Go 1.26.5 maps each test binary's stdout and stderr to the same product writer. Teach the captured executor to share one open file description for byte-equal output paths, then make directory test products emit that one ordered capture on coordinator stdout. Build captures and inherited-stdio routes remain unchanged.\n\nKeep the executor mechanism, package policy, native Cstage/WWstage proof, and normative contract together so every commit preserves the observable test-output behavior.
Go compiles only _test.go files as tests; discovery now keys on the
_test.ww suffix alone. The line-leading-@test compatibility allowance
(noncanonical filenames admitted as test sources) is removed from both
driver stages and the coordinator. An @test declaration outside a
*_test.ww file is rejected loudly ("@test declaration outside
*_test.ww", wording byte-identical cs/ww) instead of silently running
under compose or silently dropping in a non-T build (#6). Tree audit
found zero real carriers; the two allowance fixtures flip canonical
(dep_test.ww, widget_test.ww). New pins: direnum attest-noncanon
reject row (both-stage stderr parity) and the coordinator
noncanonical_attest_rejected package row.
Go has no per-file test targets: the PACKAGE is the unit of testing
(`go test ./...`) and same-package test files compose together. The
59-target LIBRARY_TESTS fan-out (per-file -w workdirs, one Make rule
per suite) collapses to one line: `ww test -j $(JOBS) -w
out/wwbuild/wwtest-lib -I lib/ww lib/...` — the regex dir-route
precedent generalized. The lib/regex line and the per-file pattern
rule instance dissolve with it (test/lang keeps its own rule).
Measured before committing (-j4, strings edit row): old per-file
warm 4.7-4.9s; bare walk 5.2-5.6s — a real regression, so the
coordinator first gained the brief's persistent per-package workdir:
`-w DIR` on a package target forwards to wwpackage, which keys
DIR/<dir>_<pkg> per group and hands it to each inner `ww test -c`
build. Reuse stays entirely with the driver's existing
content-identity contract — the coordinator adds pure path policy,
no cache machinery. Both driver stages drop their package-target -w
rejects (forward instead); -w with -c stays rejected at the
coordinator (two ownership contracts). After: 3.9-4.0s on the edit
row, 1.0s warm no-op, 2.5s cold — faster than the old flow on every
row.
package_test's tree -w reject row becomes the positive contract
(cold+warm byte-stable stream, cs/ww same) plus the -c conflict
reject. libbyteid roster shape DECIDED: per-file fx entries stay —
every enrolled file is still standalone-buildable, so coverage is
byte-for-byte unchanged; the dir-mode entry form arrives only with
the B3 shared-helper split that first needs it. Docs: owner table,
target table, -w contract paragraph.
Go does not distinguish bootstrap observers from unit tests: an
in-process test is a unit test. 738 is pure parsefile calls (arena +
lexinit + parserinit, AST assertions, zero subprocess or artifact
claims) — the 200_parse class, misfiled among the residuals. The
fleet partition now reads six units / six bootstrap / three platform
and the residual class is empty; COMPILER_WRAPPER_SOURCES drops it
by the existing filter-out. Doc roster updated (also refreshes the
corpus prose to the current 1745/3490 pin and retires the stale
794-era paragraph).
The last Open-driver-work bullet. `-c -o <name>` replaces the fixed
<package>.test stem for exactly one package; the coordinator rejects a
multi-package fan-out ("cannot use -o with multiple packages", Go's
`go test -o` rule) and -o without -c is rejected at the driver
("needs -c for a package target" — a plain run executes from the temp
root, so a caller-owned name has nothing to name). Both driver stages
byte-identical wording; -S keeps its single-file-only reject.
package_test gains the contract row (naming, fixed-stem absence,
artifact runs, both rejects, both drivers); the tree-mode -o row's
pinned wording follows the contract.
Four ww tests carry the last eight native byteid carriers' assertions:
asmwindow (753 convwrap beta/main/alpha order + window polarity, 754
slice stride + negative scale scan, 755 amp-dot-idx four rows, 758
first-CALL-line extraction + tab-framed disp literals, direct
w6c/w6c_ww never ww build), freenoop (930 byte-id strengthened to
per-stream compare + negative 'free' grep), structabi (946 param/ret
MOVSD windows with polarity tables and SEND/RECV agreement), mangle
(989_m1mangle needles + concat byte-id, glob order strengthened to
byte-lexicographic). Dead want/stage_mask row fields documented, not
invented into runtime legs; the w6c_ww-absent skip gates drop because
the Make target declares the tools.
BYTEID_WRAPPER_SOURCES, its bins, and test-native-byteid are deleted —
the native byte/artifact partition is EMPTY (11 -> 0 this session);
docs counts move to 123 carriers.
25 renames (git mv, content untouched). _test.ww is what the package
coordinator's test detection and the sep loader's canonical exclusion
key on; the old *test.ww spellings survived only through the
line-leading-@test compatibility scan. Consumers updated in place:
LIBRARY_TESTS, the libbyteid roster, the 901/974/975/976 carriers that
copy or invoke these files, and the check.c/check.ww + path/ftos
comments that cite them. Closes the open-driver-work migration bullet.
test/byteid/wwi_test.ww carries every assertion of both carriers: the
M2 producer gate (ascii/strings/getopt driver-combined positives with
byte-id + wwdump -a re-parse, the synth decl-kind/type-node sweep with
the #47 @symbol round-trip, the #48 types export-def gate, and the
negative private-nominal leak with both-stage identical reject
diagnostics), plus the #11 wwileaf decl-less-leaf rows (package-line
leaf, cross-stage .wwi byte-id, importer resolution + .s byte-id).
Byteid carriers 10 -> 8; docs counts move; the byteid ww target gains
the wwdump prereq its re-parse legs invoke.
test/testenv (package testenv) is the shared plumbing for ww-native
test drivers — subprocess launch with captured output, file IO, string
search, byte-sorted directory listing, scratch ownership — lifted from
the proven package_test.ww idiom; a utility, not a framework.
test/byteid/libbyteid_test.ww carries every assertion of the retired C
carrier: the 44-entry roster (fixtures + sentinel-guarded import probes
+ the zero-dep root-only build), both-stage sep builds with the
resolved-unit proof, per-package .s concat compare with the empty-concat
guard, the ID/DIVERGE/WWREJECT pin discipline, and the lib/ corpus
completeness scan (negative-verified against a planted un-enrolled
module). Concat order strengthened from shell-glob to explicit
byte-lexicographic. The roster fills by cursor-driven field writes
because append() rejects struct-call-result sources (#34).
Byteid carriers 11 -> 10; docs counts move; test-byteid runs the ww
test via the wwtest/ pattern with WW_TEST_REPO.
A -w DIR workdir replaces the fresh .sepwork scratch with a caller-owned
persistent package-artifact tree. A package is reused only when its
freshly composed unit byte-equals the committed unit and byte copies of
the compiler/assembler recorded in the dir equal the live tools — pure
content identity, no mtimes, no hashes, every decision reproducible
with cmp against plain files. Recompiles stage at .new names and commit
by rename, unit strictly last, so an interrupted build forces a
recompile and can never leave a committed unit vouching for uncommitted
artifacts; .o/.a additionally reject zero size (ELF/ar are never
empty), while .s/.wwi accept legitimate empties (FFI-only rt). A mode
stamp pins the -T/-S shape and the artifact protocol revision. Classic
scratch keeps its exact acquire/refuse/cleanup contract; run rejects
-w; dir-mode test rejects -w; both driver stages implement identical
behavior and wording.
Ten of the thirteen were already fully twinned — their runtime rows
migrated in earlier waves (r56_def_*, r79_*, r78_*, r926_*,
r989_m1union_*) and the carriers survived only for a cstage-vs-wwstage
byte-compare the blanket test-data-byteid comparator owns once the
source is a corpus fixture. The other three needed their remaining
inline sources added first: r756_* (4, alias-chain unwrap),
r844_bid_src (size(u64) untyped-int compare), r989_structframe_* (3,
struct-local frame layout). Every new fixture was verified both-stage:
compile exit 0 twice, .s byte-identical, runtime exit as pinned.
The five ww_ww-driver-leg carriers among them (785/786/789/790/
m1union) follow the 815/940/951 precedent: content identity owned by
their corpus twins, driver-leg identity owned by 989_lib_byteid's
sweep. Byte/artifact partition 24 -> 11; carriers 147 -> 134; corpus
pin 1495/2990; the stale 915 sweep figure in the docs corrected to
the true 1,157 non-error count.
fnptrcalleetfn's N_DOT arm required an N_IDENT base, so a.b.cb(...)
fell to the name registry with an empty module hint and emitted
CALL cb(SB) (undefined symbol; cstage calls the stamped ptr indirect).
dotbasestructinfo resolves the base chain through the struct registry
— each link a struct- or *struct-typed field — and the single-dot
path routes through the same resolver unchanged. Closes the last open
shape of the #59.8 name-keyed callee family. Fixture fnptrfield_chain;
corpus pin 1487/2974.
installparams normalizes a decl's `T...` param lhs to []T in place
(cstage instead sets the resolved tp->type and never mutates AST), so
typeeqast compared a decl-synthesized fn type's []i64 against a fn
TYPE expr's surface i64 and confidently rejected
`let f: fn(args: i64...) void = sum;` — cstage accepts and runs. The
normalization wrapper now carries an op marker and typeeqast peels
exactly it, so both sides compare at the declared element type without
admitting genuinely different element depths. Fixture fnptrlet_variadic;
corpus pin 1486/2972.
The last frontend-gap pin: wwstage had no Hare struct embedding
(struct { hash.hash, ... }), rejecting lib/crypto/sha256 at parse.
- parse.ww: the three member forms (named / anonymous struct / bare
dotted-ident embed), consume-then-branch since this parser has no
peek; embeds carry f.str == "" and the type in f.lhs.
- check.ww N_TSTRUCT flatten: promote the inner struct's flattened
fields at base+src.offset (check.c:961-990); the embed is one
nested-struct unit in the slot ladder; the resolved inner AST is
planted on the TFIELD rhs for cgen.
- check.ww walkers: astoffset / exprtype N_DOT / #251 struct-lit
field lookups descend embeds through shared helpers; the collision
and non-struct-embed rejects live in validatestructfields (the
once-per-decl diagnostic site).
- cgenutil.ww registerstruct: regfieldrun walks the AST against the
flattened tfield cursor, descending embeds via the planted inner
AST so promoted fieldinfo entries keep the inner field's own name
and type node.
- wwi printers unchanged (both stages already emit nameless fields).
sha256_test compiles byte-identically end to end and its 6 tests
pass; 989_lib_byteid is now 44 id / 0 divergent / 0 wwreject.
Fixtures r5913_* (promoted rw, offset shift, anonymous embed,
two-level embed + promoted fn-ptr callee, three rejects); corpus pin
1485/2970.
fnptrcalleetfn's N_DOT arm accepted only a bare N_TFN field tnode, so
a call through a field declared `cb: *fn(...)` missed, fell through to
the name registry, and emitted CALL cb(SB) — an undefined symbol
(loud at link; cstage calls the stamped ptr-to-fn indirectly). Same
TPTR peel the N_IDENT arm already had. Fixture fnptrfield_call covers
the by-value and via-pointer shapes; corpus pin 1478/2956.
Audit-driven migrate-and-retire wave (one read-only auditor per
carrier batch, verdicts row-checked through the full corpus gate):
every retired carrier's assertions are now owned by declarative
fixtures or an existing owner. 238 new fixtures land (corpus 1,239 ->
1,477; cells 2,954), covering enum/str/slice/tagged array elements,
inferred-length arrays, global slice/str zero-init and literals,
global array/pointer field reads and stores, struct-literal slice
fields, struct returns, float arithmetic, size/int limits, sort and
log-vstream behavior, place-addressed stores, and append places.
Rejection rows STRENGTHEN the old nonzero-exit checks to required
diagnostic fragments; all byte-id claims fold into the blanket
test-data-byteid sweep (now 1,142 compared, 0 pinned-divergent).
Four carriers were fully redundant with existing fixtures/lang tests
(749, 722, 765, and 771's byteid rows) and retire without new rows.
Native carriers 171 -> 147 (residual 135 -> 111).
PROJECT_PLAN.md was a historical disclaimer wrapping five still-open
T1 driver items — those move to docs/test-system-v2.md (Open driver
work); the rest is dated 2026-08-05 design history, archived in git.
Also retired-mechanism cleanup: both drivers drop the *.combined.ww
enumeration skip (the amalgamator is gone; leftover debris now fails
loudly instead of being silently accommodated), the consumer-less
m4_combined_epoch.md5 pin is deleted, and the wwi.ww/ww.h/check.ww
comments that still described combined.ww as the live path are
re-cited to the sep reality.
Wwstage-driver-leg byte identity gets one owner. The three carriers'
content identity is already blanket-owned by their r815_/r940_/r951_
corpus twins via test-data-byteid; their only non-redundant assertion
was the ww_ww driver leg, which 989_lib_byteid's 43-unit sweep covers
apart from the zero-dep root-only edge — folded in as a new probe row
(44 units). r940 fixture comments re-cited off the retired carrier.
The four graduated pin families byte-compared under test-data-byteid
but their original fixtures masked the miscompiles at runtime (lone
frame slot, dead code path, pointer accident). Add the discriminating
programs as run fixtures: adjacent-element clobber for the 2-byte
deref store, second-variant match dispatch for fn-type dedup, payload
integrity across the tagged widen of a padded struct, and base64
clear() actually zeroing its buffers. Corpus pins move to 1,229
fixtures / 137 run / 766 run-exit / 2,458 cells with the new identity
hash; the data-byteid floor moves to 915.
Every probe's assertion is owned by a current gate: the compile and
link probes by make all and the bootstrap fixed point; build/run and
cs/ww byte identity by the fixture corpus, test-data-byteid, and
989_lib_byteid; wwstage driver and toolchain parity by 993/995;
checker-diagnostic parity by the corpus' both-stage //ww:error rows.
The wwdump -t/-a dump-parity probes gated the frontend port's
convergence, which the compiler-output identity gates now own end to
end; carrier ran green at retirement.
What was still uniquely alive migrates: smoke.ww becomes corpus
fixture selfhost_smoke (upgraded from a cstage-only build to both
frontends, byte-identical, exit 42 on both toolchains; corpus pins
move to 1,225/763/2,450 with the new identity hash in the same
commit), and sym_link.ww's scope/sym behavior rows become in-language
lib/ww/syntax/symtest.ww under LIBRARY_TESTS. uses.ww (parser-stub-era
-a fixture) and the already-orphaned tagged_ptr_ret.ww/trypromote.ww
retire with the probe corpus. Bootstrap native gates drop to six;
frontend numeric-sync comments now cite the rule-6 mirror instead of
the retired diff probe.
test-system-v2.md is the authoritative owner table and target graph;
CLAUDE.md rule 13 moves from the sizelint gate to authoritative layout
helpers and rule 14 names the new targets; PROJECT_PLAN.md is marked
historical where superseded.
