The wcc test drivers ran `ww build <bare-/tmp src>` with no -o, so the
compiler's <stem>.sepwork scratch landed beside the source and was never
cleaned: unbounded /tmp growth (2195 stale dirs observed) that fills tmpfs
and fabricates phantom test failures + silent harness aborts, and for
in-repo fixture builds leaked .sepwork into the tracked tree.
Each leaking build now writes its source + output inside a per-invocation
tmpdir, passes -o <tmpdir>/<stem> so the .sepwork lands inside it, and
rm -rf's the tmpdir on every exit path -- including fopen-fail and the
expected-fail reject builds (scratch is mkdir'd before the build can fail).
`ww run` and explicit-`-o`/byte-id helpers are left as-is; the 990/993
byte-id comparison logic is byte-for-byte unchanged.
Two items filed separately (this commit holds the no-Makefile / no-main.c
rail):
- #13: a stale <src>.s byte-id readback (749) silently no-ops since
separate-compile emits .s to <ostem>.sepwork/__root.s; documented inline.
- #14: build-system Makefile recipes build selfhost/cmd/*/main.ww with no
-o and leak main.sepwork in-tree (bounded, gitignored; own commit).
One concern -- sepwork leak hygiene -- across 228 drivers; uniform
transform applied per-file and two-round reviewed. make test: all 402
passed, zero net-new /tmp scratch, zero test-driven in-repo .sepwork.
ww half of the #77+#78 fused g-fold train; completes the family. cs
half landed as the previous commit (G1) — the two ship together, one
gated train, per the fuse ruling on both tasks.
Root: the global DATA emit walk dispatched on the UNCHASED decl tnode —
a NO-PEEL consumer (zero `.under` tokens on the path; it never learned
aliases exist). An alias-typed global array's N_TNAME matched no arm
and the documented skip-policy ate the decl: w6c_ww referenced
main.g(SB) but emitted zero DATAW → loud `w6l: undefined reference to
main.g` on every direct alias-global array row (ken NEW-1, all k_gidx*
shapes). Every other kind was already chased (letvarisstr/isslice/
isfloat/isstruct walk aliaslookup chains; the tuple gate walks tnodes;
emitarraydata/emitslicedata chase tinfo internally; letemitsize walks —
registration was never the gap), probe-confirmed: only array rows
failed ww-side.
Fix: ONE tichase at the dispatch entry, per the spec's entry-point rule
— not per-arm. Dispatch arms touched (enumerated):
emitletdataw (cgen.ww): hoisted `dti = tichase(d.lhs.type_)` at the
per-decl entry; the isarr8 scalar-shortcut gate and the array arm
now key on dti.kind == TY_ARRAY (were d.lhs.kind == N_TARRAY) and
emitarraydata receives dti; the struct zero-fill arm's inline
TY_NAMED loop collapses into the same dti (ef93b16 precedent,
byte-neutral). str/float/struct/slice/tuple gates unchanged.
letpreintern (cgen.ww): the #18 [N]str array-leg gate keyed on the
N_TARRAY tnode while its body already chased the tinfo — gate now
keys on the chased kind, so alias-typed [N]str globals pre-intern
their _S_ labels in decl order (label-order parity with cstage;
the inner elem chase collapses into tichase).
For non-alias decls tichase is identity (same tinfo pointer) — the
emitted bytes are unchanged by construction; full byte-id invariant
holds (test-unit 288/288 incl. the new table).
main.combined.ww (w6c + wwdump) regenerated by `make` — diff verified
content-identical to the cgen.ww hunks, nothing else.
Graduation table committed as test/wcc/944_alias_global_decl_run.c —
23 rows x {cs run, ww run, byte-id} = 69 checks green. This table IS
the permanent guard: the path is lint-invisible (NO-PEEL — nothing for
the future peellint to see), so only a runtime+byte-id row pins it.
Rows: plain control; alias array 1-lvl read/write/decl-order; 2-lvl
read + order-permuted write (the #78 silent saved-BP rows); [4]u32
narrow-esz; scalar/str/f64/f32 2-lvl; alias-of-named-struct field w/r
(the cs SEGV-at-one-user-level row) + 3-layer + STRUCTLIT init; slice
2-lvl literal; [2]str 1-lvl/2-lvl (letpreintern label leg); def-side
2-lvl array/struct/float; no-regression holds (alias global SLICE,
alias ELEMENT [2]row). Values >255, LAST element asserted.
Probe-OUT rows documented in the test header, filed not pinned: #86
(named-tuple global init: cs checker loud-reject vs ww accept), #87
(plain tagged global: cs silent-wrong vs ww loud-reject, non-alias).
