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.
An empty zero-length array diverged cs!=ww in asm (both ran correct=7):
a [0]int global emitted a spurious DATAW main.X(SB),"", and a [0]int local
reserved a $16 frame slot. cstage emits neither. wwstage-only, byte-id-only.
The global DATAW emit is now gated on sz > 0 (skips the empty array).
The local frame: localreserve dropped its sub-8 floor (if asz<8 asz=8) to
mirror cstage's localslot formula (frame+sz+7)&~7 -- but that floor was
MASKING slotsize(TY_VOID)=0 (a void local), which cstage defaults to 8B;
removing the floor alone collided the zero-size void slot with a spilled
param (a real miscompile -- 1132 self-compile hunks). So letslotsize now
returns 8 for a void local, while empty-struct / [0]-array stay genuine 0.
The frame formula is byte-id-neutral for every sz>=1 local (round8 already
>= 8); only true zero-size cases change.
cstage unchanged (w6c md5 unchanged). byte-id 990-997 8/8 (the full
self-compile is what caught the void-local class); test/wcc/820 un-carves
the #9 empty-[0] byte-id exclusion + adds void-local/local-[0] rows.
An explicit [N]T = [init] with more initializers than N silently
mis-compiled for N==0: the over-fill length-mismatch check was suppressed
when alen==0, because alen==0 doubles as the [_] infer-sentinel after
resolve_type collapses the two. So def/let [0]int=[1,2] silently resized
(cstage exit 2) or OOB-read/segfaulted (wwstage) instead of the loud
length-mismatch that [N]=[init>N] gets everywhere else.
The AST keeps the distinction the Type loses: [_] leaves the N_TARRAY
length-child NULL, an explicit [N] carries N_INTLIT. cstage adds an
is_infer_arr() helper, drops the alen>0 exemption at the over-fill check,
and gates the 4 infer-resize/no-init sites on is_infer_arr so an explicit
[0] flows to the over-fill -> loud. wwstage flips the one shared count
gate (checkarrlitfits) from declen>0 to arrtn.rhs!=nil, which also
dissolves a wwstage local-resize/module-OOB inconsistency.
[_] inference, [0]=[] empty, and [_]-no-init louding all preserved.
Under-long (count<N) stays out of scope (#10). byte-id 990-997 8/8.
test/wcc/820 table-driven; its one empty-[0] global row carves out
byte-id (pre-existing spurious-DATAW divergence, task #15).
