Hojun-Cho
74195ac4dc
F2a batch-2 c4, per rob's FINAL RULING (i) (banked, .ai/rob-batch2-spec.md):
ww-only — the cs twin (cmd/wcc/type.c:369-385) is already harec-shaped
since F1 9bd0d8b. Port of ref/harec/src/types.c:989-996
type_is_assignable structure: full dealias of the dst when its chased
kind is not tagged (du = tichase(dst); the STR/BOOL/NIL arms read du);
the tagged variant drill keeps the ORIGINAL p.type_ and detects a
variant's taggedness via the full chase (was one-level — a 2-level
alias-tagged variant slipped INTO the recursion); INT/FLOAT/RUNE keep
the recursive predicates on the unpeeled dst, exactly as cs.
STRUCTURAL ALIGNMENT, ZERO ACCEPTANCE GRADUATIONS, GRADUATIONS RIDE
#90: on wwstage no bool/str literal reaches this funnel today — the
checker stamps N_TRUE/N_FALSE CONCRETE bool (check.ww:2440) where cs
stamps ty_untyped_bool (check.c:1234), the bool member of the #14
N_STRLIT stamp cluster (task #90, filed this train; its metadata
carries the graduation obligations: BOTH layers, the stamp AND the
flatvariantidxt pass-2 one-level pu.under guard at :2868/#17, must
clear before the bound rows flip). The live silent-wrong this would
close — `type bb2 = bb1 = bool; let v: (void|bb2) = true; v is bb2`
runs cs 0 / ww 1, single-line asm diff (tag store MOVQ $1 vs $0) —
therefore does NOT graduate here: the row dies one layer up
(pins-follow-the-layer).
test 944 (+7 rows): untyped_hold33_pin ((void|T) untyped-int init,
the #33 tagged-dst drill HOLD) + untyped_tagdst_2lvl_hold (2-level
alias tagged dst) + untyped_str2lvl_ctl (masked twin — the shape
fallback rescues str, documents the rescue) + untyped_int2lvl_ctl +
untyped_bool1lvl_ctl all K_RUN 0/0; untyped_bool2lvl_bound90
K_RUN_CS BOUND citing #90; untyped_nestvar_bound K_BUILDERR_CS BOUND
expected-state row (#199-alpha + #90) — untyped literal into a
NESTED-tagged alias-wrapped variant: cs CHECKER loud-rejects
("not assignable", the #199-alpha ww-stricter no-transitive-drill
rule, type.c:316-324) while ww's node-keyed checker accepts and the
binary runs 0 OBSERVED post-chase; the acceptance gate is check.ww
(batch-4 aligns ww down; the row flips to loud-BOTH then; ww half
pinned via #90 metadata — the harness has no cs-reject+ww-runs
mode). Every row's cs column OBSERVED (built+run) at dispatch-SHA cs
per the ratified condition; probes at /tmp/b2/c4.
989 lib-byteid ratchet: ZERO flips (31 ID / 9 DIVERGE / 3 WWREJECT
hold exactly as pinned). Byte-id: five-mains NEUTRAL on identical
inputs vs the c3 build; cs==ww on the regenerated five; test-unit 288
green. combined.ww regens ride along.