Files
BFM-decomp/docs
Drew T 69d1d37262 feat(phase-25): T7 F-band wave 3 — func_80164930 cracked + swept ×134 (read-global fix); fleet 73.85->73.89%, R22 136/136
- func_80164930 (81 ins) CRACKED + swept x134 = 134 fns (133/0 siblings, family_sweep --reconcile).
  The crack = the read-global fix: flip the file-scope decl s16->u16 (byte-neutral to the store-only
  caller func_801647A4) + reference the global directly, so the read lowers to direct-addressed `lhu`.
- TWO DURABLE FINDINGS (cookbook §42b):
  (1) THE STALE-OBJECT GATE TRAP: a piped `make build >/dev/null` that FAILS leaves a stale .o, and
      `asm-differ -o` then reports a phantom score-0. This invalidated wave-2's "iso-drift" labels --
      a rigorous re-check (rm .o + build exit-code + real whole-binary SHA) shows all 4 remaining
      iso-drift drafts NOCOMPILE (unreconciled callee externs vs the TU canonical-sig layer). Every
      gate MUST rm the split .o + check the exit code (compounds the §42a --out gotcha).
  (2) canon_sig_reconcile `*(T*)&D_sym` READ-global drift: &sym forces the address into a held register
      (kills direct %hi/%lo -> schedule drift); write-only globals unaffected. Fix = file-scope exact-type
      decl + direct ref (a block-scoped `extern u16` vs ambient s16 is a hard cc1 conflicting-types error).
- Frontier reassessed: the 9 remaining wave-3 targets each need real-TU reconcile-cracking (NOT gating the
  broken wave-2 drafts); each cracks -> ~134 fns (all x134 families). ~1,200 fleet potential.
- R22 clean-fleet 136/136 BYTE-IDENTICAL from a fully clean tree; NON_MATCHING 7 (0 in default build, G4).
2026-07-10 14:58:19 -06:00
..