databasev2 4 part A, task 6. Mostly documentation, plus one real fix the
full battery caught.
THE FIX. The drain held EVERY DB reply until the barrier — including
reads, which stage nothing and have no stake in durability. That parked
readers behind an fsync for no reason: durable.sN.mixread.p99 rose from
~1043us to 4057us. Only a statement that actually staged a record now has
its reply held. Caught by the gate, not by review.
THE TRADE, recorded rather than smoothed over. What remains is inherent: a
barrier blocks the owner shard LONGER (more records per fsync) though LESS
OFTEN, so anything queued behind one waits. Three full runs of the same
build gave durable.sN.mixread.p99 of 1043 / 2318 / 4147us and wmix.p99 of
8758 / 20000us — a 2-4x spread with the box near idle. So part A buys ~3x
write throughput at the cost of a longer, noisier tail on the owner shard,
and that is the strongest argument for part B (submit and keep serving).
- durable.sN.*.p99us tolerance widened to 100% WITH the reason in the
code: a 2-4x-variable tail gated at 50% gates the disk, not the engine.
The floor is the real guard and is not slack — mixread's (4172us) came
within 25us of tripping on the worst run. Baseline refreshed; a fresh
full run then passed 106 checks 0 failures
EXIT STATUS MOVED 3 -> 74 (sysexits EX_IOERR). 3 and 4 are already used by
SAMPLES for their own meanings — db-bench's own `verify` exits 3 on a
checksum mismatch, and it is the gate that exercises durability, so a
durability abort exiting 3 would have been indistinguishable from the
mismatch it should help diagnose. The low range belongs to programs.
Docs:
- story: progress, the payoff measured two ways, the cost side, criteria
split met/outstanding, and a "part B — its premise changed" section:
it was justified by "close the 66x gap", but that gap is two problems
and only the concurrent one was a batching problem
- board: standup entry in the six-question shape; both databasev2 4 rows
rewritten. They had said "close the 66x gap" — recorded as MIS-STATED
rather than quietly renumbered
- 00-wob-format.md and 04-db-binding.md: the normative failure contract
("a failed WAL commit traps WO_T_IO after un-applying the row") was
false; corrected, along with the tick-scoped group commit that never
happened
- database/src/CODE-LOGIC.md: where the barrier runs and why there, why
replies are held, why the inline path is asymmetric, the one failure
rule, and how to measure it
- db-bench README: the wmix mode, the env knobs, and the tmpfs warning
Battery: wovm-test 36 suites 0 fail, woc-test, oop-e2e 119/0,
db-bench 106/0, linkcheck clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Float full stack: literals (fraction/exponent; `0..10` still a range), f64
opcodes 34-41, @table column, WAL bit-exact replay, json fractions in and
shortest-round-trip out. IEEE-quiet — FDIV never traps where DIV does.
- Bytes: a wo_str with its own class id, so alloc/free/copy are shared but no
Text builtin accepts one; len/at/slice/eq/concat, base64 both ways, json
boundary as base64; TEXT_COPY preserves the kind.
- No implicit Int/Float mixing (WO-E201 in the typechecker, not the emitter,
which picks the opcode from one side and would misread the other).
- One IEEE deviation: float_cmp total order (NaN last, -0.0 == +0.0) for
indexes and order-by, keys canonicalized to match. `?Float` nil is a
reserved quiet NaN — the zero word is +0.0, WO_NIL_SCALAR's bits are -2.0.
- Renderer prefers fixed over exponential in 1e-6..1e21: pure shortest makes
a price of 900.0 read `9e+02`. One renderer for interp/json/float_to_text.
- Fixed en route: lexer double-counted the leading digit; is_scalar_shaped
took Float/Bytes as Int-shaped; Bytes ownership needed a shared heap-scalar
predicate or temps never dropped; order-by bit-compared negatives backwards.
- Iteration 17: `kind = "library"` (absent = program; bad value = WO-E109),
entry-less check mode retiring the `--emit` workaround, Go's `internal/` as
WO-E108 at the consumer's `use`. Driver-only; VM/.wob/GC untouched.
- Framework reorg: internal/{parse,serve}.wo; http/form.wo split out to keep
media_type/form_values public (parse.wo had grown public surface).
- Docs: link audit (97 -> 88 broken, conflict markers resolved, 2 duplicate
stories removed), 00-code-review verified 26/27, iterations re-sequenced.
- Also carries the pre-staged pub(read)/using/#if work from the index.
- Gates: corpus 103/0, test_wal 156/0, web-app 26/0, oop-accept ALL MET.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Closes json's two documented fidelity limits (iteration 5 strictness):
- field_class gains WOB_FIELD_BOOL (a plain `Bool` field) and
WOB_FIELD_NIL_BOOL (a `?Bool`: WO_NIL_SCALAR nil + bool encoding) — the
kind byte alone cannot tell a Bool slot from an Int slot, so the metadata
carries it. Emitter writes them (field_class_meta); loader whitelists
them; json.c encodes `true`/`false` (and `null` for a ?Bool nil), decode's
null/omitted-key pre-write covers NIL_BOOL.
- A JSON number with a fraction or exponent is MALFORMED for an Int field:
the checked decode (`json.decode(t) as T`) yields nil for the whole
document instead of silently truncating 3.7 to 3 — the language has no
float, and corrupting data quietly was the one thing a "checked decode"
must never do. Floats stay representable through a raw `json.Value` field.
- corpus: run/json-bool-fidelity pins the round-trip (true/false both ways,
?Bool null both ways, fraction AND exponent rejected).
- Board's two known-gap entries struck; format doc's field_class marker list
extended.
Verified: oop-e2e 87/0; runtime test + test-iso OK; woc-test 540/0;
log-watcher 7/0; employee 8/0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The spec's §8 migration table, applied:
- 00-principles.md P3: "@gc is a per-class opt-in, reference-counted" ->
GC-ness is inferred; incremental per-shard mark-sweep in budgeted slices;
still no global pause by construction.
- OOP spec: decision-table GC row -> inferred (hybrid rule named); §3 rule 5
-> traced classes alias freely, which classes is inferred; §4 memory model
-> the RC + Bacon-Rajan paragraph replaced by tracing (snapshot roots,
Yuasa barrier, born-black, budgeted slices); header rc comment -> union'd
sweep link; mixing rule restated for tracing.
- 00-wob-format.md: header says version 4; opcodes 27-28 -> reserved (loader
rejects); the owned-temporary rule's @gc exclusion restated for tracing.
- 08-builtin-surface.md: the push RC_INC special case and the set(m,k,v)
retention gap DELETED — neither exists without RC; the corpus cycle is
collected by tracing.
- story 07b: status -> LANDED 2026-08-18 (with the historical note kept);
board: 7b row ✅ (supersedes iteration 2's RC memory model), pending row
removed.
- gc-cycle README: Phase 3 flipped to landed (the ring runs, is reclaimed,
ASan-clean; the ?Node RC_DEC-on-nil trap no longer exists); the barrier
prose corrected to the as-built design (snapshot-at-beginning + deletion
barrier + born-black, not per-slice root re-reads).
- plan 2026-08-18: all checkboxes ticked + a completion banner recording the
four deviations from the plan as written.
(Error catalog was already amended with the keyword-removal commit: WO-E104
added, WO-W201 retired.)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- compiler: `insert Class { ... }` is a typed Ast.Insert in statement
AND expression position, sharing the ctor literal's field grammar;
typechecked with the ctor's omittable rule; result = the row id (Int)
- owner pass: the engine copies at the row API, so an insert BORROWS
its field values -- no transfer, no E304; node is trap-capable and
carries a live-mask drop entry like DbStub did
- emit: builtin 61 window = class-id const + one slot per DECLARED
field in declaration order; omitted defaults emitted, omitted ?scalar
gets WO_NIL_SCALAR, other omitted optionals the zero word; fresh
argument values reaped after (the push/set copy semantics)
- runtime: database/src/db.c executes via the choke-point row API;
rt.db/rt.wal opaque handles on wo_rt; WO_DATA=<dir> = replay
<dir>/shard-0.wal at boot + commit-before-ack per statement (the
builtin's return IS the ack until iteration 8 ticks); failed commit
un-applies the row and traps WO_T_IO; loader validates the class-id
slot (variable window documented in wob.h + format doc)
- the promised diff: trap/pricing-set-price-db-stub is now
run/pricing-set-price-insert printing engine-allocated ids;
durability smoke prints 1,2 then 3,4 across two WO_DATA runs
- old "bare insert is an Ident" unit test rewritten to the new
contract; runner's loader mirror accepts id 61; goldens re-blessed
- gates: oop-accept ALL CRITERIA MET, oop-e2e 71/0, woc-test 566/0,
wovm-test green, log-watcher 7/0
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Found by running the compiled log-watcher, not by reading code: the supervisor
rejected every cron line ("malformed schedule: * * * * *") because a `*` field
expands to 0 and `?Int`'s nil was also 0, so `a == nil` was true for a real
value. Both log-watcher subcommands now behave: `watch` alerts on a live file,
`run` reports SCHEDULE /var/log/backup.log: * * * * *. corpus 71/0, woc 565/0,
wovm gates green.
- a nullable SCALAR (?Int/?Bool/?Timestamp/?Id) spells nil as WO_NIL_SCALAR
(-2^62), not the zero word. Heap-shaped optionals keep 0 — a null pointer is
unambiguous. The value is -2^62 and NOT INT64_MIN on purpose: the compiler's
integers are OCaml's 63-bit natives, so INT64_MIN is not expressible there
(and `min_int * 2` silently wraps to 0 — the first attempt did exactly that)
- the class table marks such fields (WOB_FIELD_NIL_SCALAR in field_class), so
the runtime writes the right absence where it produces absence itself:
json.decode leaving a key absent or seeing `null`, and parse_int on
unparseable input (so parse_int("0") is now distinguishable from a failure).
json.encode renders a nil scalar as JSON null
- emit.ml: `nil` takes its word from its destination (annotation, field,
return type); a comparison against `nil` emits the literal with the other
operand's type, so ?scalar compares against the sentinel and ?heap against 0
- vm.c: EQS accepts a nil operand — two `?Text` values compare with it, and the
answer is "both absent is equal, one absent is not". Trapping there made
`a != b` on optionals unusable (it was trapping BOUNDS "null text" in the
supervisor's rescan). A non-nil operand must still be a real Text
- docs: both normative docs now state the heap-vs-scalar nil split and the EQS
rule; the stale duplicate vm_unwind comment is gone
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
docs/examples/log-watcher (1285 lines, 7 files) now compiles clean: 0
diagnostics, a 35KB .wob written. corpus 71/0, woc runtest 565/0, every wovm
unit gate green (both dispatch flavors).
- .wob v2: each class row gains three u32 per-field arrays — the field's NAME
constant, the CLASS it refers to (or the json-raw marker), and a container
field's ELEMENT kinds. json is then a runtime service driven by metadata
instead of per-type generated code. loader/emitter/disassembler/test
assembler all read and write v2; field-name constants are interned with the
rest of the pool (interning during serialization silently loses them)
- runtime/src/json.c (new): encode by static kind + object headers + class
table (nested records need no static knowledge); decode parses and BINDS
straight into the target class — keys matched to field names, nested objects
built as the field's class, arrays as a multi of the field's element kind,
unknown keys skipped, absent keys nil. Malformed input is nil, never a trap
- `as`: `json.decode(text) as T` is the one cast this language has (WO-E403
for any other `as`, and for a bare json.decode with no target type). Its
result is `?T`, which is why the decode and the target are one instruction
- json.Value: a reserved type name for a value the source does not inspect —
the raw JSON slice, kind TEXT, re-emitted verbatim by encode
- docs: 00-wob-format.md is now the v2 reference (class metadata, TRY/ENDTRY,
the whole builtin surface, WO_T_IO); 08-builtin-surface.md documents the
text/container builtins, the OS modules with their predeclared records, and
json's two documented limits (Bool encodes 0/1, floats truncate)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Modules: `use`/`pub`, directory-as-module, per-module symbol resolution (a
flat first-wins merge silently ran the wrong `pub fn` body), six reserved
stdlib namespaces typed UNKNOWN-BUT-RESERVED.
- Surface: `and`/`or` (own precedence tier, short-circuit, Bool-only), `${}`
interpolation desugared at parse time, `const`, break/continue with
drop-correct exits, do-while, inline-fn rejection.
- switch expr/stmt: required `default` over scalars/Text, arm unification,
EQ/EQS+JZ lowering, per-arm drop scopes with N-way JOIN-DROP; `default`
sorted last by a shared lowering order (textual order made arms dead).
- typedef records: structural, same shape = one class entry; `?name: T`
nullable-by-shape; emit_ctor fills omitted defaults; `type` as field name.
- Enum variants: all-bare unions = int ordinals; any-payload = one class
entry per variant, tag IS the header class_id (no header field, no format
bump); exhaustive switch without `default`; arity checked both directions.
- Payload escape modeled as move-out (pointer-kind fields only — a scalar
escape is a copy); caller reaps owned heap temps passed by borrow: two
unbounded LSan-blind leaks, 10.5 MB -> 1.5 MB flat over 300k iterations.
- Fixed en route, each with a RED repro: dead E209 builtin-arg check and
`int_to_text` missing from both types.ml builtin tables (both segfaulted
wovm), multi-file phantom double-report, emit_ctor's field temp clobbering
dst in tail position (pre-existing), warnings swallowed without an error.
- Two fenced VM builtins: `int_to_text` (13), `variant_tag` (14).
- 14+565 unit (was 14+401), corpus 71 (was 32) plain and under wovm_asan,
wovm-test + cli_smoke green. Log-watcher 307 -> 93 diagnostics (85 E101 /
4 E207 / 1 E208 / 3 W202); the 5 non-E101 residuals await Task 7 grammar.
- `woc` now emits `.wob` that `wovm` runs: emit.ml lowers the typed,
owner-annotated AST (scope-stack registers with a >64 WO-E401 diagnostic,
Lua-style call windows, ICALL by slot, dedup const pool, drop maps, line
tables, implicit terminators); disasm.ml backs `--dump-bc` goldens.
- Ownership lowering consumes the four owner tables verbatim; RESIDUAL is the
only source of borrow ops, coalesced per operand. Review caught the emitter
consuming only 2 of owner.ml's 4 residual producers — an assignment-anchored
aliasing violation ran to exit 0 instead of trapping; fixed, plus a backstop
raising WO-E404 for any residual region left unconsumed.
- Conformance harness `scripts/oop-e2e.sh` (`just oop-e2e`): four fixture
kinds with exact outcomes — byte-exact stdout, one WO-E### anchored on
`error CODE:`, numeric trap code, gc trace. 25 fixtures incl. pricing-demo
logic, the ownership suite, and DB_STUB's parse-but-trap. `tests/` un-ignored
so the corpus is actually tracked.
- `woc build` produces a self-contained binary: wovm copy + appended image +
20-byte trailer, self-exec via /proc/self/exe. Verified relocated outside
the repo, argless, and against adversarial trailer corruption.
- Milestone 1's five spec criteria all MET (`just oop-accept`). Criterion 3
closed by WO-E405 — the entry must return `Int`, since program mode already
says its return value is the exit code — which deletes the leak class
without adding return-type metadata to the format. `gc/held-cycle` retired:
an externally-held cycle is not expressible in a post-exit pump.
- New spec: inferred GC + incremental per-shard tri-color mark-sweep, retiring
`@gc` and reference counting. Story gains iterations 7b (that work) and 9b
(`@table`, relations, compiler-checked query); `.dev/reference` gains a
sparse System.Linq checkout. Priority: 5→6→7 (log-watcher) then 7b, 8, 9, 9b.
Completes plan 2 Tasks 7-8. owner.ml: mutable-value-semantics flow analysis
producing the four plan-3 emitter tables (moves, drops incl. LIVE-MASK for trap
unwinding, rc with elision, residual borrow sites) plus WO-E301-304 two-site
diagnostics. Alias questions run over canonicalized places, so a double-mut
reached through let-bound aliases lands in the residual table like the direct
form; dump.ml's contract notes the emitter must coalesce guards per operand.
main.ml: directory discovery, cross-file programs (symbols merge before bodies
check), diagnostics ordered by (file,line,col), new WO-E214 for a name declared
in two files. New docs/plan/oop-vm/01-error-catalog.md (14 emitted + 10 reserved
codes), un-ignored so both plan tracks can cite it; justfile regains woc-*.
builtin_scalars is now the five that work: Int, Bool, Text, Timestamp, Id.
Money/SKU/Float and the abstract_types allowlist are gone — `abstract` never
lexed, and Float had no literal syntax and no wob kind, so no value could exist.
Fixtures and samples retype Money->Int, SKU->Text. The abstract newtype feature
is rejected outright (verdict row adopt->reject); haxe-parity Task 7 keeps `is`.
nullable-types-implementation.md corrected: ?T is plumbed but UNENFORCED
(E211-213 declared, never emitted; probe exits 0), handed to haxe-parity Task 6
as next work item. Records all 10 dead codes incl. E205 — interface satisfaction
is unchecked. crates/rt keeps its Money/SKU fixtures (opaque strings, Stage 2).
Gate: build warning-clean, 14 + 264 checks 0 failures, pricing golden exit 0,
docs/examples histograms unchanged (13/70, zero WO-E225).
