_MAIN_011
|
||||
[ OK ] md_SC03_073
|
||||
[ OK ] ov_MAIN_012
|
||||
[ OK ] ov_SC07_006
|
||||
[ OK ] md_SC03_074
|
||||
[ OK ] md_SC03_075
|
||||
[ OK ] ov_SC06_032
|
||||
[ OK ] ov_SC06_033
|
||||
[ OK ] md_SC03_078
|
||||
[ OK ] md_SC03_077
|
||||
[ OK ] ov_SC07_011
|
||||
[ OK ] md_SC03_079
|
||||
[ OK ] md_SC03_076
|
||||
[ OK ] md_SC03_132
|
||||
[ OK ] md_SC03_133
|
||||
[ OK ] md_SC07_003
|
||||
[ OK ] md_SC03_134
|
||||
[ OK ] md_SC03_137
|
||||
[ OK ] md_SC03_138
|
||||
[ OK ] md_MAIN_003
|
||||
[ OK ] md_SC04_025
|
||||
[ OK ] md_SC04_024
|
||||
[ OK ] md_SC03_136
|
||||
[ OK ] md_SC04_029
|
||||
[ OK ] md_SC04_026
|
||||
[ OK ] md_SC03_135
|
||||
[ OK ] md_SC04_030
|
||||
[ OK ] md_SC05_023
|
||||
File diff suppressed because one or more lines are too long
@@ -31,7 +31,7 @@
|
||||
},
|
||||
"duplicate_text_classes": 6398,
|
||||
"duplicate_text_sites": 17779,
|
||||
"elapsed_s": 26.6,
|
||||
"elapsed_s": 22.6,
|
||||
"flags": {
|
||||
"ALIAS": 32,
|
||||
"E": 3826,
|
||||
|
||||
@@ -37,4 +37,4 @@ share_census: 218 binaries · 362,389 sig instances · 362,389 classified · 0 U
|
||||
B E,F 11852 x 2 0x80014554,0x8001455C,0x8001513C func_80014554,func_8001455c
|
||||
A - 141 x 167 0x80163C2C func_80163c2c
|
||||
A - 141 x 165 0x80178004 func_80178004
|
||||
macro sites 0 · duplicate-text classes 6,398 (17,779 sites) · elapsed 26.6 s
|
||||
macro sites 0 · duplicate-text classes 6,398 (17,779 sites) · elapsed 22.6 s
|
||||
|
||||
@@ -41,7 +41,13 @@ instrument findings fixed (cdecl's scratch dir, two wiki citations, kit_coverage
|
||||
**Pending by name:** the preset's creation by decomp.me's maintainers, a reply from Archipelago, frogress. Candidates (a)–(d) proposed. **Gen3 opens at Phase 35** — its seed now includes the S91-b types doctrine (R95, the canonical
|
||||
type layer).
|
||||
|
||||
**Phase 35 OPEN (2026-09-08, S94; gate 1 in plan mode at Max) — Gen3's first phase, the DEDUP phase: "one source per unique function."**
|
||||
**Phase 35 CLOSED 2026-09-08 (v2.1.0; `PhaseEnd_Phase35.md`) — Gen3's first phase, the DEDUP phase: "one source per unique function."**
|
||||
Three sessions (S94–S96) / 48 commits: every shared body one plain-C header under `src/shared/` included at each site (3,175 headers, the
|
||||
8.8 MB macro header gone), the five twins from one source directory, the same-address backlog 1,099 classes / 4,755 copies → 51 ledgered
|
||||
/ 160, a new `h_text` registry tier for 38 byte-variant functions, S1 a tools-health gate with a second oracle, 14 macro-era tools frozen +
|
||||
4 retired behind a guard, the published counts corrected (+459 functions; REAL 360,744 → 350,533 with EMPTY up the same). R96–R99 ratified;
|
||||
candidates (a)–(g) proposed. **Gen3 continues at Phase 36 — pins off** (Drew's order dedup → pins → structs → names). The open paragraph:
|
||||
*Phase 35 OPEN (2026-09-08, S94; gate 1 in plan mode at Max)*
|
||||
Drew set the Gen3 order dedup → pins → structs → names (one phase each, planned one at a time). The phase converts the 3,516 distinct
|
||||
`DEFINE_func_` macro bodies (5,147 define lines) into per-function plain-C headers under `src/shared/<space>/` included at each member's
|
||||
site (sotn's shape, verified from its tree), collapses the five identical-payload twin overlays onto one source directory each, shares
|
||||
@@ -193,6 +199,15 @@ dry-runs in a throwaway judged by script (run 5 PASS 20/20). Findings: the kit h
|
||||
the S91-b types hindsight (a banking + width lever → the canonical type layer at Phase 6, DK-65/G62); a typed verify figure wrong twice;
|
||||
four false judge flags in five runs, all the instrument's. Rules: R74–R83 ratified at gate 1; candidates (i)–(xii) proposed.
|
||||
|
||||
**P35 (2026-09-08, v2.1.0) GEN3 OPENS — THE DEDUP PHASE.** 3 sessions (S94–S96) / 48 commits. Gate 1 reversed the 2026-09-02 "leave the
|
||||
backlog" decision on sotn's tree (it shares once, by an include). T1 `share_census` (362,389 instances, 0 unaccounted, 10,180 classes);
|
||||
T2 the health chain learns the include + twin forms; T3 five twin pairs → one source dir (166 files deleted); T4 `macro_to_header`: 246,347
|
||||
sites → 2,215 headers, 213/213 binaries and 4,121/4,121 objects byte-identical, `engine_core.h` deleted; T5 `share_body` (bucket 0 183 →
|
||||
135 shared / 48 ledgered; bucket new 915 → 914 / 3; a re-exemplar), the S94 death at 91 % and the S95 transcript recovery, four tool
|
||||
defects fixed at their cause; T5b the `h_text` tier (38 classes / 2,030 sites); T6 14 frozen / 4 retired / the guard; T7 S1 strict + C2c/C2d
|
||||
in tools-health, the digest corrected; T8 the record (wiki, cookbook §453, decision log). Close: R22 218/218; S1 10,180/10,180, 0
|
||||
violations; fleet clean-run 157 s → 84–90 s. Rules: R96–R99 ratified; candidates (a)–(g).
|
||||
|
||||
## 3. Every rule, in full (transcribe these at session start; P/G/H/X are in PROJECT_CONTEXT.md)
|
||||
- **R1 — H1 relaxed while private.** ROM-derived content (`asm/`, `assets/`, `extracted/`, decompressed `.CD`) MAY be
|
||||
committed while the repo is private; accidental inclusion is not a violation. The raw multi-GB disc dump stays ignored
|
||||
|
||||
@@ -0,0 +1,205 @@
|
||||
# PhaseEnd — Phase 35: Gen3 opens — the dedup phase, "one source per unique function"
|
||||
**Date:** 2026-09-08 · **Project Version:** 2.1.0 · **Phase Status:** Complete (milestone: see §Milestone — every item with its literal output; gate 2 = Drew's close commit) · **Generation:** Gen3 (the first phase; Drew's Gen3 order: dedup → pins → structs → names, one phase each, planned one at a time)
|
||||
|
||||
> Chartered at gate 1 (S94, 2026-09-08, plan mode, Max) from `docs/gen3-handoff.md` + `docs/gen3-standards.md`, on evidence read from
|
||||
> sotn-decomp's tree that reversed the 2026-09-02 "leave the dedup backlog" decision. Ran **three sessions (S94–S96, all 2026-09-08)** and
|
||||
> **48 commits** (`48170fd7f..HEAD`, plus this close). The granular trail — every task's verify lines, the S94 death and the S95
|
||||
> recovery, every tool defect with its cause — is at **`phase-ends/logs/Phase35.md`** (R19, on demand). Owner decisions (Drew): gate 1's
|
||||
> four (twins → one source directory; cross-address classes censused and deferred; macro-era tools frozen with a loud refusal; R96–R99
|
||||
> ratified); no "trivial" exception; the tooling must know there is one source per unique function (→ the S1 gate); the recovery session's
|
||||
> scope ("capture only"); the effort levels per step (Max for two decisions delegated to Claude, high/xHigh otherwise); T9 run in the same
|
||||
> session at xHigh.
|
||||
|
||||
## Build Log
|
||||
|
||||
**The phase in one line:** every shared function body became one plain-C header included at each member's site (sotn's shape), the
|
||||
five identical-payload overlay twins build from one source directory, the 4,755 already-matched duplicate copies of 1,099 same-address
|
||||
classes were shared down to 160 copies in 51 ledgered classes, 38 functions whose bytes vary per overlay share one text through a new
|
||||
registry tier, the invariant "one source per unique function" is a `tools-health` gate with a second, disagreeing oracle, the macro-era
|
||||
tools are frozen or retired behind a guard, the published numbers were corrected where the old counting was blind — and every step was
|
||||
proven by the 218 hashes from a clean rebuild.
|
||||
|
||||
**Files created/changed and complete — do not recreate** (category summaries; the per-commit list is `git log 48170fd7f..HEAD` and the
|
||||
archived log):
|
||||
|
||||
- **The shape (T3, T4, T5, T5b).** `src/shared/engine_prelude.h` (what every overlay/module TU includes at line 2), `src/shared/{ov,main,
|
||||
slot_800CAE08,slot_801E25E8,slot_801EF468}/func_<VRAM>[__<h8>|__t<h8>].h` — **3,175 headers**, one per shared body (2,215 converted from
|
||||
the macro header at T4, 914 + 1 + 38 written by the share tool); `src/shared/engine_core.h`, `ov_setters.h`, `clearTbl40.h` deleted;
|
||||
the whale `func_80144B9C.h` moved under `ov/`; **3,173 registry groups** in `config/dedup.us.yaml` (2,220 at the open; 38 of them
|
||||
`tier: h_text` in the verbose member form with per-member `h_exact`); `config/dedup_exceptions.tsv` (the S1 ledger: 51 rows, each with
|
||||
the compiler's diagnostic and a reason code); `config/overlays.mk` `<twin>_TWIN_OF/_SRC_DIR` for SC01_006, SC03_119, SC02_003, SC04_019,
|
||||
SC03_015 (their `src/` directories deleted, 166 files; the twins' yamls = the primaries' carves with `create_c_files: False`, their
|
||||
`JTBL_PADS` lines regenerated); the Makefile's twin block.
|
||||
- **The tools (T1, T2, T4, T5, T5b, T6, T7).** `tools/share_census.py` (the S1 census + gate: every sig instance resolves to one source form;
|
||||
verdicts A/B/C/D/M, flags E/F/TWIN/PINS/ALIAS; `--selftest` fixture; `--check --strict-macros --strict-text`; `text_classes`; the second,
|
||||
sig-blind oracle keyed by (text, address)), `tools/macro_to_header.py` (the T4 converter, kept as `share_body`'s library), **`tools/
|
||||
share_body.py`** (the permanent share tool: `--plan / --apply --bucket extend|new|text / --reexemplar / --repair-registry`, the
|
||||
per-binary gate with object comparison and a snapshot-based bisect, the ledger, `--batches 1`), `tools/share_body_cycle.sh` (the
|
||||
unattended batch cycle: batch → exit code → phase-log entry → commit → periodic clean fleet run), `tools/frozen.py` (the one refusal
|
||||
14 FROZEN tools print from `main()`), `tool_census.py` (status FROZEN; the `ast` macro-form guard with `--guard-root` for the negative
|
||||
control), `dedup_integrate.py` (C2c/C2d; the `h_text` tier with C1 per member), `progress.py` (the include form as a definition, the
|
||||
source-dir oracle, the empty-shared fold, `unique_function_bodies`, `duplicate_source_copies/_classes`, `text_tier_*`,
|
||||
`deferred_same_address_*`, a dated `corrections` list, the README sentence), `corpus.py` (`src_dir/twin_of`, the one source-dir oracle),
|
||||
16 health-chain tools taught the include + twin forms (T2), 12 tools' dead macro branches dropped (T6); `tools/sunset/`: `dedup_propagate`,
|
||||
`dedup_extend` (their library surface moved verbatim into `share_body`), `macro_draft`, `test_reconcile_ledger`; the Makefile's
|
||||
`tools-health` gained `share_census --selftest` and the strict check before `report`.
|
||||
- **The record (T8).** `docs/wiki/The-dedup-engine.md` rewritten for the shared-source model; `Repository-layout.md`,
|
||||
`Where-the-project-goes-next.md`, `Verification-and-progress.md`, how-to ch.10, README prose and the generated block (with the
|
||||
"One source per unique function" sentence), `docs/gen3-standards.md` (§4 row struck, the DoD's "0 macro bodies" met), `docs/gen3-
|
||||
handoff.md` (dated snapshots), cookbook **§453** + a retirement banner on §14, `docs/decision-log.md` P35 (R31), `docs/accelerators.md`
|
||||
P35 (five), DIGEST §0/§2/§3/§4, `docs/SETUP.md` P35 sections (T1, T3–T5, T6, T7/T5b, the context guard), `docs/tool-index.md` and the
|
||||
kit corpus regenerated, the memory `dedup-backlog-leave-it` rewritten as REVERSED.
|
||||
- **Evidence (`.run/P35/`, allowlisted text):** `baseline/` (every R22 and tools-health log, the kit-corpus logs), `census/` (the census
|
||||
json/txt, `check_t5.txt`), `share/` (every batch record `batch_<label>.json`, every run log, the 146 per-binary gate logs, the registry
|
||||
repair record, the old header of the re-exemplar), `convert/` (T4's batch scripts, ledger and apply logs), `probe/` (T0's object A/B).
|
||||
- **Outside the repository:** `~/.claude/statusline.sh` writes the context percentage; `~/.claude/ctx_guard.sh` (a user-settings
|
||||
`PostToolUse` hook) injects "90% context. decide on the best place to checkpoint for a fresh session" once per session — built after
|
||||
S94 died at 91 % without a checkpoint; it fired in S96 exactly at the right moment.
|
||||
|
||||
**Tools/packages installed:** none (no apt/pip). Agents: three at gate 1 (two Explore, one Plan; ~680k tokens); none during execution —
|
||||
every task was the coordinator's own work and deterministic gates.
|
||||
|
||||
**Verification results (literal, at close — S96, HEAD `c1a901373` + this close):**
|
||||
- T9 R22 from clean: `extract-all: 217 extracted, 0 failed of 217 (+ main, serial)` · **`check-all: 218 passed, 0 failed of 218`** · `wall=87.62 s` · exit 0` (`.run/P35/baseline/r22_t9.log`).
|
||||
- T9 `make tools-health` (foreground): `tools-health: OK — sigs fresh; corpus(+resident) + cdecl + binaries + report(lint+dedup) + cookbook-index all green.` (run 2, 431 s, exit 0; run 1 was red only at the kit corpus, which had not yet copied this file and the DIGEST edit — regenerated, re-run) (`.run/P35/baseline/tools_health_t9.log`) carrying `S1: one source per unique function — 10,180 classes, 10,180 satisfied (3,507 twin-covered, 51 excepted, 3,801 deferred cross-address), 0 VIOLATION(S) — OK`, `dedup-check: 3173 validated, 0 failed | C1 coverage 262573/262573 members signed`,
|
||||
`macro-form guard: 0 LIVE tools reference the retired form (217 scanned, 14 frozen, 6 whitelisted detectors, 38 retired)`.
|
||||
- `git grep -c '^#define DEFINE_func_' -- src` → empty; `tools/macro_to_header.py --verify` → `OK — 0 macro sites, no macro header, the
|
||||
prelude present`; the census `macro sites 0` (two block-comment lines in prose still mention the token — comments are the record);
|
||||
`src/shared/engine_core.h` absent.
|
||||
- `tools/share_census.py --selftest` → `7/7 verdicts correct`; the strict check's second oracle: `0 same-address definition texts
|
||||
duplicated across TUs`, `381 … inside the deferred cross-address classes (1,668 sites)`.
|
||||
- `tools/progress.py --json --readme --check` → fresh; `timeline --check: fresh (74 rows)`; `doc_links` strict `0 pending, 0 broken`;
|
||||
`wiki_render --selftest: 12 cases, 0 failed; 32 pages, 0 unlisted`; `cookbook-index OK (1170 sections)`; `kit_coverage: OK`;
|
||||
`tool_census --check: OK` (the guard: 217 scanned, 14 frozen, 6 whitelisted detectors, 38 retired).
|
||||
- The negative controls on record (R39): the guard against a worktree of the pre-T4 tree flags exactly the 15 tools with macro-form code
|
||||
(`--guard-root`); one member's include reverted in place → `S1 … 1 VIOLATION(S) — FAIL` naming the class, C2d naming the member, restored → OK.
|
||||
|
||||
**Milestone achieved (gate 2 — Drew's close commit is the confirmation; each item with its literal output):**
|
||||
1. `git grep -c '^#define DEFINE_func_' -- src` empty; no `DEFINE_func_X()` site (`macro_to_header --verify` 0; the census 0);
|
||||
`src/shared/engine_core.h` absent; every registered body a plain-C header under `src/shared/<space>/` (`dedup-check` C2c, 3,173 groups).
|
||||
2. `tools/share_census.py --check --strict-macros --strict-text` exit 0: `S1: one source per unique function — 10,180 classes, 10,180
|
||||
satisfied (3,507 twin-covered, 51 excepted, 3,801 deferred cross-address), 0 VIOLATION(S) — OK` — same-vram duplicate copies 160, each
|
||||
in a ledgered class with its reason; the five twins built from one source directory each (`audit-binaries` CHECK 3b); the cross-vram
|
||||
classes published as a deferred count (3,801; plus the 381 tiny-body texts / 1,668 copies inside them).
|
||||
3. `make tools-health` OK with S1 strict and the macro-form guard; every consumer updated (16 + 12), frozen (14) or retired (4) per its
|
||||
dictionary row.
|
||||
4. `make clean && make extract-all && make check-all` → `check-all: 218 passed, 0 failed of 218` (R22, T9 and after every batch).
|
||||
5. README/wiki/kit regenerated (R75), decision log (R31), SETUP rows (R21), this PhaseEnd + the DIGEST; v2.1.0.
|
||||
|
||||
**Next:** **Phase 36 — pins off** (Drew's Gen3 order). Open a fresh session in **plan mode, Max** (Tier 1) from `docs/gen3-handoff.md`
|
||||
§5–§7 and `docs/gen3-standards.md`: the 44,243 `register … __asm__` pins (most come off byte-identical once the source shape is right —
|
||||
Phase 32's finding), family-batched, byte-gated per binary, R22 per batch; the S1 gate stays in the chain (a pin removed inside a shared
|
||||
header reaches every member — one edit, one gate). Inherited by name: the 51 ledgered declaration conflicts (the types phase), the 3,801
|
||||
cross-address classes and the 381 tiny-body texts (the names phase's parameterized form), the kit split, xsig v2.
|
||||
|
||||
## Roadmap delta (what Phase 35 changed about the road ahead)
|
||||
|
||||
- **The Gen3 charter's first item is met and measured:** "shared engine functions live as C, not as macros" — 0 macro bodies; the
|
||||
charter's "5,147" counted define lines (3,516 distinct macros; R14/R41 correction published in the log and the wiki).
|
||||
- **A new registry tier exists (`h_text`)** for the shape the byte tier cannot hold: identical C whose bytes vary per binary through
|
||||
the translation unit's declaration environment. The types phase must treat those 38 headers as compiling per unit by design.
|
||||
- **The names phase inherits a precise list:** 3,801 cross-address classes (E flag), 381 same-address tiny-body texts inside them, and
|
||||
the parameterized `SHARED_FN` form as the intended shape (one worked example, `clearTbl40`).
|
||||
- **The types phase inherits the ledger:** 50 declaration conflicts in the late-onboarded overlays (`config/dedup_exceptions.tsv`,
|
||||
each with its diagnostic) and one byte-variant body — exactly where a canonical declaration layer (R95) pays.
|
||||
- **Scale, honestly (R41):** the plan priced the phase at 7–8 sessions; it took three (one of them a recovery), because the tooling was
|
||||
built to gate itself and the batches ran unattended. The cost that was real: one dead session's 35 minutes rebuilt from its transcript,
|
||||
four tool defects found from their artifacts, and nine red health-chain runs across T6–T7, every one an instrument finding.
|
||||
|
||||
## Deviations
|
||||
|
||||
| Item | Plan (gate-1 plan) | Actual | Reason |
|
||||
|---|---|---|---|
|
||||
| Sessions | 7–8 | 3 (S94–S96) + this close | the gates were the tool's; batches ran unattended (`share_body_cycle.sh`) |
|
||||
| S94 | a checkpoint at every pause | died at 91 % context waiting on a background run, no checkpoint for T5; S95 rebuilt the state from the transcript with no runs (Drew's scope) | two bank commits carried no log line; rule candidate (b); the context guard built |
|
||||
| `share_body` | designed at T5 | four defects found from S94's artifacts and fixed at S96: the cause extractor (no "error" token in gcc 2.7.2), `git checkout` as restore, a twin's duplicate edit, wholesale registry extension | each caught by a gate or a known-true control; cookbook §453 |
|
||||
| Bucket 0's residue | ledgered with a message class | 48 late-overlay declaration conflicts ledgered TU-CONFLICT for the types phase; `E_func_80168B70` re-exemplared from the majority text (Claude's delegated decision at Max) | the header carried a 7-overlay minority spelling; the cast form compiles under both environments (141/141) |
|
||||
| The second oracle | "a second, disagreeing oracle" | it disagreed: 38 same-address texts / 2,030 copies byte-variant per binary → **T5b added** (the `h_text` tier; Claude's delegated decision at Max); 381 tiny-body texts inside the deferred classes stay deferred | no later phase would remove the copies; one header per address for the empties would be undone by the names phase's form |
|
||||
| T6's counts | freeze 7, retire 3 | freeze **14**, retire **4**; the guard's census measured 30 hits, not 22; `family_cousins` stays LIVE (`make atlas` runs it) with its dead branch dropped | the plan's counts came from a partial census |
|
||||
| `progress.py` | "the +219 correction" | +459 (SC03_015's 219 + SC03_118's 3 + the never-extended members of five groups) and REAL −10,211 → EMPTY (the fold); both dated in `progress.json` `corrections` | the include form made invisible sites visible; the macro-era fold had counted empty shared bodies as REAL |
|
||||
| T9 | Max, fresh session | run in S96 at xHigh (Drew) | owner's call; the checkpoint carried the brief |
|
||||
|
||||
## What we believed, what failed, and what we would do sooner (for the retrospective; detail in `docs/decision-log.md` P35)
|
||||
|
||||
- **Believed (2026-09-02):** sotn writes duplicate functions explicitly, so ~12,000 duplicate copies are the normal end state. **False:**
|
||||
read from its tree, sotn shares stage code once and instantiates it by an include. **Sooner:** read the target project's tree, never a
|
||||
memory of it, before setting a policy on its precedent.
|
||||
- **Believed:** refreshing the checkpoint at every task close was enough. **S94 refreshed it five times and still died with 35 minutes
|
||||
of state only in its transcript** — two bank commits inside a task carried no log line, and a background wait consumed the rest.
|
||||
**Sooner:** every commit that advances a task carries its log line and the headline; past ~85 % context, checkpoint before any long wait.
|
||||
- **Believed:** a green per-binary gate meant the tool's edits were what it reported. **Its restore was `git checkout`,** which wiped the
|
||||
previous batch's uncommitted shares and returned green; its cause lines read "Error 33" for 254 of 303 rejections. **Sooner:** a tool's
|
||||
restore path is part of the tool (R57); read a compiler's diagnostics by their form.
|
||||
- **Believed:** the byte join sees every duplicate. **The text oracle found 38 functions in 2,030 copies it could never class.** **Sooner:**
|
||||
build the disagreeing oracle before the batch runs, and measure the disagreement before believing either side (R34).
|
||||
- **Cost:** no model tokens beyond the coordinator and gate 1's three agents; the phase's expense was wall-clock in gates (≈2.5 h of
|
||||
`make check` and fleet runs) and the recovery (≈1 h reading a 6 MB transcript).
|
||||
|
||||
## Commit Message
|
||||
``
|
||||
(Phase 35 landed as 48 commits 48170fd7f..c1a901373, 2026-09-08 (S94–S96). This close = the archived log + this PhaseEnd + the DIGEST
|
||||
update + the kit's record copies, Drew's milestone-close commit.)
|
||||
|
||||
chore(phase-35): CLOSE — Gen3's dedup phase: one source per unique function (3,175 shared headers, 3,173 registry groups incl. the
|
||||
h_text tier, the macro header gone, the five twins from one directory, S1 a tools-health gate with a second oracle, 14 macro-era tools
|
||||
frozen and 4 retired behind a guard, the published counts corrected) — v2.1.0
|
||||
|
||||
- PhaseEnd_Phase35.md written; CURRENT_PHASE.md archived to phase-ends/logs/Phase35.md (R19); DIGEST §0/§2 appended (step 3b)
|
||||
- T0–T8 + T5b across S94–S96 (S94 died at 91 % without a checkpoint; S95 recovered it from the transcript; S96 finished)
|
||||
- close: R22 clean fleet 218/218; tools-health OK (S1 10,180/10,180, 0 violations; dedup-check 3173/0; the guard 0)
|
||||
- rules: R96–R99 ratified at gate 1; candidates (a)–(g) proposed for Phase 36 gate 1
|
||||
- v2.0.0 -> v2.1.0
|
||||
``
|
||||
|
||||
## Rules Added This Phase
|
||||
|
||||
**Ratified at gate 1 (2026-09-08):** **R96–R99** (the PhaseEnd_Phase34 candidates (a)–(d)); full text in `phase-ends/DIGEST.md` §3.
|
||||
|
||||
**Candidates recorded in the log, for Drew to accept, modify or reject at Phase 36 gate 1 (P10):**
|
||||
|
||||
| Rule | Reason |
|
||||
|---|---|
|
||||
| **(a) A shared body has exactly one source; a duplicate copy is a defect the health chain asserts, and its count is published with its rule.** | The phase's invariant (S1), a `tools-health` gate; 1,099 classes / 4,755 copies → 51 / 160, every remainder ledgered with its diagnostic. |
|
||||
| **(b) Every commit that advances a task — an intra-task bank included — carries its log line and the 🛑 headline; a checkpoint older than HEAD is a dead session's checkpoint.** | S94 refreshed the block at every task close, made two T5 bank commits with no log line, then died at 91 % context; S95 rebuilt 35 minutes of state from the transcript. |
|
||||
| **(c) A tool that restores files restores from its own snapshot, never `git checkout` on a tree it did not commit.** | `share_body`'s first bisect wiped the previous batch's uncommitted shares (~50 sites) and returned green. |
|
||||
| **(d) A failure-cause extractor is negative-controlled against the compiler's real message forms, not against the word "error".** | gcc 2.7.2 prints errors without it and exits 33; 254 of 303 rejection lines read "Error 33". |
|
||||
| **(e) Build the disagreeing oracle before the batch runs, and measure its disagreement before believing either side (R34 operated).** | The text oracle's 4,312 hits decomposed into twin false positives, the gate-1 deferral, and 38 genuine byte-variant functions — three different actions. |
|
||||
| **(f) A registry lists a member only after the gate has spoken for it; a listed member whose site does not share is a defect the check names.** | The first pass extended 788 members wholesale; 317 were private after the run (C2d exists now). |
|
||||
| **(g) A policy taken on a remembered precedent is a belief — read the target project's tree.** | The 2026-09-02 deferral rested on one parenthetical about a cross-jump idiom. |
|
||||
|
||||
## PhaseEnd Changelog
|
||||
**v2.0.0 → v2.1.0 — Phase 35 complete (Gen3's first phase).** Every shared function body in the fleet is one plain-C header included at
|
||||
each member's site, sotn's shape, verified from its tree; the 8.8 MB macro header and its 5,147 define lines are gone; the five overlay
|
||||
twins build from one source directory each; the same-address duplicate backlog fell from 1,099 classes and 4,755 private copies to 51
|
||||
ledgered classes and 160 copies, each with the compiler's own reason; 38 functions whose bytes vary per overlay share one text through a
|
||||
new registry tier; the invariant "one source per unique function" is asserted on every health-chain run by a census with a self-test and
|
||||
a second, sig-blind oracle, and by four registry checks; fourteen macro-era tools are frozen and four retired behind a guard that scans
|
||||
every live tool; the published numbers were corrected where the macro-era counting had been blind, dated and explained. Every batch was
|
||||
proven by the 218 hashes from a clean rebuild, which now takes 84–90 seconds instead of 157. R96–R99 ratified; seven candidates proposed.
|
||||
|
||||
## Plain-English Recap
|
||||
This phase made the code say each thing once. Before it, when the same function appeared in many of the game's 141 level files, our source
|
||||
held it as a macro in one enormous header, or as a private copy pasted into each level, thousands of times over. Now every shared function
|
||||
is one small ordinary C file, included wherever it is used, exactly the way the best-known PlayStation decompilation does it, and a check that
|
||||
runs with every health check proves that no function has two sources. Along the way the tool that does the sharing was caught making four
|
||||
kinds of mistake by its own gates and fixed; one session ran out of memory in the middle and the next rebuilt its state from the transcript;
|
||||
a second, independent check found 38 functions the first could never have seen, and they got their own way of being shared. Fifty-one
|
||||
functions could not be shared yet because their surrounding declarations disagree between level files, and they are listed with the
|
||||
compiler's own reason for the next phase to settle. The numbers on the public page were corrected where the old counting had been blind.
|
||||
All 218 game files still rebuild byte for byte, and faster than before.
|
||||
|
||||
## 🛑 Stop Here
|
||||
PhaseEnd written; `CURRENT_PHASE.md` archived → `phase-ends/logs/Phase35.md` (R19, via `git mv`); `phase-ends/DIGEST.md` §0/§2 appended
|
||||
(step 3b); the kit's record corpus regenerated (`make kit-corpus`) — **all left uncommitted for Drew's milestone-close commit** (R6 gate-2
|
||||
sign-off; the message is above). The Phase-35 work is committed across the sessions (`48170fd7f..c1a901373`; nothing pushed). The headless
|
||||
Ghidra MCP was stopped via the sentinel (R23); nothing under `ghidra/` is tracked. **Drew, from the repository root:**
|
||||
|
||||
``
|
||||
git add -A phase-ends decomp-architect docs .run/P35/baseline .run/P35/census
|
||||
git commit -m "chore(phase-35): CLOSE — Gen3's dedup phase: one source per unique function (3,175 shared headers, 3,173 registry groups incl. the h_text tier, the macro header gone, the five twins from one directory, S1 a tools-health gate with a second oracle, 14 macro-era tools frozen and 4 retired behind a guard, the published counts corrected) — v2.1.0"
|
||||
git tag -a v2.1.0 -m "v2.1.0 — Phase 35: one source per unique function; 218 binaries byte-identical"
|
||||
git push origin main --tags
|
||||
``
|
||||
|
||||
Gen3 continues — do **NOT** start Phase 36 here. Start a **fresh session** (effort **Max**, **plan mode**) for **Phase 36 — pins off**,
|
||||
planning from `docs/gen3-handoff.md` and `docs/gen3-standards.md`. Keep this file forever.
|
||||
@@ -89,6 +89,6 @@
|
||||
| 2026-09-05 | 83 | P31 (v1.30.0) | 218 | 100.0% | 100.0% | 100.0% | 97.7% | 4 | 787 / 2091 |
|
||||
| 2026-09-06 | 36 | P32 (v1.31.0) | 218 | 100.0% | 100.0% | 100.0% | 100.0% | 0 | 789 / 2091 |
|
||||
| 2026-09-07 | 61 | P33 (v1.32.0) | 218 | 100.0% | 100.0% | 100.0% | 100.0% | 0 | — |
|
||||
| 2026-09-08 | 72 | P33.5 (v1.32.1), P34 (v2.0.0) | 218 | 100.0% | 100.0% | 100.0% | 100.0% | 0 | — |
|
||||
| 2026-09-08 | 75 | P33.5 (v1.32.1), P34 (v2.0.0), P35 (v2.1.0) | 218 | 100.0% | 100.0% | 100.0% | 100.0% | 0 | — |
|
||||
|
||||
74 dated rows · phase ticks from the 35 PhaseEnds · the chart: `docs/story-timeline.svg`.
|
||||
74 dated rows · phase ticks from the 36 PhaseEnds · the chart: `docs/story-timeline.svg`.
|
||||
|
||||
@@ -50,6 +50,7 @@
|
||||
<line x1="969.3" y1="360" x2="969.3" y2="352" stroke="#888"/>
|
||||
<line x1="980.0" y1="360" x2="980.0" y2="352" stroke="#888"/>
|
||||
<line x1="980.0" y1="360" x2="980.0" y2="352" stroke="#888"/>
|
||||
<line x1="980.0" y1="360" x2="980.0" y2="352" stroke="#888"/>
|
||||
<line x1="466.5" y1="30" x2="466.5" y2="360" stroke="#c33" stroke-dasharray="4,3"/><text x="470.5" y="44" font-size="11" fill="#c33">2026-07-22: main enters the denominators</text>
|
||||
<polyline fill="none" stroke="#1f77b4" stroke-width="2" points="348.8,169.4 359.5,152.3 370.2,149.5 380.9,133.5 391.6,135.5 402.3,125.8 413.0,122.1 423.7,111.7 434.4,108.2 445.1,102.6 455.8,102.5 466.5,100.5 477.2,98.1 487.9,96.3 498.6,94.1 509.3,90.4 520.0,81.5 530.7,74.4 541.4,72.3 552.1,68.9 562.8,65.8 573.5,55.0 594.9,50.8 605.6,48.1 616.3,46.5 627.0,49.7 637.7,50.5 669.8,48.7 680.5,47.0 691.2,46.0 701.9,45.9 712.6,45.5 723.3,44.8 734.0,44.6 744.7,43.6 755.3,42.9 808.8,42.8 830.2,38.6 840.9,34.9 851.6,34.2 873.0,33.6 883.7,33.3 894.4,32.1 905.1,31.6 915.8,31.0 926.5,30.9 937.2,30.1 947.9,30.0 958.6,30.0 969.3,30.0 980.0,30.0"/><text x="70" y="352" font-size="12" fill="#1f77b4">instruction-weighted (all binaries)</text>
|
||||
<polyline fill="none" stroke="#2ca02c" stroke-width="2" points="348.8,263.2 359.5,230.9 370.2,225.5 380.9,197.6 391.6,196.8 402.3,184.2 413.0,181.7 423.7,159.9 434.4,152.9 445.1,140.5 455.8,140.4 466.5,136.8 477.2,136.7 487.9,136.7 498.6,134.8 509.3,131.5 520.0,114.5 530.7,106.2 541.4,103.4 552.1,100.4 562.8,94.4 573.5,77.8 594.9,69.5 605.6,65.5 616.3,64.8 627.0,70.9 637.7,72.4 669.8,69.1 680.5,65.6 691.2,63.7 701.9,63.3 712.6,62.7 723.3,61.3 734.0,60.8 744.7,58.5 755.3,57.0 808.8,56.9 830.2,47.8 840.9,40.3 851.6,38.8 873.0,37.9 883.7,37.4 894.4,34.8 905.1,33.8 915.8,32.2 926.5,32.0 937.2,30.3 947.9,30.1 958.6,30.0 969.3,30.0 980.0,30.0"/><text x="70" y="338" font-size="12" fill="#2ca02c">distinct code (each body once)</text>
|
||||
|
||||
|
Before Width: | Height: | Size: 6.7 KiB After Width: | Height: | Size: 6.8 KiB |
+16
-1
@@ -41,7 +41,13 @@ instrument findings fixed (cdecl's scratch dir, two wiki citations, kit_coverage
|
||||
**Pending by name:** the preset's creation by decomp.me's maintainers, a reply from Archipelago, frogress. Candidates (a)–(d) proposed. **Gen3 opens at Phase 35** — its seed now includes the S91-b types doctrine (R95, the canonical
|
||||
type layer).
|
||||
|
||||
**Phase 35 OPEN (2026-09-08, S94; gate 1 in plan mode at Max) — Gen3's first phase, the DEDUP phase: "one source per unique function."**
|
||||
**Phase 35 CLOSED 2026-09-08 (v2.1.0; `PhaseEnd_Phase35.md`) — Gen3's first phase, the DEDUP phase: "one source per unique function."**
|
||||
Three sessions (S94–S96) / 48 commits: every shared body one plain-C header under `src/shared/` included at each site (3,175 headers, the
|
||||
8.8 MB macro header gone), the five twins from one source directory, the same-address backlog 1,099 classes / 4,755 copies → 51 ledgered
|
||||
/ 160, a new `h_text` registry tier for 38 byte-variant functions, S1 a tools-health gate with a second oracle, 14 macro-era tools frozen +
|
||||
4 retired behind a guard, the published counts corrected (+459 functions; REAL 360,744 → 350,533 with EMPTY up the same). R96–R99 ratified;
|
||||
candidates (a)–(g) proposed. **Gen3 continues at Phase 36 — pins off** (Drew's order dedup → pins → structs → names). The open paragraph:
|
||||
*Phase 35 OPEN (2026-09-08, S94; gate 1 in plan mode at Max)*
|
||||
Drew set the Gen3 order dedup → pins → structs → names (one phase each, planned one at a time). The phase converts the 3,516 distinct
|
||||
`DEFINE_func_` macro bodies (5,147 define lines) into per-function plain-C headers under `src/shared/<space>/` included at each member's
|
||||
site (sotn's shape, verified from its tree), collapses the five identical-payload twin overlays onto one source directory each, shares
|
||||
@@ -193,6 +199,15 @@ dry-runs in a throwaway judged by script (run 5 PASS 20/20). Findings: the kit h
|
||||
the S91-b types hindsight (a banking + width lever → the canonical type layer at Phase 6, DK-65/G62); a typed verify figure wrong twice;
|
||||
four false judge flags in five runs, all the instrument's. Rules: R74–R83 ratified at gate 1; candidates (i)–(xii) proposed.
|
||||
|
||||
**P35 (2026-09-08, v2.1.0) GEN3 OPENS — THE DEDUP PHASE.** 3 sessions (S94–S96) / 48 commits. Gate 1 reversed the 2026-09-02 "leave the
|
||||
backlog" decision on sotn's tree (it shares once, by an include). T1 `share_census` (362,389 instances, 0 unaccounted, 10,180 classes);
|
||||
T2 the health chain learns the include + twin forms; T3 five twin pairs → one source dir (166 files deleted); T4 `macro_to_header`: 246,347
|
||||
sites → 2,215 headers, 213/213 binaries and 4,121/4,121 objects byte-identical, `engine_core.h` deleted; T5 `share_body` (bucket 0 183 →
|
||||
135 shared / 48 ledgered; bucket new 915 → 914 / 3; a re-exemplar), the S94 death at 91 % and the S95 transcript recovery, four tool
|
||||
defects fixed at their cause; T5b the `h_text` tier (38 classes / 2,030 sites); T6 14 frozen / 4 retired / the guard; T7 S1 strict + C2c/C2d
|
||||
in tools-health, the digest corrected; T8 the record (wiki, cookbook §453, decision log). Close: R22 218/218; S1 10,180/10,180, 0
|
||||
violations; fleet clean-run 157 s → 84–90 s. Rules: R96–R99 ratified; candidates (a)–(g).
|
||||
|
||||
## 3. Every rule, in full (transcribe these at session start; P/G/H/X are in PROJECT_CONTEXT.md)
|
||||
- **R1 — H1 relaxed while private.** ROM-derived content (`asm/`, `assets/`, `extracted/`, decompressed `.CD`) MAY be
|
||||
committed while the repo is private; accidental inclusion is not a violation. The raw multi-GB disc dump stays ignored
|
||||
|
||||
@@ -0,0 +1,205 @@
|
||||
# PhaseEnd — Phase 35: Gen3 opens — the dedup phase, "one source per unique function"
|
||||
**Date:** 2026-09-08 · **Project Version:** 2.1.0 · **Phase Status:** Complete (milestone: see §Milestone — every item with its literal output; gate 2 = Drew's close commit) · **Generation:** Gen3 (the first phase; Drew's Gen3 order: dedup → pins → structs → names, one phase each, planned one at a time)
|
||||
|
||||
> Chartered at gate 1 (S94, 2026-09-08, plan mode, Max) from `docs/gen3-handoff.md` + `docs/gen3-standards.md`, on evidence read from
|
||||
> sotn-decomp's tree that reversed the 2026-09-02 "leave the dedup backlog" decision. Ran **three sessions (S94–S96, all 2026-09-08)** and
|
||||
> **48 commits** (`48170fd7f..HEAD`, plus this close). The granular trail — every task's verify lines, the S94 death and the S95
|
||||
> recovery, every tool defect with its cause — is at **`phase-ends/logs/Phase35.md`** (R19, on demand). Owner decisions (Drew): gate 1's
|
||||
> four (twins → one source directory; cross-address classes censused and deferred; macro-era tools frozen with a loud refusal; R96–R99
|
||||
> ratified); no "trivial" exception; the tooling must know there is one source per unique function (→ the S1 gate); the recovery session's
|
||||
> scope ("capture only"); the effort levels per step (Max for two decisions delegated to Claude, high/xHigh otherwise); T9 run in the same
|
||||
> session at xHigh.
|
||||
|
||||
## Build Log
|
||||
|
||||
**The phase in one line:** every shared function body became one plain-C header included at each member's site (sotn's shape), the
|
||||
five identical-payload overlay twins build from one source directory, the 4,755 already-matched duplicate copies of 1,099 same-address
|
||||
classes were shared down to 160 copies in 51 ledgered classes, 38 functions whose bytes vary per overlay share one text through a new
|
||||
registry tier, the invariant "one source per unique function" is a `tools-health` gate with a second, disagreeing oracle, the macro-era
|
||||
tools are frozen or retired behind a guard, the published numbers were corrected where the old counting was blind — and every step was
|
||||
proven by the 218 hashes from a clean rebuild.
|
||||
|
||||
**Files created/changed and complete — do not recreate** (category summaries; the per-commit list is `git log 48170fd7f..HEAD` and the
|
||||
archived log):
|
||||
|
||||
- **The shape (T3, T4, T5, T5b).** `src/shared/engine_prelude.h` (what every overlay/module TU includes at line 2), `src/shared/{ov,main,
|
||||
slot_800CAE08,slot_801E25E8,slot_801EF468}/func_<VRAM>[__<h8>|__t<h8>].h` — **3,175 headers**, one per shared body (2,215 converted from
|
||||
the macro header at T4, 914 + 1 + 38 written by the share tool); `src/shared/engine_core.h`, `ov_setters.h`, `clearTbl40.h` deleted;
|
||||
the whale `func_80144B9C.h` moved under `ov/`; **3,173 registry groups** in `config/dedup.us.yaml` (2,220 at the open; 38 of them
|
||||
`tier: h_text` in the verbose member form with per-member `h_exact`); `config/dedup_exceptions.tsv` (the S1 ledger: 51 rows, each with
|
||||
the compiler's diagnostic and a reason code); `config/overlays.mk` `<twin>_TWIN_OF/_SRC_DIR` for SC01_006, SC03_119, SC02_003, SC04_019,
|
||||
SC03_015 (their `src/` directories deleted, 166 files; the twins' yamls = the primaries' carves with `create_c_files: False`, their
|
||||
`JTBL_PADS` lines regenerated); the Makefile's twin block.
|
||||
- **The tools (T1, T2, T4, T5, T5b, T6, T7).** `tools/share_census.py` (the S1 census + gate: every sig instance resolves to one source form;
|
||||
verdicts A/B/C/D/M, flags E/F/TWIN/PINS/ALIAS; `--selftest` fixture; `--check --strict-macros --strict-text`; `text_classes`; the second,
|
||||
sig-blind oracle keyed by (text, address)), `tools/macro_to_header.py` (the T4 converter, kept as `share_body`'s library), **`tools/
|
||||
share_body.py`** (the permanent share tool: `--plan / --apply --bucket extend|new|text / --reexemplar / --repair-registry`, the
|
||||
per-binary gate with object comparison and a snapshot-based bisect, the ledger, `--batches 1`), `tools/share_body_cycle.sh` (the
|
||||
unattended batch cycle: batch → exit code → phase-log entry → commit → periodic clean fleet run), `tools/frozen.py` (the one refusal
|
||||
14 FROZEN tools print from `main()`), `tool_census.py` (status FROZEN; the `ast` macro-form guard with `--guard-root` for the negative
|
||||
control), `dedup_integrate.py` (C2c/C2d; the `h_text` tier with C1 per member), `progress.py` (the include form as a definition, the
|
||||
source-dir oracle, the empty-shared fold, `unique_function_bodies`, `duplicate_source_copies/_classes`, `text_tier_*`,
|
||||
`deferred_same_address_*`, a dated `corrections` list, the README sentence), `corpus.py` (`src_dir/twin_of`, the one source-dir oracle),
|
||||
16 health-chain tools taught the include + twin forms (T2), 12 tools' dead macro branches dropped (T6); `tools/sunset/`: `dedup_propagate`,
|
||||
`dedup_extend` (their library surface moved verbatim into `share_body`), `macro_draft`, `test_reconcile_ledger`; the Makefile's
|
||||
`tools-health` gained `share_census --selftest` and the strict check before `report`.
|
||||
- **The record (T8).** `docs/wiki/The-dedup-engine.md` rewritten for the shared-source model; `Repository-layout.md`,
|
||||
`Where-the-project-goes-next.md`, `Verification-and-progress.md`, how-to ch.10, README prose and the generated block (with the
|
||||
"One source per unique function" sentence), `docs/gen3-standards.md` (§4 row struck, the DoD's "0 macro bodies" met), `docs/gen3-
|
||||
handoff.md` (dated snapshots), cookbook **§453** + a retirement banner on §14, `docs/decision-log.md` P35 (R31), `docs/accelerators.md`
|
||||
P35 (five), DIGEST §0/§2/§3/§4, `docs/SETUP.md` P35 sections (T1, T3–T5, T6, T7/T5b, the context guard), `docs/tool-index.md` and the
|
||||
kit corpus regenerated, the memory `dedup-backlog-leave-it` rewritten as REVERSED.
|
||||
- **Evidence (`.run/P35/`, allowlisted text):** `baseline/` (every R22 and tools-health log, the kit-corpus logs), `census/` (the census
|
||||
json/txt, `check_t5.txt`), `share/` (every batch record `batch_<label>.json`, every run log, the 146 per-binary gate logs, the registry
|
||||
repair record, the old header of the re-exemplar), `convert/` (T4's batch scripts, ledger and apply logs), `probe/` (T0's object A/B).
|
||||
- **Outside the repository:** `~/.claude/statusline.sh` writes the context percentage; `~/.claude/ctx_guard.sh` (a user-settings
|
||||
`PostToolUse` hook) injects "90% context. decide on the best place to checkpoint for a fresh session" once per session — built after
|
||||
S94 died at 91 % without a checkpoint; it fired in S96 exactly at the right moment.
|
||||
|
||||
**Tools/packages installed:** none (no apt/pip). Agents: three at gate 1 (two Explore, one Plan; ~680k tokens); none during execution —
|
||||
every task was the coordinator's own work and deterministic gates.
|
||||
|
||||
**Verification results (literal, at close — S96, HEAD `c1a901373` + this close):**
|
||||
- T9 R22 from clean: `extract-all: 217 extracted, 0 failed of 217 (+ main, serial)` · **`check-all: 218 passed, 0 failed of 218`** · `wall=87.62 s` · exit 0` (`.run/P35/baseline/r22_t9.log`).
|
||||
- T9 `make tools-health` (foreground): `tools-health: OK — sigs fresh; corpus(+resident) + cdecl + binaries + report(lint+dedup) + cookbook-index all green.` (run 2, 431 s, exit 0; run 1 was red only at the kit corpus, which had not yet copied this file and the DIGEST edit — regenerated, re-run) (`.run/P35/baseline/tools_health_t9.log`) carrying `S1: one source per unique function — 10,180 classes, 10,180 satisfied (3,507 twin-covered, 51 excepted, 3,801 deferred cross-address), 0 VIOLATION(S) — OK`, `dedup-check: 3173 validated, 0 failed | C1 coverage 262573/262573 members signed`,
|
||||
`macro-form guard: 0 LIVE tools reference the retired form (217 scanned, 14 frozen, 6 whitelisted detectors, 38 retired)`.
|
||||
- `git grep -c '^#define DEFINE_func_' -- src` → empty; `tools/macro_to_header.py --verify` → `OK — 0 macro sites, no macro header, the
|
||||
prelude present`; the census `macro sites 0` (two block-comment lines in prose still mention the token — comments are the record);
|
||||
`src/shared/engine_core.h` absent.
|
||||
- `tools/share_census.py --selftest` → `7/7 verdicts correct`; the strict check's second oracle: `0 same-address definition texts
|
||||
duplicated across TUs`, `381 … inside the deferred cross-address classes (1,668 sites)`.
|
||||
- `tools/progress.py --json --readme --check` → fresh; `timeline --check: fresh (74 rows)`; `doc_links` strict `0 pending, 0 broken`;
|
||||
`wiki_render --selftest: 12 cases, 0 failed; 32 pages, 0 unlisted`; `cookbook-index OK (1170 sections)`; `kit_coverage: OK`;
|
||||
`tool_census --check: OK` (the guard: 217 scanned, 14 frozen, 6 whitelisted detectors, 38 retired).
|
||||
- The negative controls on record (R39): the guard against a worktree of the pre-T4 tree flags exactly the 15 tools with macro-form code
|
||||
(`--guard-root`); one member's include reverted in place → `S1 … 1 VIOLATION(S) — FAIL` naming the class, C2d naming the member, restored → OK.
|
||||
|
||||
**Milestone achieved (gate 2 — Drew's close commit is the confirmation; each item with its literal output):**
|
||||
1. `git grep -c '^#define DEFINE_func_' -- src` empty; no `DEFINE_func_X()` site (`macro_to_header --verify` 0; the census 0);
|
||||
`src/shared/engine_core.h` absent; every registered body a plain-C header under `src/shared/<space>/` (`dedup-check` C2c, 3,173 groups).
|
||||
2. `tools/share_census.py --check --strict-macros --strict-text` exit 0: `S1: one source per unique function — 10,180 classes, 10,180
|
||||
satisfied (3,507 twin-covered, 51 excepted, 3,801 deferred cross-address), 0 VIOLATION(S) — OK` — same-vram duplicate copies 160, each
|
||||
in a ledgered class with its reason; the five twins built from one source directory each (`audit-binaries` CHECK 3b); the cross-vram
|
||||
classes published as a deferred count (3,801; plus the 381 tiny-body texts / 1,668 copies inside them).
|
||||
3. `make tools-health` OK with S1 strict and the macro-form guard; every consumer updated (16 + 12), frozen (14) or retired (4) per its
|
||||
dictionary row.
|
||||
4. `make clean && make extract-all && make check-all` → `check-all: 218 passed, 0 failed of 218` (R22, T9 and after every batch).
|
||||
5. README/wiki/kit regenerated (R75), decision log (R31), SETUP rows (R21), this PhaseEnd + the DIGEST; v2.1.0.
|
||||
|
||||
**Next:** **Phase 36 — pins off** (Drew's Gen3 order). Open a fresh session in **plan mode, Max** (Tier 1) from `docs/gen3-handoff.md`
|
||||
§5–§7 and `docs/gen3-standards.md`: the 44,243 `register … __asm__` pins (most come off byte-identical once the source shape is right —
|
||||
Phase 32's finding), family-batched, byte-gated per binary, R22 per batch; the S1 gate stays in the chain (a pin removed inside a shared
|
||||
header reaches every member — one edit, one gate). Inherited by name: the 51 ledgered declaration conflicts (the types phase), the 3,801
|
||||
cross-address classes and the 381 tiny-body texts (the names phase's parameterized form), the kit split, xsig v2.
|
||||
|
||||
## Roadmap delta (what Phase 35 changed about the road ahead)
|
||||
|
||||
- **The Gen3 charter's first item is met and measured:** "shared engine functions live as C, not as macros" — 0 macro bodies; the
|
||||
charter's "5,147" counted define lines (3,516 distinct macros; R14/R41 correction published in the log and the wiki).
|
||||
- **A new registry tier exists (`h_text`)** for the shape the byte tier cannot hold: identical C whose bytes vary per binary through
|
||||
the translation unit's declaration environment. The types phase must treat those 38 headers as compiling per unit by design.
|
||||
- **The names phase inherits a precise list:** 3,801 cross-address classes (E flag), 381 same-address tiny-body texts inside them, and
|
||||
the parameterized `SHARED_FN` form as the intended shape (one worked example, `clearTbl40`).
|
||||
- **The types phase inherits the ledger:** 50 declaration conflicts in the late-onboarded overlays (`config/dedup_exceptions.tsv`,
|
||||
each with its diagnostic) and one byte-variant body — exactly where a canonical declaration layer (R95) pays.
|
||||
- **Scale, honestly (R41):** the plan priced the phase at 7–8 sessions; it took three (one of them a recovery), because the tooling was
|
||||
built to gate itself and the batches ran unattended. The cost that was real: one dead session's 35 minutes rebuilt from its transcript,
|
||||
four tool defects found from their artifacts, and nine red health-chain runs across T6–T7, every one an instrument finding.
|
||||
|
||||
## Deviations
|
||||
|
||||
| Item | Plan (gate-1 plan) | Actual | Reason |
|
||||
|---|---|---|---|
|
||||
| Sessions | 7–8 | 3 (S94–S96) + this close | the gates were the tool's; batches ran unattended (`share_body_cycle.sh`) |
|
||||
| S94 | a checkpoint at every pause | died at 91 % context waiting on a background run, no checkpoint for T5; S95 rebuilt the state from the transcript with no runs (Drew's scope) | two bank commits carried no log line; rule candidate (b); the context guard built |
|
||||
| `share_body` | designed at T5 | four defects found from S94's artifacts and fixed at S96: the cause extractor (no "error" token in gcc 2.7.2), `git checkout` as restore, a twin's duplicate edit, wholesale registry extension | each caught by a gate or a known-true control; cookbook §453 |
|
||||
| Bucket 0's residue | ledgered with a message class | 48 late-overlay declaration conflicts ledgered TU-CONFLICT for the types phase; `E_func_80168B70` re-exemplared from the majority text (Claude's delegated decision at Max) | the header carried a 7-overlay minority spelling; the cast form compiles under both environments (141/141) |
|
||||
| The second oracle | "a second, disagreeing oracle" | it disagreed: 38 same-address texts / 2,030 copies byte-variant per binary → **T5b added** (the `h_text` tier; Claude's delegated decision at Max); 381 tiny-body texts inside the deferred classes stay deferred | no later phase would remove the copies; one header per address for the empties would be undone by the names phase's form |
|
||||
| T6's counts | freeze 7, retire 3 | freeze **14**, retire **4**; the guard's census measured 30 hits, not 22; `family_cousins` stays LIVE (`make atlas` runs it) with its dead branch dropped | the plan's counts came from a partial census |
|
||||
| `progress.py` | "the +219 correction" | +459 (SC03_015's 219 + SC03_118's 3 + the never-extended members of five groups) and REAL −10,211 → EMPTY (the fold); both dated in `progress.json` `corrections` | the include form made invisible sites visible; the macro-era fold had counted empty shared bodies as REAL |
|
||||
| T9 | Max, fresh session | run in S96 at xHigh (Drew) | owner's call; the checkpoint carried the brief |
|
||||
|
||||
## What we believed, what failed, and what we would do sooner (for the retrospective; detail in `docs/decision-log.md` P35)
|
||||
|
||||
- **Believed (2026-09-02):** sotn writes duplicate functions explicitly, so ~12,000 duplicate copies are the normal end state. **False:**
|
||||
read from its tree, sotn shares stage code once and instantiates it by an include. **Sooner:** read the target project's tree, never a
|
||||
memory of it, before setting a policy on its precedent.
|
||||
- **Believed:** refreshing the checkpoint at every task close was enough. **S94 refreshed it five times and still died with 35 minutes
|
||||
of state only in its transcript** — two bank commits inside a task carried no log line, and a background wait consumed the rest.
|
||||
**Sooner:** every commit that advances a task carries its log line and the headline; past ~85 % context, checkpoint before any long wait.
|
||||
- **Believed:** a green per-binary gate meant the tool's edits were what it reported. **Its restore was `git checkout`,** which wiped the
|
||||
previous batch's uncommitted shares and returned green; its cause lines read "Error 33" for 254 of 303 rejections. **Sooner:** a tool's
|
||||
restore path is part of the tool (R57); read a compiler's diagnostics by their form.
|
||||
- **Believed:** the byte join sees every duplicate. **The text oracle found 38 functions in 2,030 copies it could never class.** **Sooner:**
|
||||
build the disagreeing oracle before the batch runs, and measure the disagreement before believing either side (R34).
|
||||
- **Cost:** no model tokens beyond the coordinator and gate 1's three agents; the phase's expense was wall-clock in gates (≈2.5 h of
|
||||
`make check` and fleet runs) and the recovery (≈1 h reading a 6 MB transcript).
|
||||
|
||||
## Commit Message
|
||||
``
|
||||
(Phase 35 landed as 48 commits 48170fd7f..c1a901373, 2026-09-08 (S94–S96). This close = the archived log + this PhaseEnd + the DIGEST
|
||||
update + the kit's record copies, Drew's milestone-close commit.)
|
||||
|
||||
chore(phase-35): CLOSE — Gen3's dedup phase: one source per unique function (3,175 shared headers, 3,173 registry groups incl. the
|
||||
h_text tier, the macro header gone, the five twins from one directory, S1 a tools-health gate with a second oracle, 14 macro-era tools
|
||||
frozen and 4 retired behind a guard, the published counts corrected) — v2.1.0
|
||||
|
||||
- PhaseEnd_Phase35.md written; CURRENT_PHASE.md archived to phase-ends/logs/Phase35.md (R19); DIGEST §0/§2 appended (step 3b)
|
||||
- T0–T8 + T5b across S94–S96 (S94 died at 91 % without a checkpoint; S95 recovered it from the transcript; S96 finished)
|
||||
- close: R22 clean fleet 218/218; tools-health OK (S1 10,180/10,180, 0 violations; dedup-check 3173/0; the guard 0)
|
||||
- rules: R96–R99 ratified at gate 1; candidates (a)–(g) proposed for Phase 36 gate 1
|
||||
- v2.0.0 -> v2.1.0
|
||||
``
|
||||
|
||||
## Rules Added This Phase
|
||||
|
||||
**Ratified at gate 1 (2026-09-08):** **R96–R99** (the PhaseEnd_Phase34 candidates (a)–(d)); full text in `phase-ends/DIGEST.md` §3.
|
||||
|
||||
**Candidates recorded in the log, for Drew to accept, modify or reject at Phase 36 gate 1 (P10):**
|
||||
|
||||
| Rule | Reason |
|
||||
|---|---|
|
||||
| **(a) A shared body has exactly one source; a duplicate copy is a defect the health chain asserts, and its count is published with its rule.** | The phase's invariant (S1), a `tools-health` gate; 1,099 classes / 4,755 copies → 51 / 160, every remainder ledgered with its diagnostic. |
|
||||
| **(b) Every commit that advances a task — an intra-task bank included — carries its log line and the 🛑 headline; a checkpoint older than HEAD is a dead session's checkpoint.** | S94 refreshed the block at every task close, made two T5 bank commits with no log line, then died at 91 % context; S95 rebuilt 35 minutes of state from the transcript. |
|
||||
| **(c) A tool that restores files restores from its own snapshot, never `git checkout` on a tree it did not commit.** | `share_body`'s first bisect wiped the previous batch's uncommitted shares (~50 sites) and returned green. |
|
||||
| **(d) A failure-cause extractor is negative-controlled against the compiler's real message forms, not against the word "error".** | gcc 2.7.2 prints errors without it and exits 33; 254 of 303 rejection lines read "Error 33". |
|
||||
| **(e) Build the disagreeing oracle before the batch runs, and measure its disagreement before believing either side (R34 operated).** | The text oracle's 4,312 hits decomposed into twin false positives, the gate-1 deferral, and 38 genuine byte-variant functions — three different actions. |
|
||||
| **(f) A registry lists a member only after the gate has spoken for it; a listed member whose site does not share is a defect the check names.** | The first pass extended 788 members wholesale; 317 were private after the run (C2d exists now). |
|
||||
| **(g) A policy taken on a remembered precedent is a belief — read the target project's tree.** | The 2026-09-02 deferral rested on one parenthetical about a cross-jump idiom. |
|
||||
|
||||
## PhaseEnd Changelog
|
||||
**v2.0.0 → v2.1.0 — Phase 35 complete (Gen3's first phase).** Every shared function body in the fleet is one plain-C header included at
|
||||
each member's site, sotn's shape, verified from its tree; the 8.8 MB macro header and its 5,147 define lines are gone; the five overlay
|
||||
twins build from one source directory each; the same-address duplicate backlog fell from 1,099 classes and 4,755 private copies to 51
|
||||
ledgered classes and 160 copies, each with the compiler's own reason; 38 functions whose bytes vary per overlay share one text through a
|
||||
new registry tier; the invariant "one source per unique function" is asserted on every health-chain run by a census with a self-test and
|
||||
a second, sig-blind oracle, and by four registry checks; fourteen macro-era tools are frozen and four retired behind a guard that scans
|
||||
every live tool; the published numbers were corrected where the macro-era counting had been blind, dated and explained. Every batch was
|
||||
proven by the 218 hashes from a clean rebuild, which now takes 84–90 seconds instead of 157. R96–R99 ratified; seven candidates proposed.
|
||||
|
||||
## Plain-English Recap
|
||||
This phase made the code say each thing once. Before it, when the same function appeared in many of the game's 141 level files, our source
|
||||
held it as a macro in one enormous header, or as a private copy pasted into each level, thousands of times over. Now every shared function
|
||||
is one small ordinary C file, included wherever it is used, exactly the way the best-known PlayStation decompilation does it, and a check that
|
||||
runs with every health check proves that no function has two sources. Along the way the tool that does the sharing was caught making four
|
||||
kinds of mistake by its own gates and fixed; one session ran out of memory in the middle and the next rebuilt its state from the transcript;
|
||||
a second, independent check found 38 functions the first could never have seen, and they got their own way of being shared. Fifty-one
|
||||
functions could not be shared yet because their surrounding declarations disagree between level files, and they are listed with the
|
||||
compiler's own reason for the next phase to settle. The numbers on the public page were corrected where the old counting had been blind.
|
||||
All 218 game files still rebuild byte for byte, and faster than before.
|
||||
|
||||
## 🛑 Stop Here
|
||||
PhaseEnd written; `CURRENT_PHASE.md` archived → `phase-ends/logs/Phase35.md` (R19, via `git mv`); `phase-ends/DIGEST.md` §0/§2 appended
|
||||
(step 3b); the kit's record corpus regenerated (`make kit-corpus`) — **all left uncommitted for Drew's milestone-close commit** (R6 gate-2
|
||||
sign-off; the message is above). The Phase-35 work is committed across the sessions (`48170fd7f..c1a901373`; nothing pushed). The headless
|
||||
Ghidra MCP was stopped via the sentinel (R23); nothing under `ghidra/` is tracked. **Drew, from the repository root:**
|
||||
|
||||
``
|
||||
git add -A phase-ends decomp-architect docs .run/P35/baseline .run/P35/census
|
||||
git commit -m "chore(phase-35): CLOSE — Gen3's dedup phase: one source per unique function (3,175 shared headers, 3,173 registry groups incl. the h_text tier, the macro header gone, the five twins from one directory, S1 a tools-health gate with a second oracle, 14 macro-era tools frozen and 4 retired behind a guard, the published counts corrected) — v2.1.0"
|
||||
git tag -a v2.1.0 -m "v2.1.0 — Phase 35: one source per unique function; 218 binaries byte-identical"
|
||||
git push origin main --tags
|
||||
``
|
||||
|
||||
Gen3 continues — do **NOT** start Phase 36 here. Start a **fresh session** (effort **Max**, **plan mode**) for **Phase 36 — pins off**,
|
||||
planning from `docs/gen3-handoff.md` and `docs/gen3-standards.md`. Keep this file forever.
|
||||
@@ -75,7 +75,7 @@
|
||||
digest + README sentence. Effort: high (Drew's default).
|
||||
- ☑ **T8** (S96; doc_links strict 0 broken, wiki 32/32, cookbook §453 + index, decision log, accelerators, DIGEST §4, the memory) — the record: wiki (4 pages), how-to ch.10, README:117, the charter rows as dated snapshots, the cookbook section, decision log
|
||||
(R31), accelerators, DIGEST §4, sunset rows, the memory rewritten; doc_links/wiki_render/cookbook_index/kit_coverage green.
|
||||
- ☐ **T9** — close (Tier 1): R22 → 218/218; tools-health OK; the metrics table; the reviewer sequence; PhaseEnd + DIGEST + log archived; v2.1.0.
|
||||
- ☑ **T9** (S96, xHigh by Drew's choice) — close (Tier 1): R22 → 218/218; tools-health OK; the metrics table; the reviewer sequence; PhaseEnd + DIGEST + log archived; v2.1.0.
|
||||
- Rules check (P6): after T3 and after T7.
|
||||
|
||||
## Decisions (owner's words, in order)
|
||||
@@ -583,6 +583,19 @@
|
||||
index regenerated (`1170 sections, 14 symptom buckets`) and `cookbook-index OK` · `kit_coverage: OK` · kit corpus + `tool_census
|
||||
--check: OK`. **T8 ☑.** (`tools/sunset/README.md` rows were written at T6.)
|
||||
|
||||
- **S96 — T9, the close (run in S96 at xHigh by Drew's choice; the plan said Max in a fresh session).** R22 from clean:
|
||||
`extract-all: 217 extracted, 0 failed of 217 (+ main, serial)` · **`check-all: 218 passed, 0 failed of 218`** · `wall=87.62 s` · exit 0
|
||||
(`.run/P35/baseline/r22_t9.log`). `make tools-health` run 1 (467 s) was green through S1 / dedup-check / the guard and red only at
|
||||
`tool_census --check` — the kit's record copies of `DIGEST.md` and the new `PhaseEnd_Phase35.md` (both written while the chain ran);
|
||||
`make kit-corpus` regenerated them and the chain was re-run (its line is appended below). The headless Ghidra MCP stopped via the
|
||||
sentinel: `Save succeeded for processed file: /SLUS_007.26` (R23; unused all phase). `PhaseEnd_Phase35.md` written (the metrics table,
|
||||
the milestone items with literal output, the reviewer sequence, seven rule candidates, the recap); DIGEST §0 refreshed + §2's P35
|
||||
synopsis appended (step 3b). **The chain, run 2:** `tools-health: OK — sigs fresh; corpus(+resident) + cdecl + binaries +
|
||||
report(lint+dedup) + cookbook-index all green.` (`.run/P35/baseline/tools_health_t9.log`, 431 s, exit 0, 0 gaps) with `S1 … 10,180
|
||||
satisfied … 0 VIOLATION(S) — OK`, `dedup-check: 3173 validated, 0 failed`, `macro-form guard: 0 LIVE …`. The log archived to
|
||||
`phase-ends/logs/Phase35.md` (R19); `make kit-corpus` + `tool_census --check` after the PhaseEnd's last edit; everything left
|
||||
uncommitted for Drew's close commit (R6). **T9 ☑ — Phase 35 milestone met; v2.1.0.**
|
||||
|
||||
## Approved plan (verbatim, gate 1 — 2026-09-08)
|
||||
|
||||
# Phase 35 — Gen3 opens: the dedup phase, "one source per unique function" (v2.0.0 → v2.1.0)
|
||||
@@ -882,7 +895,7 @@ snapshot" (share_body's bisect wiped the previous batch's uncommitted shares; R4
|
||||
negative-controlled against the compiler's real message forms, not against the word error" (gcc 2.7.2 prints errors without it;
|
||||
254 of 303 rejection lines read `Error 33`).
|
||||
|
||||
## 🛑 SESSION CHECKPOINT — S96 close (2026-09-08, written at the 90 % context guard): Phase 35 T0–T8 ☑ (+ T5b); NEXT = T9, THE CLOSE (Tier 1, Max) in a FRESH session
|
||||
## 🛑 SESSION CHECKPOINT — S96 close (2026-09-08): Phase 35 COMPLETE — T0–T9 ☑ (+ T5b); PhaseEnd_Phase35.md written, this log archived; NEXT = Drew's close commit, then Phase 36 (pins) in a fresh session at Max, plan mode
|
||||
|
||||
### 0. How to use this block
|
||||
A fresh session (S97) reads CLAUDE.md's load order, replays THIS block verbatim, asks Drew for `/effort max` (T9 is Tier 1: the PhaseEnd
|
||||
Reference in New Issue
Block a user