Files
BFM-decomp/tools/t7_bank.py
T
Drew T 5b94a8247b feat(phase-25): T7-M1 — 37 exemplars banked ×1 via canon_sig_reconcile v3.2 (giants incl. func_8016A290/284)
- tools/t7_bank.py: the M1 driver (reconcile-at-bank-time against the CURRENT TU + harvest_verify
  whole-binary gate; chunk-bet + per-round re-reconcile for cross-fn ambient mutation; giants first)
- canon_sig_reconcile v3.2: uniquify ALL draft-defined type names AND struct/union TAGS to <name>_<addr>
  (byte-neutral) -> collision-proof when many exemplars bank into one TU (the 'redefinition of struct Fr'
  class); banked func_8016A290 (284-ins giant) + fixed the inter-draft collisions
- 37/40 non-jumptable M1 exemplars banked BYTE-IDENTICAL (ov_SC01_077 d19c9580, clean rebuild); each is
  currently ×1 (M2 sweeps them ×134 next)
- R14 CORRECTION (cookbook §41b): the T6 object-only probe OVER-counted BANKABLE by 7 -- it is blind to
  rodata + link. (a) 4 jump-table fns (3 _o0 giants + func_8012ACE0): .text byte-perfect but a switch
  jump table in rodata diverges -> REFUTES the T6 'Q3 -O0 REFUTED' claim; F-band jump-table workflow.
  (b) 3 last-referencer link-walls (func_8016D688/D1D8/165240): C-ifying the only asm referencer of a
  scratch data symbol drops splat's auto-symbol -> ld undefined reference; M-linkwall tier, deferred
- ov_SC01_077 REAL 32 stub-exemplars -> 37 more defined; whole binary byte-identical throughout (G3/P9)
2026-07-10 00:02:55 -06:00

110 lines
4.9 KiB
Python

#!/usr/bin/env python3
"""t7_bank.py — Phase-25 T7 tier-M1 driver: reconcile-at-bank-time + whole-binary gate.
The T6 curriculum (docs/phase25-t6-curriculum.md) proved 44 family exemplars bank via
canon_sig_reconcile v3.1. This drives the banking with the curriculum's discipline:
* reconcile EACH still-stubbed target against the CURRENT TU (each bank mutates ambient
truth for the next — so re-reconcile every iteration, clearing csr's caches);
* stage the reconciled drafts, run tools/harvest_verify.py (the whole-binary byte-gate,
G3/P9, the SOLE arbiter — the probe's in-TU masked compare is jal-symbol-blind);
* --good-sha from config/check.<bin>.sha (NEVER hand-typed);
* loop per split until no new banks (handles cross-fn conflicts: a fn that fails because
a sibling banked below it re-reconciles against the updated TU next round);
* giants first within each split.
Usage:
tools/t7_bank.py --tier M1 # all M1 splits
tools/t7_bank.py --tier M1 --split ov_SC01_077.c # one split (pilot)
tools/t7_bank.py --tier M1 --chunk 12
"""
import json, os, re, sys, shutil, subprocess, importlib.util, argparse
from collections import defaultdict
REPO = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
os.chdir(REPO)
spec = importlib.util.spec_from_file_location('csr', 'tools/canon_sig_reconcile.py')
csr = importlib.util.module_from_spec(spec)
spec.loader.exec_module(csr)
BIN = 'ov_SC01_077'
GOOD_SHA = open(f'config/check.{BIN}.sha').read().split()[0]
def subdir_of(split): # ov_SC01_077_a.c -> ov_SC01_077_a
return split[:-2]
def stubbed_in(tu_path, subdir, fns):
tu = open(tu_path).read()
return [f for f in fns
if re.search(r'INCLUDE_ASM\("asm/' + re.escape(BIN) + '/nonmatchings/'
+ re.escape(subdir) + r'",\s*' + f + r'\);', tu)]
def main():
ap = argparse.ArgumentParser()
ap.add_argument('--tier', default='M1')
ap.add_argument('--split', default=None, help='limit to one split .c (pilot)')
ap.add_argument('--chunk', type=int, default=12)
ap.add_argument('--worklist', default='.run/t6_worklist.json')
a = ap.parse_args()
wl = json.load(open(a.worklist))
targets = wl[a.tier]
by_split = defaultdict(list)
for t in targets:
by_split[t['src']].append(t)
banked_all, failed_all = [], []
splits = [a.split] if a.split else list(by_split)
for split in splits:
items = sorted(by_split[split], key=lambda t: -t['nins']) # giants first
subdir = subdir_of(split)
tu_path = f'src/{BIN}/{split}'
stage = f'.run/t7_stage_{subdir}'
print(f'\n########## split {split} ({len(items)} M1 targets) ##########')
rounds = 0
while True:
rounds += 1
stub = stubbed_in(tu_path, subdir, [t['fn'] for t in items])
if not stub:
break
by_fn = {t['fn']: t for t in items}
# reconcile each still-stubbed target against the CURRENT TU
csr._AMBIENT_CACHE.clear(); csr._VISIBLE_CACHE.clear()
shutil.rmtree(stage, ignore_errors=True); os.makedirs(stage)
n_recon = 0
for fn in stub:
try:
out = csr.reconcile(fn, open(by_fn[fn]['draft']).read(), None, tu_path)
open(f'{stage}/{fn}.c', 'w').write(out)
n_recon += 1
except Exception as e:
print(f' ! reconcile {fn}: {e}')
print(f' round {rounds}: {len(stub)} stubbed, {n_recon} reconciled -> gate')
before = set(stubbed_in(tu_path, subdir, [t['fn'] for t in items]))
subprocess.run(['python3', 'tools/harvest_verify.py', '--binary', BIN,
'--src', tu_path, '--asm-subdir',
f'asm/{BIN}/nonmatchings/{subdir}', '--drafts', stage,
'--good-sha', GOOD_SHA, '--chunk', str(a.chunk),
'--out', f'build/{BIN}/{BIN}', # NOT the resident default
'--verified-out', f'.run/t7_verified_{subdir}.txt',
'--failed-out', f'.run/t7_failed_{subdir}.txt'],
check=False)
after = set(stubbed_in(tu_path, subdir, [t['fn'] for t in items]))
newly = before - after
print(f' round {rounds}: banked {len(newly)}')
if not newly: # no progress -> the rest are genuine residue
failed_all += sorted(after)
break
banked_all += sorted(newly)
print('\n==================== T7 %s SUMMARY ====================' % a.tier)
print('banked : %d %s' % (len(banked_all), ' '.join(sorted(banked_all))))
print('residue: %d %s' % (len(failed_all), ' '.join(sorted(failed_all))))
if __name__ == '__main__':
main()