- runtime/src/CODE-LOGIC.md: gc.c row describes the tri-color mark-sweep
(traced list, snapshot roots, Yuasa barrier, budgeted slices) instead of
RC + Bacon-Rajan; obj.c row gains the traced-list/may-gcref notes; main.c
row gains the post-exit pump.
- compiler/src/CODE-LOGIC.md: pipeline diagram gains gcinfer.ml between
types and owner; the owner-tables section drops rc sites and names the
inference pass as the source of GC-ness.
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>
The compiler no longer emits reference-counting ops anywhere, and the format
reserves them. With Phase 3a's collector this completes the runtime half of
iteration 7b: spec success criteria 3 (no RC ops in any image, opcodes
reserved) and 6 (corpus ASan-clean) are met — `just oop-accept` is fully green.
- owner.ml: the rc machinery is deleted outright — rc_site/rc_op types, the
rcs table, fn_rcs/rc_groups/rc_escaped, record_rc, release_gc, gc_escape,
resolve_rc, and the clobber rule (its only consumer was elision). The
`push`-of-a-gc-value RC_INC special case is gone (the bug class cannot recur
without RC). Drop tables (owned + LGc kinds) are untouched — the gc mask is
what feeds the collector's root maps.
- emit.ml: emit_rc, the v_rc view, the escape-acquire anchor, and every
caller deleted; assignment displacing a traced value emits nothing (the VM's
store barrier owns it); scope-ended LGc handles clear their gc-mask bit so
root maps stay precise.
- .wob v4: WOB_VERSION 3 -> 4 in wob.h + emit.ml + disasm.ml + the runner's
loader battery; opcodes 27-28 removed from the enum/jump table/interpreter
and REJECTED by the loader like any unknown opcode.
- dump.ml: the == RC == owner-dump section is gone; 6 goldens re-blessed
(owner dumps lose the section, elision.wo's bc dump loses its RC ops).
- runner.ml: rc-table/ELIDED assertions deleted; the elision test now asserts
the WHOLE image contains no RC op; the table-contract sweep asserts rc ops
never appear.
- test_unwind.c: the rc-opcodes test becomes two — the loader rejects reserved
opcode 27, and an abandoned traced instance is freed by rt_destroy
(ASan-proven).
Verified: woc-test 540/0 + test_diag 14/0; runtime test + test-iso all suites
ASan/UBSan (test_unwind 12/0); cli_smoke; oop-e2e 79/0 (v4 images end to end);
employee 8/0; log-watcher 7/0; ring runs + reclaimed (freed=3) with zero RC
ops in its image; `just oop-accept` ALL CRITERIA MET.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reference counting and Bacon-Rajan trial deletion are gone from the runtime.
Traced (inferred-gc) objects now die only by the collector; owned values keep
deterministic drops exactly as before.
- wo_hdr: rc retired; borrow and the freed 4 bytes become a union — non-traced
values keep the borrow word, traced objects use the 8 bytes as the intrusive
sweep-list link. Header stays exactly 16 bytes. WHITE is now the all-zero
color (allocations born white by memset); WO_F_BUF retired.
- gc.c rewritten: snapshot-at-beginning tri-color mark-sweep. Roots (frames'
gc+owned masks) shaded atomically at cycle start; Yuasa deletion barrier
shades the OLD target of every gcref edge deleted while marking (SETF
overwrites + every owned-death path, which all funnel through wo_drop_kind's
GCREF case); allocations mid-cycle born black. Mark AND sweep budgeted
(WO_GC_BUDGET objects/slice), sweep resumes via a cursor; gray-worklist OOM
degrades to a blacken-all cycle (frees nothing, never wrong). Owned interiors
walked eagerly (single-owner trees), pruned by a per-class may-gcref bit
computed at rt_init (fixpoint over kinds + v2 field_class/field_elem;
conservative when metadata is absent).
- vm.c: safepoints at NEW (the heap-goal trigger), CALL, and backward JMP;
root scan follows vm_unwind's governing-pc convention. Unwind's gc-mask
branch just nulls the register. RC_INC/RC_DEC are accepted as no-ops until
the emitter stops producing them (next commit) — which also deletes the old
RC_DEC-on-nil trap that broke `?Node` gcref field stores.
- main.c pump: post-exit, a rootless cycle frees everything unreachable in
budgeted slices; the trace line moved into wo_gc_slice (one format for pump
and in-program slices). rt_destroy frees traced remnants (trap paths, tests).
- WO_GC_GOAL joins WO_GC_BUDGET/WO_GC_TRACE as an rt-owned knob (default 256
KiB; a tiny goal forces mid-program cycles for testing).
- tests: test_cycle.c rewritten (abandoned cycle freed, rooted cycle survives,
slices bounded, cycle-through-multi, repeated-cycle leak-freedom, and the
spec's load-bearing DELETION-BARRIER test: an object hidden behind a black
object mid-mark must survive). test_rc.c re-pinned to owned drops + the
owned/traced boundary; test_obj.c asserts tracked-white-linked instead of
rc=1.
Verified: make test + test-iso (all suites, ASan/UBSan; test_cycle 42/0,
test_rc 14/0) + cli_smoke; oop-e2e 79/0 (gc corpus traces unchanged: the new
slice math reproduces steps=1/freed=2 and steps=2/freed=4); employee 8/0;
log-watcher 7/0. THE RING RUNS: docs/examples/gc-cycle prints
`ring a -> b -> c -> a`, is reclaimed post-exit (freed=3 remaining=0), is ASan
clean, and survives an in-program cycle while rooted (WO_GC_GOAL=64: mid-run
slice frees 0, post-exit frees 3).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`@gc` is no longer part of the language: a developer never writes or mentions
it. GC-ness is decided entirely by inference (structural cycles + demand
promotion), which the earlier 7b commits made complete and precise.
- parser: `@gc` on a class is now WO-E104 ("GC-ness is inferred; run
`woc --dump-gc`. Remove it."). `is_gc` stays false; the class classifies by
inference. No `.wo` in the repo carries `@gc` anymore.
- types.ml: retired the WO-W201 machinery (suggest_gc_annotation,
has_recursive_structure(_type), has_unique_field, gc_suggestion_code) — it
suggested `@gc`, now obsolete since inference traces exactly those classes.
- runner.ml: deleted the 8 WO-W201 gc-suggestion test blocks; the @gc-exemption
test's `Cache` is made self-referential so inference classifies it gc without
an annotation.
- fixtures: dropped `@gc` from rc.wo (Cache demand-promotes via its escape),
elision.wo (Cache given a self-ref to stay structurally gc for the rc-elision
dump), pricing-demo.wo (PriceCache doesn't escape -> now owned), and the
abandoned-cycle/budget-steps corpus (Node is structurally gc). rc.wo keeps a
placeholder comment line so its line-indexed rc assertions hold. Goldens
re-blessed.
- docs: error catalog gains WO-E104 and marks WO-W201 retired; gc-cycle README
records the keyword removal.
Verified: woc-test 553/0 (was 566 minus the 13 retired WO-W201 checks),
test_diag 14/0, oop-e2e 79/0, employee 8/0, log-watcher 7/0. `git grep '@gc'`
finds only comments — success criterion 1 met.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Wire Gcinfer.infer into runner.ml's owner_str/emit_str so the unit tests
classify GC-ness identically to the driver (prerequisite for removing @gc: its
tests read the golden .wo through the library).
That exposed owner-err/borrow-escape.wo, which tested WO-E304 on *class*
escapes — now legally demand-promoted. Retargeted it to CONTAINER (`multi Text`)
escapes, which are owned and never promoted, so it still exercises the three
WO-E304 shapes (stored-in-field, returned, moved-to-take). Assertion positions
+ golden re-blessed.
Verified: woc-test 566/0, test_diag 14/0, oop-e2e 79/0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The escape hook promoted the borrow-root local's class, so `return h.box`
over-promoted the container `Holder` alongside `Box`. Now `owner.ml`'s
`transfer` passes the escaping place's type (`place_ty p`) as `~promote_class`,
so only the value that actually escapes is promoted.
- escape: takes ~promote_class; records it in collect mode, else reports WO-E304.
- borrow-escape.wo now promotes only Box (was Box + Holder); rc.wo sans @gc
still promotes Cache.
Verified: woc-test 566/0, test_diag 14/0, oop-e2e 79/0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Extract the structural+demand classification out of main.ml's typecheck_all
into `Gcinfer.infer : parsed -> symbols -> symbols`, so a single library entry
point runs the whole pass. The driver calls it once (before typecheck); the
unit-test helpers can call the same function, so is_gc_class classifies
identically in the binary and in tests (prerequisite for removing the @gc
keyword, whose tests read the golden fixtures through the library directly).
- dune: gcinfer moved after owner (it now runs ownership in collect mode).
- main.ml typecheck_all: the split structural-inject + post-typecheck demand
loop become one `Gcinfer.infer parsed syms` call.
- Documented the known demand-promotion imprecision (promotes the escaping root
local's class, over-promoting the container) to refine with Phase 3.
Behavior-neutral: woc-test 566/0, test_diag 14/0, oop-e2e 79/0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Structural inference (2a) covers cyclic classes; this adds the DEMAND half for
the acyclic-but-aliased case, so @gc is now redundant everywhere.
- owner.ml: a `promote` sink on ctx. In collect mode, a class value that would
raise WO-E304 (escape) records its class instead of erroring — because a
class that MUST escape cannot be owned (second-class borrows can't be stored
or returned), so it must be traced. `class_of_ty` extracts the class from the
escaping local's type. analyze/analyze_fn take ?promote.
- main.ml typecheck_all: after structural injection, a fixpoint runs ownership
in collect mode over every file, unioning promotions into syms.traced (and
every module table), before owner/emit see it. Terminates (promotions only
grow, bounded by class count).
- gcinfer.render_final: --dump-gc now reads the authoritative is_gc_class
(structural + demand + the remaining @gc bridge), with the reason.
Effect: a class that escapes is inferred `gc` with no annotation — e.g. `Cache`
(rc.wo sans @gc) shows `gc (alias escape (demand))`; `Box` returned out of
`leak` is promoted and the program is valid. No false positives: employee's
Department/Employee stay owned; the 566 goldens + 14 test_diag unchanged.
Corpus: compile-fail/borrow-escape-return reclassified to run/ (prints 1) —
returning a borrowed class is now legal under demand promotion; the fixture
encoded pre-7b behavior.
Verified: woc-test 566/0 + test_diag 14/0; oop-e2e 79/0; employee 8/0;
log-watcher 7/0.
NOT in this slice: removing the `@gc` KEYWORD (parser rejection + rewriting the
RC/@gc golden + test_diag assertions + moving the inference injection into the
library so unit tests see it) — coupled to Phase 3, which deletes the RC
machinery those tests cover. The ring still needs Phase 3 to RUN (nullable
`?Node` gcref path).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
GC-ness now comes from the inference pass, not only the annotation. A class is
traced if the structural SCC put it in `syms.traced`, OR (temporary bridge
until demand-promotion lands) it still carries `@gc`.
- Types.symbols gains a `traced : StringSet.t`; is_gc_class reads it (union'd
with the surviving @gc annotation). All symbols literals + both merges carry
the field.
- typecheck_all injects the classification once (Gcinfer.classify -> traced)
into the merged table AND every module table, before typecheck/owner/emit.
- emit.ml routes the class gc-flag and the union/drop decision through
is_gc_class instead of the raw `.is_gc`, so structurally-inferred gc classes
get the runtime flag. Field-kind derivation already routed through is_gc_class.
- gcinfer.traced_names exposes the traced set for injection.
Effect: docs/examples/gc-cycle now COMPILES with no annotation (the WO-E301
use-after-move at the ring-closing store is gone) — traced classes alias
freely. Bytecode is byte-identical to writing `@gc class Node`.
Verified: woc-test 566/0 (goldens unchanged — every current @gc class stays gc
via the annotation branch, and no golden has a structural-gc-non-annotated
class); oop-e2e 79/0 (gc corpus green).
Not in this slice: demand-promotion (the acyclic-aliased PriceCache case still
needs the @gc bridge) and @gc-in-source-as-error (Phase 2b); the ring RUNNING
(the RC runtime doesn't implement nullable-gcref `?Node` fields — Phase 3).
WO-W201 still fires on gc-cycle (retired in Phase 4).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Structural half of the inference pass, additive — it backs `woc --dump-gc`
and does NOT yet feed field-kind derivation (that is Phase 2 at the
Types.is_gc_class seam), so no emitted bytecode or golden changes.
- compiler/src/gcinfer.ml: build the class-reference graph (edges from
Scalar/Multi/Map fields, unwrapping ?; ref and backlink contribute NO edge),
run Tarjan SCC, classify any class in a non-trivial SCC or with a self-loop
as `gc`, else `owned`; carry a cycle-path reason.
- --dump-gc mode in bin/main.ml (mirrors --dump-owner) + usage line + dune.
Verified:
- docs/examples/gc-cycle -> `Node gc (cycle Node -> Node)`, `Segment owned`.
- docs/examples/employee -> Department/Employee both `owned` (ref/backlink
make no edge, so no false cycle) — the load-bearing correctness case.
- just woc-test 566/0 (goldens untouched, build clean both flavors).
Remaining Phase 1: a --dump-gc golden fixture (deferred — verified manually to
avoid golden-harness churn this slice). Phases 2-4 per the plan.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Design-first deliverable for iteration 7b (no runtime/compiler code yet).
docs/examples/gc-cycle explains, by example, how pointers flow through the heap
and the collector's mark-sweep logic:
- types.wo: Node (self-referential ?Node -> inferred `gc`/traced) vs Segment
(acyclic -> `owned`, deterministically dropped)
- main.wo: ring_demo builds a->b->c->a and abandons it; owned_demo shows the
drop path with no collector
- README.md: the model (ownership frees the 99%, tracing only the cyclic/
aliased residue, inference decides), the 16-byte header rewrite (retire
rc+borrow -> 8-byte sweep-list link, colors in flag bits), where a traced
pointer lives (root via pc gc-mask / GCREF field / container), and the
tri-color incremental algorithm with the Yuasa deletion barrier. Two mermaid
step diagrams (heap+roots, collector cycle) + the owned contrast.
Grounded in the approved spec (2026-08-11-inferred-gc-mark-sweep-design.md) and
the real runtime structures (obj.h/wob.h: wo_hdr, WO_K_GCREF, arena, wo_obj_size).
Run status: honest — the sample does NOT build today; woc reports WO-E301
(use-after-move at the ring-closing store), which is exactly the aliasing that
"traced classes alias freely" unblocks under 7b. README records this.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Master now reflects only the current woc/wovm project. The Rust `wo` runtime
was the prior, abandoned architecture; it is fully independent of the woc/wovm
stack (dune + make, no Cargo dependency), so it lifts out cleanly. Recoverable
via git history.
Removed:
- crates/ (29 files) + Cargo.toml + Cargo.lock — the Rust runtime workspace
- prototypes/ — wo-rt-c (a stale duplicate of runtime/) + wo-db C++ ref
- justfile: the rt-c-demo / rt-c-bench recipes (drove prototypes/wo-rt-c)
- docs/plan/05..16 (11 Rust engineering plans) + docs/plan/done/ (4 Rust
Stage-2 done plans)
- docs/runtime/ (11): the old runtime overview + 7-phase DB design series +
async/fibers/gc/surreal concept essays
- docs/cm.md — legacy scratch note
Kept: compiler/, runtime/ (C VM), database/, the current-track docs
(stories, superpowers, plan/{oop-vm,compiler,exploration}, examples), the
discarded/learnings registers, and the syscall/postgres/assembly/c-runtime
studies. Follow-up commits fix the status board, project-structure doc, and
any dangling links to the removed docs.
Verified: woc + wovm still build; woc/wovm --version green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The ##ui/.htmlx/LIVE track (plan 7) was removed as stale, so drop its dangling
references in 08-project-structure.md: the `ws, sub, htmlx, assets (plan 7)`
runtime/src line, the `06-ui-live` oop-vm contract entry, the `plan 7
UI/.htmlx/LIVE` server-track step, and `ui` in the build rule-of-thumb.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`client/` (wo-live.js live-patch runtime) never existed and was tied to the
removed plan 7 (##ui/.htmlx UI track). Dropped its tree entry, its proof-layer
section header + bullet, and its lifecycle-table mention (plans 4–7 -> 4–6,
http/ui -> http).
Left as-is: the Rust-track phase prose still names a `ui` crate scaffold and
plan-7 sequencing — that describes the maintained (non-advancing) Rust
roadmap, not a removed file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cleanup off master. None of this is referenced by the current woc/wovm
toolchain or the maintained Rust crates; it belongs to abandoned/scratch
state.
Removed:
- .planboard/ + .planboard.json — planboard task-tracking state for the
haxe-parity plan (tooling scratch)
- prompt.md — a stray one-line note about the old runtime's 13c task
- infra/ (deploy.sh, setup.sh, sync.sh) — deploy scripts for the old Rust
runtime (cargo build -p wo-rt, scp to writeonce.de, nginx/SSL/systemd)
- static/ + templates/ — the v1 blog's .htmlx frontend assets (svg/css +
about/article/home/layout templates); the UI track that used them was
already removed as stale
- content/ data/ wo-data/ — empty untracked v1 runtime dirs
Updated docs/08-project-structure.md: dropped the "v1 blog operating assets"
tree line and section. (cm.md keeps a historical mention of infra/sync.sh as
a legacy note — left as-is.)
Nothing kept references the removed paths (verified tree-wide; crates/rt hits
were `'static` lifetimes, not file reads).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The manifest pinned `[build] runtime = "../../../runtime/wovm"`, a repo-only
relative path that overrides woc's runtime resolution — so `woc <copied-dir>`
failed off-repo (e.g. an installed toolchain on a test server) with "runtime
binary not found".
- Drop the [build] section: the project no longer hardcodes a machine path, so
an installed `woc` self-locates `wovm` beside its own binary.
- The module justfile's `build` recipe now sets WO_RUNTIME=<repo>/runtime/wovm
so the in-repo build still uses the freshly built VM.
- Acceptance is unaffected (it compiles via `woc --emit` + an explicit $WOVM,
never the manifest [build] key).
Verified: just log-watcher::build OK; just log-watcher 7/0; and building a
copied tree with the installed-layout woc from an unrelated cwd self-locates
the sibling wovm and produces a runnable binary.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Close the 3 gaps between "builds in the repo" and "installs from a tarball
like Go", so writeonce can ship to other developers.
- VERSION file at repo root single-sources the toolchain version (0.1.0).
- `woc version` -> "writeonce 0.1.0 linux/amd64"; `wovm --version` -> "wovm
0.1.0" (stamped by the Makefile from VERSION; --version handled only in the
plain-wovm path so a built app never shadows its own `version` arg).
- wo.toml `[runtime] wo = ">= X.Y"` is now ENFORCED: woc refuses a project
requiring a newer toolchain than itself (>= and bare version parsed;
unknown operators accepted forward-compatibly). Was parsed-and-ignored.
- woc self-locates wovm: --runtime > [build] runtime > $WO_RUNTIME > a `wovm`
beside the woc binary (Sys.executable_name) > runtime/wovm rel CWD. An
installed woc in <prefix>/bin finds its sibling wovm from any cwd.
- `just dist` (scripts/mkdist.sh) packages writeonce-<ver>-linux-amd64.tar.gz,
Go-shaped (archive root writeonce/, bin/{woc,wovm}, README, VERSION), with a
drift guard asserting VERSION == woc == wovm. dist/ gitignored.
- `just install-accept` (scripts/install-accept.sh) is the gate: extract, PATH,
version, build+run a project from an unrelated cwd, constraint refusal — 6/0.
Verified: install-accept 6/0; woc-test 565/0; wovm suites + cli_smoke;
log-watcher 7/0. Linux-amd64 only (a cross matrix is future work).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Analyzed the full 131-file docs tree (4 parallel classifiers) against the
shipped woc/wovm toolchain. Removed 21 stale docs, kept all intentional
history (Rust-track plans/done, the runtime/database design series cited by
current specs, syscall/postgres/assembly/c-runtime studies, discarded/
learnings). Deleted:
- old-runtime "front door": writeonce-pl.md, runtime/wo-language.md
(pitched the Rust wo runtime -- REST/LiveView/SQL+Cypher -- as the current
language; contradicted the new README)
- v1 design set: 02-recovery, 03-data, 04-ui, 05-datalayer,
06-markdown-render, 07-ssl; runtime/database/05-go-sdk
- future-scope/ai-agents-content-management (unfinished old-runtime CMS)
- the ##ui/.htmlx LiveView frontend track (product decision to abandon):
9 plan/exploration/ui/*, plan/14-mvc-ui-implementation,
superpowers/plans/2026-08-01-ui-htmlx-live; 13d pricing-UI board row
Tree left link-clean: 46 dead links to the removed docs neutralized to plain
text or deleted as pure see-also bullets across 20 kept docs; whole-tree
link-resolving scan reports zero links to any deleted file. Removal recorded
in discarded.md; board Frontend section + project-structure tree updated.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- README.md now IS the getting-started page (moved from
docs/writeonce.md; single source, no duplication): current
woc -> .wob -> wovm toolchain, system requirements, build, hello-
world, language + stdlib, embedded database, samples, roadmap
- deletes the old README's Rust `wo` runtime pitch (cargo run, axum
REST, Postgres mirror, blog/ecommerce) — that runtime is not
advancing and no longer the project's front door
- content unchanged from the verified doc (hello-world + switch
compiled live before the prior commit)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- docs/writeonce.md — external-developer front door for the current
woc -> .wob -> wovm toolchain (NOT the stale root README's old Rust
wo runtime): what writeonce is, system requirements, how to build
the toolchain, hello-world + the two build paths, language surface,
stdlib, the embedded database, project/manifest layout, samples
- shipped-only by decision: every feature described compiles and runs
today; hello-world + switch snippet verified live before commit;
employee/log-watcher cited as the working acceptance samples
- unshipped roadmap (aggregates, HTTP service, concurrency/fibers,
cross-program attach + keypair auth, blue-green, @derive) kept in a
clearly-separated Roadmap section, plus named current limits (net
TCP-only, proc.run no timeout, no stdin/stdout, no FFI)
- note: root README.md is stale (documents the older Rust axum/REST
runtime, no mention of woc/wovm); left untouched, flagged for the
developer
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- frames a writeonce port of ~/projects/skillhost (C++ MCP host that
links libllama in-process, discovers filesystem skills, runs their
scripts confined) as the sample that drives host capabilities into
the open — the way log-watcher drove the systems stdlib
- names each gap as a candidate iteration (surveyed 2026-08-16 vs the
running compiler + skillhost source):
- Blocker A: in-process native-lib FFI (no FFI today) — fork:
FFI-as-language vs out-of-process model driver over proc/net+json
(llama-server, needs nothing new); leaning out-of-process
- Blocker B: stdio transport — no stdin/stdout builtins; port uses
a TCP socket meanwhile; io.stdin_read/stdout_write a candidate
- Blocker C: bounded/killable subprocess — proc.run has no timeout/
signal/process-group kill; smallest + most broadly useful, do 1st
- partials: recursive fs walk, exec-bit check, symlink-resolving
confinement (realpath) — one small fs-metadata iteration
- records what is already expressible (catalog via @table/9g skill-
catalog, discovery, frontmatter text-parse, config, single-thread
serve, context-gate arithmetic — no VRAM query needed)
- out of scope: in-process libllama/CUDA, VRAM introspection, exact
sampler chain / per-turn memory clear
- roadmap + board rows added
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- method: an embedded-SQL app is a grammar corpus; catalogue what it
actually uses and add only that, translating its statements 1:1 as
the acceptance (the postgres/System.Linq reference pattern applied to
a whole application)
- corpus #1 = ~/projects/skillhost (C++ MCP host, embedded SQLite skill
catalog): surveyed, entire SQL footprint is one file — 1 table, a
single-row parameterized INSERT, 4 SELECTs. 3 of 5 statements already
run on the 9b surface (insert, where name==?, order by name; PK ≈
@unique). Exactly 2 are the real gap:
- whole-query `count` (group-free; the degenerate aggregate, NOT
the parked group-by)
- correlated `not exists` subquery (skillhost's roots-of-the-tree)
- notable finding: skillhost's NOT EXISTS is naturally a `backlink`
emptiness in writeonce (children backlink + len==0), so the corpus
may be fully expressible once len(query) is confirmed — the iteration
may collapse to "confirm len(query) + add exists"; forks record this
- explicitly parks everything skillhost does NOT use (join/having/
offset/distinct/CTE/window/union/upsert/returning/json/fts/triggers)
and the full group-by; each enters only when a corpus demands it
- acceptance: docs/examples/skill-catalog mirroring skillhost's schema
+ its 5 catalog ops as writeonce translations
- roadmap + board rows added
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- FK restrict: deleting a row a non-nullable `ref` still points at traps
WO_T_FK (11), catchable. The compiler now records a `ref` field's
target class in the class-table field_class metadata; the engine
(wo_row_has_referrers) scans referencing scalar columns before a
delete. Correctness-first full scan; the backlink-index optimization
is recorded for later
- docs/examples/employee now COMPILES AND RUNS all six modes against a
WAL-durable database: seed (+@unique trap across restart), report
(per-dept aggregates + payroll), staff (unique probe + backlink +
ref nav), raise (update-through-row), drop (FK restrict), and
persistence via replay
- group-by SYNTAX parked to a future iteration (user decision): the
report mode is hand-rolled from the shipped primitives meanwhile
(same numbers). "table relations and FK" is complete
- scripts/employee-accept.sh (8 checks) + a `just employee` module;
manifest parser tolerates iteration 9c's [share]/[[share.clients]]
sections so `woc .` builds the sample on this branch
- fixtures trap/db-fk-restrict (code 11) + run/db-fk-restrict-catch;
oop-e2e 79/0, woc-test 566/0, 15 runtime suites, log-watcher 7/0,
employee-accept 8/0
- 9b story + status board updated
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- `e.field = v` where e is a table row lowers to DB_UPDATE_FIELD
(class, id, field, value) — the row's indexes maintained at the choke
point; heap-object field assignment still emits SETF unchanged
- `delete <row>` expression: DB_DELETE(class, id), yields the id so it
composes in `try delete x catch (e) nil` (restrict/trap surfaces
catchably); Delete AST node threaded through type/owner/dump/emit
- disassembly-caught bug fixed: in tail position dst == the builtin
window's first reg, so moving the id into dst clobbered the class id
— reserve dst past the window (the emit_ctor guard)
- run/db-update-delete fixture; oop-e2e up, woc-test 566/0,
log-watcher 7/0
- employee seed/list/staff/raise/drop now compile+run; only `report`
(group-by aggregation + projection record) remains
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- `order by <field> [desc]` on whole-row queries: a selection sort over
the result multi, re-reading the key per element via the range var
(DB_GET_FIELD); O(n^2), KISS, no cost planner — the result sets are
small by design
- Text order keys use a new WO_B_STR_LT builtin (content compare, reusing
the WO_B_SORT elem_cmp); scalar keys use the LT opcode. The bug this
fixes: op_lt on two Text pointers compares ADDRESSES
- `take N`: clamp to count, slice [0,N). `take` is the KwTake keyword,
not an Ident — matched as the token
- two bugs found + fixed while testing: multi-line query clauses (skip
the separating newlines) and the key-kind read (must bind the range
var BEFORE ty_of_expr of the order key, or a Text key silently uses
op_lt); Index typechecks to the container's element type (`ds[0]`)
- fixture run/db-query-order; oop-e2e 75/0, woc-test 566/0, 15 runtime
suites, log-watcher 7/0
- employee `seed`/`list`/`staff` modes now compile and run; report
(group-by+projection), raise (update), drop (delete) remain
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- `backlink C.f` field type: parsed, typed as `multi C`, and VIRTUAL —
filtered out of the stored row layout (no column, omittable in
ctor/insert), collected in clsrec.cr_backlinks
- reading a backlink (`d.staff`) lowers to DB_PROBE on the source
class's index for the backing column (backlink_target resolves the
(source class, index number); a backlink with no backing index has
no efficient read)
- `ref C` navigation (`e.dept.name`) chains: a ref value is the target
row's id, so a `Ref C` base navigates into C's fields exactly like a
table-class value, routing to DB_GET_FIELD both in typecheck and emit
- query navigation source `from s in d.staff`: emit_query evaluates the
nav expr to get its id-list instead of DB_SCAN; QNav typechecks with
the range var bound to the navigation's element class
- fixture run/db-query-relations proves both directions; oop-e2e 75/0,
woc-test 566/0, log-watcher 7/0
- still ahead for employee: order/take, group-by aggregates, projection
records, delete + update-through-row
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- captures the toCSV/reflection thread: principle 13 forbids runtime
reflection, so a generic serializer can't be a user-written function;
Rust answers with derive macros (compile-time codegen), and
json.encode is already a single hand-built instance of exactly that
- iteration generalizes json.encode's mechanism into a reusable derive
facility: @derive(Json/Csv/Eq/Hash/Show) -> the compiler generates
per-type routines from the class-table metadata it already emits,
monomorphic, no runtime type tag, no dynamic dispatch
- closes the query-result-serialization gap
(csv.encode(from e in Employee ... select e)) that has no expression
today; acceptance requires json.encode retrofitted onto the framework
with byte-identical output, and disassembly proving no reflection
- four forks: request surface (lean @derive annotation), invocation
(lean compiler-recognized encode builtins, no UFCS/methods), Eq/Hash
sharing the engine's key comparison, static applicability checking
- out of scope: full trait/typeclass system, user proc-macros, general
generics, cross-channel derive -- a CLOSED compiler-known derivable
set, the pragmatic 80% without the type-system weight
- numbered 13 to echo the principle it lives inside; roadmap + board
rows added
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Ast.Query node + parser: `from <v> in <src> where* [group..into] [order
by] [take] select <e>`, positional `from` trigger so it stays a usable
identifier; select stays grammar-owned (no DbStub conflict)
- typecheck: range var bound to the source table class; a table-class
value is its row id at runtime but TYPES as the class, so `e.field`
checks against the class fields; result is `multi <select-type>`;
group/order/take/navigation diagnosed WO-E250 not-yet (honest edge)
- emit: `from/where/select` lowers to a bytecode LOOP over DB_SCAN's
materialized id list — DB_GET_FIELD per column read, where-guards skip
the push, select projects, result is a fresh multi; no plan tree, no
SQL text (disassembly-provable)
- field access on a @table-class value routes to DB_GET_FIELD instead of
GETF (clsrec.cr_is_table + is_table_class); non-table classes unchanged
so log-watcher is unaffected
- the ASan-caught bug kept in a comment: a table-class query element is a
SCALAR id, not an OWNED pointer — tagging the result multi OWNED dropped
an id as a pointer (SEGV in wo_drop_obj)
- fixture run/db-query-scan (where-filter + select-whole + select-field);
oop-e2e 74/0, woc-test 566/0, log-watcher 7/0
- SLICE scope: group-by aggregation, order/take, and ref/backlink
navigation are the next chunk (employee report/staff/raise need them)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- DB_SCAN(64): class -> multi<Int> of every row id, materialized up
front (the 9b cursor-stability rule: the loop body point-reads, so a
row updated mid-loop cannot disturb iteration)
- DB_GET_FIELD(65): class,id,field -> the field decoded to a fresh VM
value (the out-gate copy); a table-class value IS its row id at
runtime, so this is how a compiled query reads a column, and a ref
field decodes to the target id for navigation
- DB_PROBE(66): class,index,key -> multi<Int> of ids whose first
indexed column equals key (backlink + indexed where)
- wo_val_decode_vm wrapper exposed; dispatch range 61..66, loader
arities, runner mirror updated
- 15 runtime suites green, oop-e2e 73/0
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- 9e durability/throughput/scale: the measurement backbone -- run the
employee program, restart to prove persistence, benchmark read/write
through compiled .wo, ~1M-row mixed load with throughput floor + p99
ceiling + flat RSS; the gate every optimization signs (before/after
delta required, no measured delta = not accepted)
- 9f io_uring group-commit: replace fsync-per-commit with batched
io_uring durability overlapped on shard threads; same ack-after-
durable contract, crash battery unchanged, automatic fsync fallback
on kernels without it; deliberately LAST (needs 8's threads to
overlap and 9e's baseline to beat)
- wired the existing levers into the arc: iteration 8 (thread-per-core)
= "optimize multithreading", 7b (mark-sweep) = "implement GC" --
each now gated by re-running 9e and recording the delta
- explicit sequence recorded in 9e: 9b lands -> 9e baseline -> 7b
re-bench -> 8 re-bench -> 9f re-bench
- roadmap + board rows for 9e/9f; four forks each for the specs
(load generator, absolute vs relative budgets, what "1M" means,
durable vs RAM headline; ring model, liburing vs raw, batch
boundary, fallback testing)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Task 6 note: db fixtures live in existing corpus kinds; crash battery
proven at unit level; select fixtures wait on 9b's read surface
- board row updated
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- wo_row_update_field: encode new value, unique re-check against a
shadow BEFORE any mutation (violating update leaves the row
untouched, DB_ERR_UNIQUE), index entries moved old-hash -> new-hash,
old engine value freed; proven by test_table (unique refusal keeps
the row, released key becomes insertable)
- WAL UPDATE record: full-row re-log, replay = replace (remove +
re-create same id); prefix/suffix delta recorded as later
optimization; test_wal replays insert+update to the updated state
- builtins 62 DB_UPDATE_FIELD (cid,id,field,value) and 63 DB_DELETE
(cid,id), commit-before-ack like insert, WO_T_UNIQUE/WO_T_DB/WO_T_IO
mapping; dispatch range 61..63; loader arities; runner mirror
- plan Task 5 marked superseded-in-part with the recorded deviation:
the language surface (reads, queries, row views, delete statement)
is 9b's, where the comprehension design put it -- no interim brace-
select grammar to retire later
- gates: test_table 839/0, test_wal 102/0, 15 suites, oop-e2e 73/0
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- .wob v3: class records carry an index tail (flags bit0 = unique,
col_cnt, columns) -- @table(index:[a,b]) entries plus one unique
single-column entry per @unique field; loader validates columns in
range and scalar/Text-kinded; emitter validates the declarations
(unknown column, un-indexable kind => diagnostic)
- engine: db_index hash multimap per table, built from the class
table at first touch, maintained ONLY inside wo_row_insert/
wo_row_remove; unique checks re-compare actual column values (a
hash is a hint); replay re-indexes via wo_row_raw_commit AFTER
slots are filled, so recovered tables carry their indexes
- WO_T_UNIQUE = 10; a violating insert is un-applied whole (bitmap,
hash, count, and the never-observable id reclaimed) and traps
catchably -- the employee SEED-DUP pattern
- wo_row_insert gains err_kind so db.c maps UNIQUE/OOM/other to the
right trap; test images and the runner's loader mirror speak v3
- fixtures: trap/db-unique-violation (code 10 exact) and
run/db-unique-catch (catchable dup, composite index accepts
duplicates, next id dense after a refusal)
- gates: oop-e2e 73/0, all 15 runtime suites, woc-test green,
log-watcher 7/0
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- docs/examples/employee-list: attaches to the running employee
program; modes list / report (byte-identical to A's own) /
staff <dept> / probe-write (registered read-only, insert must trap
access-denied, exit 4, A unchanged)
- the manifests ARE the design, written as a pair: A's [share] gains
listen = unix socket beside WO_DATA plus [[share.clients]] naming
B's public-key fingerprint with rights = "read"; B's
[connect.employee] carries A's ipc string, A's PINNED fingerprint,
and project = ../employee for compile-time shapes -- the connect
section's name is the code's namespace (employee.Employee)
- fingerprints are PASTE-HERE placeholders by design: keys generate
into WO_DATA at first boot (9d), tomls carry fingerprints only,
printed by --identity
- sample-first: compiles after 9/9b/9c/9d; README maps each mode to
the acceptance line it exists for; 9c/9d stories now name this
sample as their workload
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- promotes 9c's identity fork to its own iteration: program identity
is a keypair (first-boot generated into WO_DATA, 0600, printable
fingerprint); A's [share] grants name PUBLIC KEYS, B pins A's key
in [connect.a]; mutual challenge-response, fresh nonces, transcript-
hash signing (protocol tag + fingerprints + nonces + channel)
- acceptance criteria: registered-key attach carries 9c rights
unchanged; unregistered key refused pre-statement with fingerprint
logged; same-uid-wrong-key refused (uid SUPERSEDED, not
supplemented); impostor on A's socket path aborted by B's pinned-key
check; handshake replay refused; rotation = manifest change
- four forks recorded: crypto provenance (lean: vendored compact
Ed25519 as the one sanctioned vendored component), keygen home
(lean: first-boot into WO_DATA), signed-transcript layout, uid
survival (lean: keys only, peer-cred demoted to log enrichment)
- out of scope: transport encryption, CA machinery, key escrow,
root-attacker protection
- plan folds into 9c's when specced (neither ships alone); 9c fork 3
marked superseded-as-end-state; roadmap + board rows added
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- program B attaches to running program A's persistent database via an
IPC string in B's wo.toml [connect.<name>]; A registers clients by
name with read / read+write rights in its [share] manifest section;
unregistered = refused at connect, under-privileged = catchable trap
- doctrine preserved: A stays the single writer -- B's statements
execute inside A through the same choke-point row API, B never
touches A's WAL or slabs; typed statements checked by B's compiler
against A's table shapes, schema handshake at attach
- four forks recorded for the spec: channel carrier (lean: unix
socket + SO_PEERCRED), how B's compiler learns A's shapes (lean:
project reference + live handshake), grant granularity (lean:
whole-db rights, name+uid identity), blocking semantics (lean:
blocking round-trip, stop-flag rule applies)
- acceptance sketch: employee sample as A, a thin employee-report
client as B (read-only GroupBy over the wire) + audit-log writer
exercising the rights matrix
- slots after 9b (shares its typed surface), before 10 (HTTP is the
external face; this is the writeonce-native one); prior art:
04-client-api.md wire protocol + the WAL's value encoding
- roadmap + status board rows added
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>
- database/src/wal.{c,h}: framed records len|crc32|payload|mark
("WOL1" written last -- no mark, no record), typed-row payloads
walking the class-table kinds (nested records, containers, nil
encodings), little-endian like the loader
- commit order verbatim from the shipped phase-D pattern: RAM apply,
stage, ONE pwrite + ONE fdatasync for the batch, ack after -- group
commit is everything staged riding one sync
- replay decodes straight into engine-owned values (no VM at boot)
and re-enters rows through the choke-point row API, so Task 4's
indexes will rebuild for free; next_id advances past replayed ids
this shard owns (wo_row_create_raw)
- torn tail = short/CRC-fail/no-mark/zero-len: intact prefix applies,
tear dropped whole, wo_wal_open positions AT the tear so the next
commit overwrites it; CRC-valid-but-undecodable = corruption, loud
- wo_wal_check: offline oracle, no engine needed -- the crash
battery's verifier
- test_wal 90/0 ASan+UBSan incl. five crash-battery rounds (fork,
insert/commit/ack-over-pipe, SIGKILL mid-stream: zero acked-but-
missing, zero acked-but-wrong); all runtime suites green, oop-e2e
71/0; binding doc WAL section + CODE-LOGIC + plan Task 2 checked
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- database/src/table.{c,h}: per-shard per-class slabs (256 rows,
malloc'd, never moved -- row addresses stable for 9b's row views),
occupancy bitmap, LIFO slot reuse, open-addressing id hash with
tombstones (ids never 0, never reused)
- field encoding walks the same .wob class-table kinds the VM walks:
scalars raw (WO_NIL_SCALAR passes through), Texts copied to db_text,
owned objects flattened recursively to db_rec, containers
element-wise; GCREF refused at encode (the GC bulkhead, defensively)
- two one-way copy gates: insert copies VM values in, read allocates
fresh VM values out -- no VM pointer in a slab, no slab pointer in
the VM, proven by mutating originals after insert
- id discipline: per table per shard, S+1 step N; owner = (id-1) % N;
N-parametric, runs at N=1 until iteration 8, tested at N=3
- choke points: wo_row_insert/wo_row_remove carry the INDEX HOOK
sites Task 4 attaches to; nothing else mutates storage
- runtime/Makefile links database/src into every wovm + test binary
- test_table 827/0 ASan+UBSan; oop-e2e 71/0; log-watcher 7/0;
binding doc docs/plan/oop-vm/04-db-binding.md; CODE-LOGIC.md beside
the code; plan Task 1 checked off
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- 9b spec gains section 6 "Ownership, borrows, and GC across the
engine boundary": two one-way copy gates (no VM pointer enters a
row, everything a select returns is copied out), so the collector
never traces engine memory and the engine never touches refcounts
- row views are borrows WITHOUT a runtime net: rows share the VM's
field encoding but not its header, so no borrow word backs them --
the compile-time escape rule is load-bearing alone
- cursor stability settled: scans materialize their id list before
the body, row updates through the view stay legal (raise mode
updates an indexed column mid-scan and is the proving fixture),
insert/delete on a table with an open cursor is a new WO-E5xx
- GC-pause interaction recorded: collector runs between statements,
a long scan delays slices -- accepted, documented
- iteration-7b ordering constraint: GC inference must classify before
table-field validation, diagnostic names the inference reason --
noted in 7b story, iteration-9 plan constraints, 9b plan tasks
- stories 09/09b Info sections point at the analysis; 9b plan Tasks
3/5 carry the enforceable checkboxes (ASan boundary assertion)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- docs/examples/employee: Department/Employee @table classes with
@unique, [dept] and [dept, salary] indexes, ref/backlink pair;
modes seed/report/staff/raise/drop per the 9b spec section 6 —
written AHEAD of the features (sample-first, like log-watcher);
README states it does not compile on today's toolchain and links
the plans that compile toward it
- report mode is the GroupBy showcase: group-and-reduce projection
{headcount, avg, min, max} ordered by avg desc, whole-query sum
for payroll; staff proves both navigation directions + index probe;
drop proves delete restrict (trap asserted)
- engine directory decision (user, 2026-08-15): database/ is its own
top-level dir, statically linked into wovm — file structures updated
in the iteration-9 plan and the 9b plan
- gap found by writing the sample: iteration 9's subset lacks a
`delete` statement and restrict needs one — added to 9b plan Task 3
- 9b plan Task 6 notes the sample is pre-authored and authoritative
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
