The bare-`main` carve-out (which keeps the link entry's main unmangled)
keyed on `leaf == "main" && imported == 0`. Under the combined path a
dependency's body folds in with imported==1, so only the root's main
stayed bare. Under separate compilation each package is its own unit and
a dependency's body carries a path-mangling module-reset but imported==0
(#57) — so an imported `fn main` matched the carve-out, emitted a bare
`TEXT main`, and collided with the root entry (`w6l: duplicate symbol
main`). The combined path was unaffected, so this only surfaced under sep.
Gate the carve-out with sep_isdep = (wwiout != NULL): the producer emits a
.wwi output only for dependency units, never for the root/link-entry unit
(root stripped, #69), symmetric on both stages. Only the root unit's main
now stays bare; an imported package's main mangles on its import path
(e.g. aa.bb.main). Both stages.
Gate: test/wcc/989_depmain_sep.c (table-driven, dotted + single-component
shapes, both stages; asserts the mangled dep main + a single bare root
main + cs==ww byte-id; combined path stays neutral).
A package-less primary's bare fn (module="") whose leaf collided with an
imported module's same-leaf exported fn was mis-mangled to the imported
qualified name (a user `fn run` emitted as `test.run`), producing a
dead-duplicate symbol the linker silently shadowed -- a #263-class silent
miscompile, gate-blind and symmetric across both stages. cgen now registers
bare-module fns and resolves a bare-ident reference to its own bare leaf:
mod_lookup_for_fn prefers the bare entry when the call carries no module
hint and skips bare entries when it does, so the moduled-caller path stays
byte-identical. The moduled `main` entry carve-out is an orthogonal rule
(the linker entry is force-bared) and is retained. Prereq for the @test
user-`run` coexist (#80). The bare non-fn (let/def/type) sibling is the
same class but hint-less; deferred as #85, noted at the retained skip.
Under M1 mangling, EXPORTED non-fn decls (let/def/type) skipped path-
qualification and emitted a BARE symbol (`types.I64_MAX` -> `I64_MAX`).
Under separate compilation two packages exporting the same data leaf
would then collide at w6l. Masked in-tree only because no two packages
export the same non-fn leaf.
§7-A (USER-locked, harec's model): path-qualify EVERY exported decl
(fn AND data) at the single mangle choke-point — mod_collect /
collectmods. Retire the `!isfn && d->export` (cstage) and `exported==0`
(wwstage) skips: every decl with a module now mangles `<mod>.<name>`.
The ONLY bare symbols left are @symbol FFI overrides (ffi_resolve at
emit) and the ROOT unit's `main` — both already carved out before the
map insert.
With exported decls in the map the exact-(name,hint)-or-bare value
dance is dead — its sole purpose was the bare-exported case. Delete
mod_lookup_value / mod_mangle_value / mahint (cstage) and
modlookupvalue / emitsymnamehint (wwstage); the value-global sites now
route through the same hint-aware-with-fallback lookup as fns
(mod_mangle_fn/mafn, emitfnname). Net negative LOC in the mangler.
Transparent rename on the live combined path: ref and def move in
lockstep, so cs==ww byte-id holds and the self-host still builds + runs
(fixed-point/995). Byte-id REBASELINE — all 5 ww binaries shift. The
w6c/wwdump combined.ww embed wcc cgen and are regenerated.
Strlit labels were emitted as `_S_<n>` from a global counter with no
per-unit prefix (cgen.c intern_strlit + wwstage internstrlit twin).
Under separate compilation two str-bearing packages both emit `_S_0`..
-> w6l link collision.
Prefix the label with the owning package PATH (`<module>._S_<n>`,
matching mklabel's spelling). The prefix is c->cur_mod, set per-fn by
cgfn and now per-decl by let_pre_intern (save/restore so the later
emit passes, which read cur_mod for fn-ptr relocs, are unaffected).
Pure function of the module path — NOT a build-nonce — so the
self-host fixed-point holds across ww2/ww3/ww4. Both stages, symmetric.
Transparent rename on the live combined path: the label is interned
once and shared by every reference, so ref and def move in lockstep.
cs==ww byte-id holds; the w6c/wwdump combined.ww embed wcc/cgen.ww and
are regenerated.
746_strdef_inline: the strdef-inline sentinel pinned the bare
`LEAQ\t_S_` shape; update to the module-prefixed form (alpha._S_ for
the in-module def, beta._S_ for the use-site-interned cross-module
inline).
989_m3sep_run: add the #49 LINK leg. The str sub-fixture (sleaf+smid)
was keystone-only — never linked — precisely because the global
counter made both emit `_S_0`. With the prefix, compile both `-c`
separately, link (w6l) + run (sroot reads a distinguishing byte through
each string's .ptr, so a collided label would corrupt the exit), both
stages + cs==ww final exe.
New `w6c -c` (both stages): separate-compile / primary-only codegen.
Emit code+DATA ONLY for a package's own (imported==0) decls; treat every
`.wwi`-sourced (imported==1) dep decl as an external. Pure addition behind
the flag — combined.ww stays the LIVE path, `-c` is off on every existing
invocation, so the 990-997 byte-id gates + all prior tests are unperturbed.
The keystone (rob): a `.wwi` is body-less/init-less prototype source, and
cgen already skips body-less fns as externs, so dep fns/types/defs emit
NOTHING for free. The single genuinely-new guard is an imported value-
global (`export let`): its DATAW would DUPLICATE the dep's own definition
(link collision), so it is skipped. The `imported==0` gate is applied at
all top-level emit sites for uniformity (close-by-construction): the fn
loop, emit_lets/emitletdataw, emit_defs/emitdefconstants, and
let_pre_intern/letpreintern — that last one because an imported dep's body
initializer interns strlits while its rhs-stripped `.wwi` does not, which
would shift the _S_ sequence; gating it keeps the strlit table a pure
function of the package's own decls. EXACTLY symmetric with M2's producer
`imported==0` filter — same predicate both directions.
Driver `--sep` build_one_sep + per-package archives + multi-.a link +
cache + BROAD real-target dual-path soak are M3-tail (#46, rob ruling B):
M3-core ships the codegen spine + a self-contained gate that proves all
codegen correctness without a production driver.
Gate 989_m3sep_run: a synth leaf->mid->root fixture carrying all four
cross-boundary fact-classes (fn signature, struct LAYOUT, `def` const
VALUE, `export let` value-global). Per package, holding `-c` constant:
`w6c -c` of (deps-as-bodies) == (deps-as-.wwi) byte-for-byte (the .wwi
conveys exactly the dep facts P's codegen needs); cs==ww at the .s AND
final-exe level (rule 10); sep-path determinism; the value-global guard
(imported origin_tag never re-emits DATAW); and behavioral identity (the
linked program's exit code is the real cross-boundary computation). COLD:
.wwi materialized fresh every run (no warm cache).
combined.ww regen'd for wwdump + w6c (both embed cgen.ww); diff is exactly
the four guards + the flag wiring, nothing spurious.
use_hint/usehint were unit-global first-leaf-match: two directory-
packages exporting the same fn leaf, each imported by a different module
aliasing the same bareword, mis-routed every qualified call to whichever
use was collected first. Identically wrong on both stages (byte-id-green
#263-class). Key the hint on (owner-module, alias) and prefer cur_mod,
mirroring the checker's use_path curmod-preference (55f54fb).
989_m1usehint_run: two same-leaf pick() across a.math/b.math, each
module's call routes to its own import (111/222) + cs.s==ww.s.
Switch symbol mangling from the import leaf clause to the full dotted import path for directory packages; single-file imports keep package-clause mangling (isdir-gate: imported<=>directory-import). The root build unit's fn main stays bare, every other top-level decl mangles, closing #31's duplicate-main hazard by construction (#32). Both stages, byte-identical.
Single commit, not split: the bare rename (f244af3) is red on its own because it unmasks cross-module resolution gaps that do not reproduce pre-M1, so the fixes are intrinsic to making the rename correct. Included: wwstage fnret/fnparamslookupmod map import alias->path (#199b cross-module union-variant scrutinee resolved the wrong fn's union); cstage use_path prefers the referencing module's import for an ambiguous leaf alias (sha256 crypto.math vs strconv math). Tests table-driven: 989_m1mangle_run/_sym, 989_m1union_run (gate-visible per-arm exit codes + cs==ww byte-id).
A tagged-union value nested in module-level array/struct static-init
mis-emitted in both stages: the lit-bytes emitters had no TY_TAGGED
arm, so a tagged element/field fell to the int path and the payload
landed in the TAG word -- match then read the wrong variant. The
zero-placeholder idiom (today the only way to declare a tagged global:
zero-init in static, write at runtime) was correct only by accident
(int-variant zero folds to (0,0), which equals the right (tag0,0)).
Extract a raw-byte core emittaggedbytes/emit_tagged_bytes -- variant
tag@+0, int payload@+8, zero-pad to the slot size; no directive, no
offset, no reloc -- and refactor the scalar tagged emitter to delegate
to it (byte-id-neutral). Add a TY_TAGGED member branch to the array
and struct lit-bytes emitters (both stages) that calls the core at the
existing full-slot stride, before the int fallthrough. Zero stays
(0,0) byte-identical; a non-zero element/field now emits (tag,payload)
correctly.
A wide (str/slice) or struct/>8B payload nested in an aggregate needs
reloc-at-member-offset machinery the aggregate byte-emitters don't
have, so it is loud-rejected (rule 7), deferred to #30; the existing
slice-of-tagged static-init reject is unchanged.
Regenerates the w6c and wwdump combined.ww. Table-driven 843 test:
non-zero array/struct (pre-fix returned the wrong variant), the
non-tag-0 bool-variant edge, byte-id-neutral zero-placeholder rows,
and wide-payload reject rows; each run row also pins cs-vs-ww asm.
A match whose scrutinee is a tagged field of a GLOBAL value-struct read
the tag/payload from the BP region (saved-BP + return-addr) instead of
g(SB) and returned garbage. Both stages were identical-wrong, so the
byte-id gate could not see it -- a gate-blind regression introduced by
M1 (#25): M1's in-place N_DOT match arm uses localfind(base), which
returns the 0 not-found sentinel for a global base, so 0+field.offset
landed in the frame.
Gate the in-place arm on a confirmed-local base -- `localfind(base)==0
&& let_islet/isletvar(base)`, verbatim from cstage's own global test at
cgen.c:2000 (both stages, same spelling). A global base now falls
through to the existing spill path, which cgexprs the scrutinee and
resolves g(SB). M1's local-field in-place ($32) path is untouched.
Regenerates the w6c and wwdump combined.ww. Table-driven 841 test
(global int/reassign/str-payload + a local-field M1 regression row),
runtime-discriminating: pre-fix returns garbage, post-fix 42 on both
stages; rob's direct-global-field spill caveat confirmed at runtime.
Compound `OP=` through an index (gs[i]/a[i]) or a bare ident (g) on a
tagged union silently misbehaved: cstage dropped the index compound and
plain-stored, and BOTH stages compiled an ident compound into an add on
the tag word -- byte-identical, so the gate stayed green while the tag
was corrupted. A compound op on a whole union is nonsense.
Gate the index plain-store arm on TK_ASSIGN so a compound falls to the
existing #133 reject (wwstage's byte-id twin); add a dedicated #21 ident
reject in both stages. This closes the compound half of the tagged-payload
write class (deref #18, dot #34 already reject).
#19 (global tagged-array static-init DATA) is a separate emitter, still open.
The indexed tagged-element assign arm computed its base without the
isglobal -> LEAQ name(SB) branch the scalar element arm already has, so
`gs[i] = v` on a global tagged array stored to a junk frame base and was
lost -- cstage rc=0 where wwstage (which has the branch) rc=42. Mirror
the scalar arm's base resolution; cstage aligns up to wwstage. Local
tagged arrays and scalar globals are unchanged.
The global tagged-array static initializer still mis-packs its DATA in
both stages -- a separate emitter path, filed as #19.
The N_UN/TK_STAR plain-deref assign arm fell to a single fldstoreop for
every pointee, so `*p = v` with p:*tagged wrote the rhs into the tag word
and never the payload -- identically in both stages, leaving the byte-id
gate green while the store corrupted the tag (#263-class, gate-blind).
Gate on TY_TAGGED and route through cg_widen_tagged_store into a scratch
slot, then word-copy to the destination -- the proven runtime-index arm.
Scalar pointees keep the single-store path unchanged.
A module-level nullable `(*T | void)` GLOBAL has no storage path in
either stage: let_emit_size / letemitsize returned 0 for the nullable
TY_TAGGED, so let_collect skipped registration and emit_lets skipped
DATA. The three READ paths then miscompiled SILENTLY and identically-
wrong (a #263-class both-wrong gap, not a wwstage align-up): match read
0(BP) = saved BP via the let_islet-gated #87 arm falling to localfind;
`g is *T` / `g as *T` emitted MOVQ name(SB) for a symbol with no DATA →
w6l undefined-reference. cstage's #87 match arm was itself `!is_nullable`-
gated, so both stages were wrong.
This is the silent→loud bridge: die loud at the size/storage layer the
instant a nullable global is declared, so all three read paths hit one
diagnostic instead of a silent miscompile. A silent gap here is exactly
what "stable before CSP" forbids — CSP's process/handle/chan singletons
(`let c: *Chan | void`) are THE canonical nullable-global consumer. The
full storage + read-class arc (real DATA, nil/void/address-of init, let-
registration, the three SB-resolution read arms) is deferred to task #15
(CSP-prereq); the `&`-init sub-problem additionally couples to the #48
static address-of relocation gap (which already bites a plain `*T` global
init the same way).
Diagnostic core text is identical both stages ("nullable-global storage
unimplemented (task #15)"); cstage's fatal() adds the harness-wide "ww: "
err.c prefix err.ww does not, the same per-stage asymmetry every existing
both-stage reject carries. Byte-id-neutral: the corpus declares zero
nullable globals (grep-verified), so the loud path is unreached in self-
compile and the emitted asm is zero-move; the embedded w6c/wwdump
combined.ww amalgamations are regenerated for the cgen.ww source change.
New 989_nullableglobal_reject: 6 reject rows (match/is/as on a &gv init,
plus nil-init and void-init match, plus an inline non-aliased nullable
form) prove rc!=0 + the shared diagnostic on both stages, init- and
form-invariant; 2 controls (non-nullable tagged global, plain nil-init
*T global) prove the reject is keyed on the nullable TY_TAGGED and the
#87 storage path is untouched.
A `x.f = o` copy of a whole struct field emits a MOVQ run for the
8-byte chunks plus a tail. Both stages inlined a tail that handled only
{4,1}: a 4-byte remainder went MOVL, a 1-byte MOVB, but {2,3,5,6,7} fell
through to an 8-byte MOVQ that OVER-READS the source and OVER-WRITES the
field's natural-offset successor. With #44 packing a successor at its
natural offset, that is a live clobber: outer2{i:inner2{u8,u8}, mark:i32}
copies i with `MOVQ -8(BP),AX; MOVQ AX,-16(BP)` and wipes mark@-12; the
correct move is a single MOVW. Same defect in cstage (cgen.c) and the
four wwstage field-copy sites (cgenexpr.ww: via-ptr, direct-BP-local,
global, and the multi-hop dot-chain CX variant).
Fix: replace each inline {4,1} tail with the descending greedy 4/2/1
(MOVL/MOVW/MOVB) the canonical aggregate-copy emitters already use, so
the tail is complete on every natural size. This is path (alpha) of the
#73 brief — a corpus-neutral, no-workaround completion of the inline
tail. Routing field copies through the shared aggcopy/cg_aggcopy choke-
point (beta) is the balloon: those emitters hardcode (SI)->(BX) at offset
k with zero base displacement, but the four field-copy dsts are
heterogeneous (foff(BX), boff+foff(BP) with no base reg, totaloff(CX)),
so routing forces per-site-per-stage LEAQ src->SI + LEAQ dst->BX rewrites
with no mechanical cross-stage mirror at the CX site = a gate-blind
cs!=ww risk. The emitter extraction is filed as a later addressing-
unification arc (#12). The ragged tail is corpus-absent (every corpus
field copy is tail in {0,4}, where greedy 4/2/1 emits exactly what the
old {4,1} tail did), so this is CLASS-N: zero corpus move on both stages,
byte-id holds by construction.
The cstage <=24 N_CALL receive site (cgen.c:5234) is a different copy
family (sret result read from AX/DX/CX, not a mem-to-mem field copy) and
already handles 4/2/1; left untouched. The str/slice/tagged/tuple 4/1
sites (#76) are likewise a separate family, filed not folded.
989_structcopytail_run pins it on both driver twins: tail2 (MOVW), tail6
(MOVL+MOVW), tail7 (the full MOVL+MOVW+MOVB ladder, the MOVB-path row),
plus an 8-aligned ctl8 (tail-0 control). Pre-fix cstage clobbers mark and
exits non-zero -> cs!=ww; post-fix 4/4 ok cs==ww.
wwstage's unguarded loop-label push wrote out of bounds at depth 17
(compiler-heap corruption); cstage guarded but emitted a wrong break
target. Loud cap error at the limit, both stages, agreeing wording.
wwstage capped defers at 16 and SILENTLY DROPPED the 17th; cstage
capped at 32. Align the cap at 32 and make exceeding it a loud
compile error in BOTH stages — the silent 16-vs-32 split was the bug
(a defer that never runs is a leaked resource). Both stages move in
one commit: one cap contract.
The per-binding copy loop moved ONE word of a 24B str/slice binding —
.len and .cap read zero/garbage in BOTH stages (byte-identical, the
deepest both-wrong-identical of the drain: the F7-era stride fix
asserted convergence without re-measuring the absolute). Copy the full
extent for an sz>8 str/slice binding; the rewritten 989_tupfieldsize
pins all three header words with sliced caps so cap!=len has teeth.
The tagged-binding arm remains open as task #53 (wwstage
paramfieldsize). Review-era task #40, recategorized #263 fused.
Both stages move in one commit: one emission contract; splitting
would leave the byte-id gates red between the halves.
cstage dereferenced 0(BP) for the base — a SEGV on every store
through a module-global struct pointer; wwstage was already correct
(the inverse-set member). Mirror ww's global-pointer load. New
989_globptrfield_run pins cs==ww both stages. Task #47.
cstage's N_TYPEASSERT assumed cgexpr filled the registers and spilled
an uninitialized payload (cs=0 for any stored value); mirror the
landed wwstage emission (tag/payload/cap from g(SB)). Flips the
residual row to cs==ww==correct. Task #46.
cstage spilled frame garbage as the box; mirror the landed wwstage
emission (copy from gi(SB)). The F8-era divergence-only rows gain
pinned cs==ww values. Task #44.
cstage zeroed the return scratch; mirror the landed wwstage emission
(copy from the g(SB) base). Flips the F8-era residual row to
cs==ww==correct. Task #42.
cstage silently dropped the store on reassigning a module-global
tagged union; mirror the landed wwstage emission (tag+payload to
g(SB) via the widener). Closes the cs half of the F8 #263 pair; the
repro row flips to cs==ww==correct. Task #41.
Widening a runtime f64 into a tagged slot pushed a stale AX as the
payload while the value sat in X0 — both stages shared the push bug
(float literals dodged it because TK_FLOAT loads AX too); the
divergent pop sides then produced different garbage. Spill the
payload from X0 (MOVSD) with the variant tag. Review item #49.
Both stages move in one commit: one emission contract; splitting
would leave the byte-id gates red between the halves.
A str field reached through a chained dot (o.i.s) emitted two loads
(ptr, len) and stored a stale CX as the cap — both stages, at any
non-zero chain depth. Emit the full header at the chained-dot leaf.
Review item #29.
Both stages move in one commit: one emission contract; splitting
would leave the byte-id gates red between the halves.
Reading a slice-typed tuple element (t.0) loaded only the pointer
word; len and cap took whatever was left in BX/CX — silent garbage in
BOTH stages once anything clobbered the registers between build and
read. Load all three header words at the tuple-element arm. Review
item #28.
Both stages move in one commit: one emission contract; splitting
would leave the byte-id gates red between the halves.
s.f *= v silently became s.f = v (and the other non-+=/-= ops dropped
likewise) in BOTH stages across five lvalue sub-arms: via-ptr field,
direct local field, str/slice pseudo-field, and the two global-field
forms. Funnel all five through a shared combine dispatch
(cgdotfieldcombine / cg_dotfield_combine) emitting the load-OP-store
sequence at field width, hard-erroring the unhandled kinds — close-by-
construction so no arm stays on the old PLUSEQ-only path (the #133
BUS-routing lesson; #227 sites A/B are the closed siblings). The
refactor routes the corpus's existing +=/-= sites through the same
helper output-identically (byte-id held). Review item #34.
Both stages move in one commit: one emission contract; splitting the
halves would leave the byte-id gates red in between.
arr[i].field /= %= <<= >>= silently dropped the op (load-combine-store
emitted plain assignment) in BOTH stages — gate-blind, the #133 class.
Route every compound op through the combine dispatch at the indexed-
field arm and hard-error the unhandled operand kinds (float/str/slice/
tagged), per the #133 template (3986818). The runtime-correct target is
the op's own algebra (a OP= b == a = a OP b). Review item #33.
Both stages move in one commit: the fix is a single emission contract —
splitting cstage cgen.c from selfhost cgenexpr.ww would leave the
byte-id gates red between the halves.
A bare 'let x: T;' with 1 <= size(T) <= 7 matched no zero-fill arm in
either stage (8B and >8B were already zeroed) - 'let c: [3]u8;' read
stack garbage. User-ruled zero-value semantics: cstage gate sz>8 ->
sz>0; wwstage zsz==8 arm hoisted above the fill-run arm (required -
8B would otherwise route into the run and diverge) and run gate
zsz>0. New 840 pin: dirty-frame probe rows, dual-dim (run + cs/ww
byte-id); discriminators fail exit-154 on pre-fix binaries.
Fused with the lib/bytes test conversion (rule 11): either half alone
turns 967 red. The old exit(signalled+10) wrapped a real 1782-count
ltrim failure to exit 0 - green depended on the garbage. Converted to
assert form (completes the 35/35 @test conversion); ltrim rows keep
the bare 'let c: [3]u8;' as the consumer proof of the fix.
Passing a module-global struct by value -- let g: pt = pt{...}; take(g)
-- was silently miscompiled, mirror-opposite on the two stages. cstage's
by-value struct-arg arm hit localfind(g)->0 and read 2 words from the
frame (MOVQ (BP)), never main.g(SB) -> returned garbage. wwstage used the
correct main.g(SB) base but fell through to the scalar single-PUSHQ
default, pushing one word for a 2-word struct -> dropped a field.
Both stages now take the off==0 global branch: LEAQ main.NAME(SB) and copy
all struct-size/8 eightbytes (reusing the GAP-A.ptr/#231 global-base
predicate), converging to one byte-identical sequence. The local path
(off!=0) is unchanged; >16B aggregates (#271) already resolved globals.
Commit A of the cluster; the cstage-only inferred-global-type Sym-repoint
(every let g = ... module-global yields <nil> downstream) is Commit B
(#18). Slice/str global-by-value args have the same wwstage field-drop --
filed (#10 G-valglobal-arg; struct closed here). byte-id 990-997 8/8.
test/wcc/822 table-driven, byte-id per stage.
def C:[N]str; C[i] was loud (undefined main.C) both stages. Three folded
fixes, one commit (splitting would ship a bisect point where wwstage
silently returns an element address instead of .len):
P0: the str-array static-init emitter dropped its vestigial directive
=="DATAW" gate so a def table rides the same DATAW-header + DATAR-reloc
path as let. A def str/slice table lives in DATAW by w6a's A_DATAR-holder
constraint -- placement only; def immutability stays checker-enforced.
P1: let_pre_intern / letpreintern walked N_LET only, so a def str-array's
element string-literals were never interned (dangling _S_n). Extracted a
pre_intern_strarray SSoT helper, called for a def str-array arm too, both
stages. Scoped to str fixed arrays; def []T / def [N][]T stay loud (#270).
P2: wwstage cgenexpr lacked a defvartnode fallback in the indexed-element
classify, so a def str-array element load returned the element address
instead of the slice header -- a silent miscompile. One line, aligning
wwstage up to cstage (which was correct). C[1].len now = 3 both stages,
byte-identical.
byte-id 990-997 8/8; w6c/w6c_ww move. test/wcc/819 table-driven. The
def-global scalar str index sibling (def S:str; S[0]) stays task #14.
A global fixed array's .ptr (= &A[0]) must take the SB base, but cstage
emitted frame-relative LEAQ off(BP) for BOTH let- and def-global arrays
-> *A.ptr read frame garbage (0 instead of the element). cstage-SILENT;
wwstage def-global was a loud link-error. The .ptr read arm now gates
off==0 && (let_islet || def_isarraydef) -> LEAQ name(SB), reusing the
def-array index base predicate (cgen.c:4367, the #94/#231/#48 class).
Locals (off != 0) stay BP-relative -- the 14 toolchain backing-ptr sites
unaffected.
wwstage let-global was already correct; this adds the missing def-global
arm (cgenexpr.ww), converging cstage/wwstage byte-identical across all
three flavors (local / let-global / def-global) and closing a latent
cstage-only let-global cs!=ww divergence.
Byte-id 990-997 8/8 (corpus has no global .ptr); w6c/w6c_ww binaries move
(cgen changed). test/wcc/818 table-driven, build+run+byte-id per flavor.
The str==/!= arm of cbinop had an N_IDENT fast-path that assumed the operand
was a local: localfind returns 0 for a module-global str, so it loaded
(BP)/8(BP) — saved-BP/retaddr garbage — into rt_streq. `p == sepstr` silently
compared garbage (returned wrong). Mirror #148's global branch at both sub-sites
(rhs/lhs): off==0 && let_islet -> LEAQ name(SB) base, load ptr/len. Distinct
per-site fast-path, not a shared choke (the by-value-global-arg family
#148/#150/#151 closes separately). cstage-only; the wwstage str== twin is #146
(-> #125 batch).
Pin test/wcc/989_strglobeq (table-driven: const+let globals, rhs+lhs ident,
==/!=, unequal + len>1 rows; teeth-proven). Surfaced by the lib/path c3
buffer-ops gate-1 oracle.
A let's own name was visible during its OWN initializer: cgen prepended the
new local into the name-keyed localfind chain BEFORE emitting the init, so
`let x = f(x)` read the fresh UNINIT slot, not the outer/param x. Both-wrong-
identical silent miscompile (gate-blind byte-id). Surfaced by path
dirname/basename (was the c3-posix path->p rename).
Align to Hare (harec check.c:1439 evals the init, then scope_insert). Fix,
both stages, IDENTICAL asm: reserve the frame slot BEFORE the init emits,
link the binding's name into the localfind chain only AFTER.
- cstage cgen.c: split localoff -> localslot(reserve)+link; N_LET's 12
case-level breaks -> goto letlink (tail links once); the inner-for break
is preserved; the 4 fatal() arms untouched.
- wwstage cgen.ww/cgenstmt.ww: new localreserve (= localalloc minus the
chain-link); cglet -> cgletbody(c,n,off) + a cglet wrapper that
reserves -> calls body -> links after.
Byte-id-safe on existing code: localfind is by-name, so deferring the link
is a no-op on every non-self-shadow let (grep = 0 self-shadow sites) — 990-997
stay green. Because both stages emit identical now-correct asm, byte-id
CANNOT catch this; the pin is a RUNTIME test, teeth-proven (revert -> pin
fails). test/wcc/989_letshadow{.ww,_run.c}: param-shadow, let-in-init shadow,
rename control, arrlit self-ref.
Embedded regen: selfhost/cmd/{w6c,wwdump}/main.combined.ww. Gate: all 325
passed, byte-id 990-997 green, w6c c587f4a1 / w6c_ww 7a69f898 (deterministic).
The slice-IDENT call-arg fast path pushed the header words off off(BP)
where off=localfind(name); for a module-global slice localfind→0, so it
read saved-BP/RIP/caller garbage instead of name(SB). Add the global
branch (LEAQ name(SB) base, push 16/8/0 off it) mirroring the sibling
N_SLICE arm; local path unchanged. cstage-only: wwstage checker-rejects
the shape (#120), so byte-id-safe and the twin defers to #125. Unblocks
path c2-stack (dot/dotdot are faithful module-global []u8). Sibling
structarg fast-path filed #150.
Probe-first find for the path c2 appendlit (buf.buf[lo..hi]=bs): a
slice-copy-assign into a struct-field array sub-range emitted ZERO code —
silent NO-OP, both stages, both-wrong-identical (#263), so runtime is the
only net. N_ASSIGN gains an N_SLICE-LHS arm (cgen.c + cgenexpr.ww
slicebaseesz twin) reusing the N_SLICE-read base/esz cascade and copying
(hi-lo)*esz bytes from rhs.ptr via a runtime loop (len is runtime; no
REP/MOVSB). esz routed through the type table (rule 13; [N]u8->1). Hare
len(bs)==hi-lo assert deferred to #149.
A void (size-0) error-singleton type-name used as a VALUE (return / let-init /
assign / call-arg) all share the N_IDENT non-local global-value load; the load
emitted MOVQ main.<singleton>(SB),AX for a payload symbol that never exists →
w6l undefined reference. Guard TY_VOID && !let && !def at the non-local
fallthrough so nothing is emitted; the enclosing widen arm stamps the variant
tag. wwstage was already tag-only correct — this aligns cstage up to it.
Pin test/wcc/949_void_error_singleton_run.c (6 rows, teeth = link-fail
pre-fix). cgen-first blocker for the path::buffer arc (error.ha is all !void).
append/insert of a struct-LITERAL value evaluated the literal's field
exprs AFTER the grow, so a field reading the destination (e.g. len(xs))
saw the grown length. Both stages, #263 gate-blind (cs==ww byte-identical,
both wrong — runtime is the only net). #50 fixed the scalar/boxing value
arm; the struct-lit arm still post-grew.
Fix (mirror #50, both stages): resolve the struct, fill the literal into a
fresh per-site scratch (@appendstructscr, sized esz, survives rt_ensure +
nested-append clobber) BEFORE the grow, then copy scratch -> post-grow slot.
The copy uses the precise descending 8/4/2/1 ladder (the proven N_IDENT
struct arm directly below), NOT a raw 8B-word block copy: a struct's size
rounds to maxalign (check.c:916), so a sub-8B struct packs at a 4/2/1B
slice stride and an 8B copy over-writes past the slot — at a power-of-2
capacity boundary that clobbers the adjacent allocation (heap corruption,
both stages). The ladder never reads past esz (no uninit high bytes) nor
writes past the slot; esz=8 stays a single MOVQ (byte-id preserved).
insert() rides by construction: both stages desugar it to append and
re-dispatch into this arm. The #49 aplace path already uses the precise
ladder (verified, not exposed). #59 closes the last composite-value
eval-order hole in append/insert.
Pin: 946_append_structlit_evalorder_run — append / insert / narrow-neighbor
(i32-field at the cap boundary with an adjacent-allocation survival assert)
rows, each base-fail at 39432f7 and post-pass with cs==ww byte-id.
A tuple LITERAL with a declared-tagged element reached the cursor-fill
helper (cg_tuple_lit_to_cursor) through the generic cgexpr(N_TUPLE) arm
with no declared type, so the element was stored stamped-keyed at its
constructed scalar width rather than widened into the declared tagged box.
Both consumers ran silent and wrong on both stages (#263 gate-blind:
cs==ww byte-identical, both wrong — runtime is the only net).
#64 massign: N_MASSIGN derives a declared tuple type from the lvalue
binding types and threads it into cg_tuple_lit_to_cursor + the receive
loop (mirror of the #57 N_LET wire); a `_` target falls back to the rhs
literal element type for cursor stride.
#68 call-arg: the send is made param-aware (fill over the PARAM tuple) and
the restage guard graduates a declared-tagged element to a real widen
(reusing cg_widen_tagged_store); nested tuple/struct/array elements and
tagged elements with no param decl stay rule-7 loud. The matching
pop/drain is made param-aware too so push count == pop count: a
param-aware send pushes the box's N words, so the drain must pop N or the
SysV arg sequence skews. This is a push/pop balance requirement of the
send change, not a separate latent under-drain (the standalone trailing-
arg drain is already correct at HEAD).
Closed by construction: the only remaining cg_tuple_lit_to_cursor caller
passing NULL/nil is the generic cgexpr(N_TUPLE) arm, provably non-widening
(constructed type == governing type). The four widening consumers — LET,
RETURN, MASSIGN, call-arg — are all decl-wired. Whole-tuple single-ident
reassign from a tuple literal is rule-7 loud (task #49), not a silent
widening consumer, so the residual NULL arm stays non-widening.
Pin: 945_tuple_lit_declblind_run — massign / call-arg / `_`-control /
call-arg-drain / nested-tuple-ERR rows, each base-fail at abd97e6 and
post-pass with cs==ww byte-id.
A cross-module `&module.fn` in a const emitted no static reloc (the
const was never defined -> w6l undefined-reference, both stages) and
wwstage's checker rejected the const fn-table. #117/#119 wired the
&fn->DATAR const-data reloc for SAME-module &fn only; charclass_map
(fold-6) needs cross-module (12x &ascii.isXXX).
cgen: add the N_DOT arm to the &fn->symbol helper (node_fnptr_sym /
nodefnptr + the two ww emit sites), emitting mafn(leaf, module-ident)
-- exactly the symbol a runtime &mod.fn or a direct cross-module call
already emits. The helper is the SSoT for both the scalar (#119) and
tuple-row (#117) const-data paths, so one arm closes both.
checker: type a cross-module `&mod.fn` as `*fn(...)` in the TK_AMP arm
(the N_DOT twin of #206's N_IDENT fn-ptr synthesis, gated on a resolved
SK_FN/N_FNDECL leaf), so isassignable affirmatively accepts the const
table -- aligning wwstage UP to cstage's actual acceptance reason
rather than by abdication. The SK_FN gate keeps a non-fn `&mod.var`
from synthesizing a fn type (the one pre-existing nonfn-scalar cs!=ww
slip is N_IDENT-base, untouched and reproduces same-module).
One consumer-coupled commit (the checker accept gates wwstage cgen, so
neither half is independently testable). Narrow: slice-row + scalar
only; fixed-array (#118) and struct-field (#129) stay separate. Both
stages emit the correct cross-module symbols at the right tuple-slot
offsets -> byte-identical (990-997 green). Pin 949_xmod_fnptr_const_run
(distinct fns so a wrong reloc is caught + the SK_FN-gate axis). This
was the last fold-6 cgen blocker; charclass_map is now unblocked.
Reading or storing a tuple element of an indexed array element was
broken across the board (the fold-6 read-path). One fused commit,
both stages, four faces of indexed tuple-element access:
- FIELD read `tbl[i].N`: was loud ("unsupported field-read shape" --
the field-read dispatch keyed on an N_IDENT base; an INDEX base fell
to a fatal). Now resolves &tbl[i] via the place-spine and reads the
field at addr+foff through the existing per-kind arms (str-triple /
scalar / fn-ptr).
- WHOLE read `let e = tbl[i]`: was a silent word0-only truncation
(plain-tuple kin of #37/#58, which covered only tagged). Now a full
cursor fill from &tbl[i].
- STORE `a[i] = (3,4)` (N_TUPLE-literal rhs): was a silent word0-only
store -- the write face of the read. The aggregate-store-into-index
site handled ident/dot/deref tuple rhs but not the literal; now it
materializes the literal and word-copies. Narrow: N_IDENT base only
(N_DOT/chained stay deferred, #270).
- for-range over a const-slice-of-tuple: was a divergent SEGV; now a
symmetric loud-stop on both stages (filed #122).
The store and read were a round-trip that passed test 809 only by luck
(broken store XOR broken read canceled). Fixing the read alone exposed
the silent store; rule-7 obliges fixing both, so 809 is now genuinely
correct, not luck-correct. Both faces are byte-id-blind (#263) -- the
net is a runtime round-trip pin with distinct-per-word values and a
real call clobbering the cursor registers between store and read, so a
word0-only store or read is caught. Both stages byte-identical
(990-997 green). Pin 947_tuple_index_read_run.
The #117 emit_slice_data TY_TUPLE arm fires for any foldable tuple row,
but wwstage's checker admits only the (str,*fn) shape and loud-rejects
the rest before cgen. Document the retained acceptance divergence at the
site (rule 7/8) with the #120 pointer. Comment-only, asm-neutral.
cg_widen_tagged_store only handled a tuple LITERAL (N_TUPLE / cast-of-
N_TUPLE) widened into a tagged box; any addressable non-literal tuple
source -- IDENT var, INDEX tbl[i], DEREF *p -- hit the `else fatal`
("tuple-typed source shape unwired"). Both stages loud-identical
(honest, no silent miscompile). This blocked indexing a const tuple
table into a union (regex charclass_map[i] -> charset union).
Add an addressable-tuple-source arm, both stages (cgen.c +
cgenutil.ww twin). It resolves the source address via the cgplaceaddr
place-spine (covering ident/index/deref -- one mechanism, so the trio
is family-closed) and block-copies the tuple's type-table ->size bytes
into the box payload (after the 8B tag), then stamps the variant tag.
No re-slotting: a tuple's in-memory layout uses the same eslot strides
(str=24B header, *fn=8B, ...) as the box payload the literal loop
fills, so source-layout == dest-layout. The existing narrow-pack and
tag-unresolved guards stay as the honest boundary; CALL/sret tuple
sources (different receive, #68-kin) stay loud.
align-BOTH: both stages were loud (no runtime reference), and byte-id
is structurally blind to an identical-wrong emission -- so correctness
is proven by a RUNTIME read-back pin (944_nonlit_tuple_widen_run, per
shape: match-extract + assert str header + call the fn-ptr elem with
distinct fns so a stale pointer is caught). 936's old reject row
graduates to a run row. Both stages byte-identical (990-997 green).
A tuple containing a tagged-union element, used as a union member
(e.g. ((void|size),(void|size),size) | error), loud-stopped in the
cgen return-store: the tuple-in-union store walk had scalar/float/
str/slice element arms but no TY_TAGGED-element arm. A PLAIN
tuple-in-union already worked -- the blocker was the tagged element.
Add the recursive two-level widen arm at both stages
(cg_widen_tagged_store / cgwidentaggedstorebp): for each tagged
element, re-enter the tagged-box store (inner tag@slot+0,
payload@slot+8) at the element's tuple-payload offset, then stamp the
outer tuple tag. Slot strides come from the type table
(roundup8(eu->size)) -- the checker already sizes the shape correctly
(tuple->size measured 40, union box 48; check.c:715-720). The
recursion descends a finite type tree (a tagged element is never a
tuple literal, so it can't re-enter the tuple arm); unsupported
deeper nesting still louds via the existing size/tag guards.
Both stages get the same arm -> byte-id (990-997 green; additive,
bootstrap-neutral). cstage runs the full b1c shape
(construct+return+match-extract) as the runtime reference; wwstage's
store rides on byte-id until gap-B. gap-B (wwstage checker
match-acceptance of the tuple-with-tagged case pattern) is a separate
commit -- wwstage still louds the match honestly at the checker.
Pin 944_tuple_tagged_union_run.
Reading or writing a tagged field of an indexed array element
(xs[i].field) was broken on BOTH stages, byte-identically and
silently (#263 gate-blind): the arr[i].field branches had arms for
array/str/slice/float but no TY_TAGGED arm, so the tagged field fell
to the single-word scalar path. READ loaded only the tag word (stale
payload -> `xs[i].min as T` read garbage); ASSIGN stored the raw
unboxed scalar into the tag slot, corrupting the box.
Insert a TY_TAGGED cursor arm before each scalar fallback, both
sites both stages (cgen.c read + assign; cgenexpr.ww cgdot N_INDEX-lhs
read + cgassign indexed-field). READ mirrors cg_tagged_memread
(payload -> DX/CX/R8, tag -> AX last). ASSIGN synthesizes the tag for
the concrete variant (taggedvariantindext) and stores tag+payload via
the str/slice 3-word store spine -- not the source-remap widener
(concrete rhs has no source tag to remap).
>32B / multi-word / float payloads are loud-stopped at all four arms
(emission not yet wired; see #114). That shape is reachable today via
a narrow-variant ctor, so it louds rather than silently miscompiling.
Both stages get the same arm -> byte-id preserved (990-997 green; the
runtime is the net for this #263 class). Pin 944_idx_tagged_field_run
(read/assign runtime rows + >32B expect-loud rows).
Drop the `!TY_ARRAY` exclusion in the bare-let no-rhs zero-fill (cgen.c
N_LET else + cgenstmt.ww cglet, both gated `sz>8 && !TY_ARRAY`) so an
uninit `[N]T` array local zero-fills like every other composite (Go-zero
per user ruling). The zero-fill extent is the array's chased ABI size
(lu->size / chased tinfo.size, rule-13 — never a hardcoded count*esz),
NOT the slot-padded letslotsize, so a non-8-multiple array ([20]u8 = 20)
zeroes its exact bytes instead of over-zeroing to the 24B slot. The
unrolled MOVQ/MOVL/MOVB run mirrors the existing composite path; the
largest real local array ([256]u8) is 32 MOVQs (pathbuf[4096] is a
module GLOBAL, BSS-filled — never on this stack path, so no large-fill
case exists).
Closes a gate-blind #263-class bug: `let a: [3]int;` (no init) read
whatever the stack held — a clean frame masked it (fresh stack = 0), a
dirtied frame exposed it (d_array=165 garbage). BOTH stages emitted no
fill, both-wrong-IDENTICAL, so the cs==ww byte-id net could not see it.
The load-bearing net is therefore a RUNTIME dirtied-stack zero-read
(944_array_zeroinit_run: array-elem / narrow [4]u32 / non-8-mult [20]u8
/ 2D + an initialized control), not asm presence.
Deliberate byte-id EVENT: every uninit-array source site gains zero-fill
insns, so the 990-997 .s MOVE vs the prior tree; cs==ww HOLDS (both add
the identical insns). The 990-997 byte-id + 995 self-rebuild staying
GREEN is the fixpoint proof — it proves every uninit compiler-array is
write-before-read, so the zero-fill is purely additive and the
ww1->ww2->ww3 self-rebuild fixpoint holds by construction. w6c/wwdump
main.combined.ww regenerated (cgenstmt.ww embeds there).
#84 is ARRAY-ONLY; the no-default reject-set (uninit tagged / plain-*T)
is split to #113, parked behind a ruling — selfhost relies on the
current (void|T) zero-fill (the "not-set-yet" idiom).
`&D[i]` over a module-level DEF array SEGV'd on BOTH stages: the
TK_AMP N_INDEX N_IDENT base classify checked only the local and let
legs, so a def-array base fell to a wrong else — cstage zero-based
the addend (XORQ BX,BX -> wild pointer, cgen.c) while wwstage
value-loaded the symbol (MOVQ name(SB) = D[0], not its address,
cgenexpr.ww complex-base fallback). Divergent asm, both wild.
Add one def-array leg per stage, mirroring the working let leg:
- cs: `def_isarraydef(base) -> LEAQ name(SB),BX` alongside let_islet.
- ww: the `defvartnode` fallback the read-side cgindex already takes
(cgenexpr.ww:1762) -> N_TARRAY classifies isglobalarr -> LEAQ
name(SB).
The def DATA symbol already exists (plain &D + D[i]-read work), so
once the base is the address the existing i*esz scale + ADDQ
round-trips. cs and ww now emit BYTE-IDENTICAL LEAQ-SB asm — the
both-broken -> both-correct convergence is the point (#263 class).
Rows (944_def_amp_idx_run, all 0/0 byte-id): amp_int [3]int,
amp_u32 [3]u32 esz=4 (narrow scale), amp_arg &D[2] as a func-arg;
controls ctrl_plain (&D), ctrl_read (D[i]), ctrl_2d (&M[1][1]) keep
working. *p spelled `let v: T = *p` — `*p: T` parses as `*(p: T)`.
OUT (filed #112): &D[..] slicing a def-array is a distinct parse
reject needing a Hare-fidelity ruling — not this leg.
