Files
marqdo-cli/docs/STATUS.md
T
cflmy 89555382e0 feat(ext): extension selection + lock v4, inspect --extensions (RFC-EXT-0001 PR-04)
- marqdo.yaml.extensions accepts only ./path or @alias/path, resolves to a
  regular file inside its owning package, never grants
- lock_version 4 records plugin/kernel/provider manifest+artifact identity;
  v3 requires explicit 'marqdo lock' migration; sync/check/run compare identity
- check stays static; run re-verifies artifact bytes; pure/strict refuse
- inspect --extensions reports declared/locked identity, plugin descriptors and
  stdlib aliases with authorization 'not_granted'
- black-box extension_selection suite (13) + package-layer tests; specs/ADR-0056
- workspace gate: fmt/clippy clean, 599 passed across 67 targets
2026-11-22 00:43:18 +08:00

41 KiB
Raw Blame History

Marqdo Implementation Status

Snapshot: 2026-11-21, workspace version 0.1.0, no release tag. Current architecture overview: architecture/CURRENT.md. Authoritative CI convergence plan: NEXT_RELEASE_CI_CONVERGENCE.md (historical baseline at c9f01aa8; refresh before use). Support matrix: ../COMPATIBILITY.md.

This file records what was actually executed on the current tree. "Implemented" does not mean "verified on every platform"; the evidence column says which commands were run and where. Do not copy a running status into another plan; architecture/CURRENT.md is the single current overview. The historical M0.5 snapshot is preserved at history/STATUS-M0.5-snapshot.md and must not be read as current status.

Status is generated from executed commands and logs; a gate row without a command, log or source path is a defect, not a state.

0.0 RFC-EXT-0001 extensibility closure (current tree)

Executed on Linux x86_64, Rust 1.100.0, branch feat/extensible-ecosystem:

cargo fmt --all -- --check                             PASS
cargo clippy --workspace --all-targets -- -D warnings  PASS
cargo test --workspace --locked                        PASS
  67 test targets: 599 passed, 0 failed
cargo build --release --locked -p marqdo-cli           PASS

Implemented and tested in this tree (PR-00…PR-07; no release claim):

Area State Tests
One CallableRegistry + one binding pass; runtime dispatch by CallSiteKey; no runtime name fallback; exact plugin reconciliation IMPLEMENTED / TESTED callable_resolution (14), plugin_worker, analysis_trace, architecture_boundaries
provider.json v2 through --provider/--allow-provider; verified before spawn; sibling helper only IMPLEMENTED / TESTED (linux/x86_64) provider_cli (12); PROV-01…10/13; macOS/Windows FAIL-CLOSED
marqdo.yaml.extensions + lock v4 identity + inspect --extensions; CLI/manifest merge and conflicts IMPLEMENTED / TESTED extension_selection (13), package_env, package crate tests
Capability handler registry; fs.read real handler; audit/unavailable IMPLEMENTED / TESTED marqdo-host capability_handlers (12)
Official stdlib public aliases; print_override/AST scan/runtime branch removed IMPLEMENTED / TESTED stdlib_alias_routing (5), print/format/Trace goldens
Independent external repo + frozen-CLI suite (pure pinned Git, plugin, kernel, provider, tool) PASS against a development binary; release-binary rerun + SHA-bound manifest pending /root/work/marqdo-extension-conformance tests/run_all.py: all 7 categories PASS

Evidence: evidence/2026-11-22-ext-implementation.md and baseline evidence/ext-baseline-595279f.md. The RFC-EXT-0001 release gate additionally requires a --release frozen-binary rerun and docs/evidence/extension/<SHA>/manifest.json on the tagged commit.

0. Language migration completed (RFC-LANG-0002)

The repository now has one language contract, implemented end to end: architecture/RFC-LANG-0002-unified-semantics.md (accepted by ADR-0052), with the normative summary in ../spec/mapping.md and ../spec/expression.ebnf.

Landed on master as one linear series (the former semantics-v2-refactor branch was fast-forwarded and deleted; only master remains):

Step Commit Content
1 4983542 docs/spec/ADR/READMEs, historical banners, migration guide and inventory
2 e029853 frontend/IR/runtime implementation, sem.* diagnostics, migrated stdlib + digest, examples, fixtures and Rust-embedded sources
3 ffd1cc9 expression-owned payload statement grammar, extra negative/CLI coverage
4 bbeb1de status/compatibility/README/migration docs flipped to the implemented state + green evidence
5 b38b769 function-wide params(...) position (across heading segments) + two negatives
6 361a6cc removed the dead Stmt::Call/CallSite/CallTarget IR
7 d5002ac stale pre-RFC role docs and review-aid fixes
8 9fb97bc signatures recognized only from Exec payloads (Knowledge sections inert)
9 a3f9770 unmatched end diagnostics point at the end payload
10 0bbe0f6 Setext headings never declare a mode; dead helper removed
11 b3429cb diagnostics sorted by source position then code
12 9c1723d coverage for the removed syntax-selector front-matter probe

The breaking behaviour is:

  • H1–H6 headings declare @exec/@knowledge; unmarked headings inherit the nearest ancestor and the document root is Knowledge;
  • only an Exec H2 is callable; an H2 Knowledge unit and H3–H6 sections are not functions, and a payload with no owning Exec H2 is sem.exec_without_function;
  • Strong, Emph and inline Code are interchangeable one-statement carriers; nested carriers and empty payloads are hard errors;
  • if/else/while/end/return/params come from the payload and lower to the existing Branch/Loop/Return/Assign/Eval IR; lists, block quotes and paragraphs are presentation containers;
  • control frames cannot cross a heading segment;
  • params(...) is legal only as the first direct Exec H2 payload across the whole function (a later ### @exec segment cannot add a signature after earlier code ran); without it the existing free-variable inference applies and inferred signatures keep the named-arguments-only rule;
  • old syntax (Strong execute / Emph return / list branch/loop / quote call) is not executable, and marqdo_semantics/syntax front-matter selectors are rejected with sem.unsupported_version_directive.

Verification at 9c1723d (Linux x86_64, Rust 1.100.0)

cargo fmt --all -- --check                            PASS
cargo clippy --workspace --all-targets -- -D warnings PASS
cargo test --workspace --locked --no-fail-fast         PASS
  507 passed, 0 failed, 2 ignored (62 green test targets)
python3 tools/pack-stdlib.py                           PASS (digest verified)

Full evidence: evidence/2026-11-21-semantics-v2-green.md. The intermediate "text migrated, implementation pending" record is kept as evidence/2026-11-21-semantic-docs-migration.md and is superseded by the green gate.

Not claimed: the remote G3 security artifact is still missing. The Linux/Windows/macOS native matrix is green, G4 is measured on the designated Woodpecker agent, and the G3 mechanism is PASS in a local disposable VM, but the RFC does not promote a local VM run to a remote gate. See §0.1, §4 and ../COMPATIBILITY.md. The release decision remains DO NOT RELEASE.

0.1 CI/security convergence (2026-11-21)

The RFC-CI-0001 series fixes the release-blocking CI defects that the 32d8fe1 GitHub run exposed. Commits on master:

Step Commit Content
1 f4a404b platform-neutral StrictChildState/inspect_strict_child; macOS/Windows no longer reference Linux-only functions (E0425)
2 f817cce strict capture/routing: one JSON document on every --json path, stable failure codes, ADR-0053, deterministic routing tests
3 c9f509e architecture guard against direct marker access; Windows cross-target check
4 990e984 G-C lane writes SHA-bound strict-evidence.json; workflow binds checkout/archive to the tested SHA and uploads evidence
5 d2267e9 correctness-verified marqdo.perf.v2 sampler and fixed-node workflow
6 0d7e11b SHA-bound release-evidence aggregator, docs drift cleanup, consistent windows-latest/macos-latest statement
7 512b9f5 evidence-generation failure fails the lane; workflow archive/verification fixes; G-C re-run
8 beb68a9 G-C evidence/status update and the cross-platform JSON smoke exit-code fix
9 3a29fce Windows PATH lookup applies PATHEXT; ANSI-stripped native failure annotations
10 b898c41 macOS parser-nesting test stack; portable process product fixture (CLI --version instead of /bin/echo)
11 4265231 CLI runs on an explicit 8 MiB thread (Windows 1 MiB overflow); macOS temp-path canonicalization; G3 adversarial tests opt-in
12 a24161c Windows foreign temp paths stay non-verbatim (Node lstat rejection)
13 5fcc111 reject rooted Windows dependency paths (has_root, not is_absolute)

Executed evidence:

Lane Result Evidence
Linux fmt/clippy/test/build (Woodpecker) PASS pipelines #9–#12, #14, #16, #17, #21–#25 on the series; #13/#15/#19 were superseded/cancelled by the next push, not failures
Linux + Windows + macOS native matrix (GitHub) PASS run 30 on 5fcc111: fmt + clippy + test (Linux x86_64), runtime contract (windows-latest), runtime contract (macos-latest), and schema consumers all PASS
G3 strict isolation PASS (local disposable VM) / remote UNVERIFIED 5fcc111, dev uid 1000, QEMU/TCG, 7/7 + strict JSON/inert-@knowledge; evidence/2026-11-21-strict-gc-5fcc111.md
G4 performance MEASURED / WARN (provisional) Woodpecker agent 1, 1199c88: 5 warmups + 20 samples, 9/9 VERIFIED, CPU 0 pinned; two runs -10.9%..+7.1% p50; shared powersave host caveat; evidence/2026-11-21-perf-woodpecker-agent-1.md

Release decision: DO NOT RELEASE. Linux, Windows and macOS native jobs are green on the same SHA, the strict isolation mechanism PASSes locally as a normal user, and G4 is MEASURED/WARN on the designated Woodpecker agent. The remaining release gate is the remote G3 artifact whose tested_sha matches; the RFC explicitly requires it (and the local VM PASS is not promoted to a remote claim). The perf gate is provisional because the agent is a shared powersave host; see performance.md.

The manifest contract is in evidence/ci/README.md; tools/ci/release_evidence.py reports DO_NOT_RELEASE for any missing/failed/SHA-mismatched gate.

1. Remote baseline and current evidence

The local language baseline for this snapshot is 9c1723d (2026-11-21, semantics v2 implementation plus follow-up fixes). The later documentation and test migration is recorded by its own commits; re-run the gate before making a release claim.

The old GitHub master c9f01aa8fb127a2dcc125103ef794832984774ad and public CI run 44692692318 are historical (2026-11-20), not current, and that run was red:

  • Linux fmt + clippy + test: FAIL — strict_mode_enforces_isolation_or_refuses_closed hit bwrap: Creating new namespace failed: Resource temporarily unavailable.
  • macOS / Windows cargo test --workspace: FAIL — E0433 on linux::probe() under cfg!.
  • Schema instance validation + consumers: PASS.
  • Release candidate artifact: SKIPPED (upstream failure; tag/dispatch only).

The remote SHA must be refreshed from CI after a push; do not read c9f01aa8 as the current tree.

This checkout's private remote is https://git.cflmy.cn/cflmy/marqdo-cli with Woodpecker CI at https://ci.cflmy.cn (.woodpecker/main.yml). The Linux Docker pipeline runs fmt/clippy/test/release-build; the docs-consistency gate runs in cargo test --workspace. The GitHub-hosted native matrix is now green at 5fcc111 (run 30). A CI-attached strict VM runner is still unavailable; the G-C evidence below remains local VM evidence.

Convergence code/security SHA: 8af72620f12e2494a59830a23abb393768b610e9 (the following docs commit only adds this report and release material; code is byte-identical).

Local static and test gate (Ubuntu 26.04, Rust 1.100.0)

cargo fmt --all -- --check                            PASS
cargo clippy --workspace --all-targets -- -D warnings PASS
cargo test --workspace --locked                       PASS
  491 passed, 0 failed, 2 ignored; includes 7/7 Linux security tests

The documentation and test migration is local work; re-run this block after it lands and update the count from the log rather than copying the number. docs_consistency, architecture_boundaries and stdlib_product run inside this workspace suite.

G-C security lane (disposable Ubuntu 26.04 VM, ordinary user dev)

Source: git archive 70d916d extracted in the VM (not the dev container). Host: QEMU/TCG, kernel 7.0.0-43-generic, systemd-detect-virt --container = none, bubblewrap 0.11.1 with bwrap-userns-restrict, uid 1000.

./tests/security/run-lane.sh
  adversarial helper matrix: 7 passed, 0 failed, 0 ignored
  positive CLI contract: PASS
  external-execution surfaces refused: PASS
  forged active marker rejected: PASS
  G-C security lane: PASS
cargo test --workspace --locked (same archive): 47 ok, 0 failed

Logs: evidence/2026-11-20-strict-resource-cm-a11.log and evidence/2026-11-20-strict-gc-8af7262.md. This is still a local VM lane, not a self-hosted GitHub Actions security URL.

2. Work package status

Historical table note. Most rows in this section record work executed before the RFC-LANG-0002 syntax migration. Language-source rows (fixtures, control mapping, semantic goldens) were re-verified after the rewrite at 9c1723d (§0); platform/security rows keep their original evidence dates and are not refreshed by the language migration.

# Work package State Evidence
0 Windows AST line endings Windows CI PASS (5fcc111) .gitattributes, CRLF fixture
1 macOS native plugin fixtures macOS CI PASS (5fcc111) host-target manifest + dylib flags
2 Cross-platform strict cfg Native CI PASS (5fcc111) #[cfg] dispatch + non-Linux regression test
3 Strict probe→launch race Linux CI PASS (5fcc111) + deterministic seam fail-closed CLI contract + probe-OK/launch-EAGAIN seam test
4 Parallel security matrix isolation Fixed unique paths + helper OnceLock; 7/7 in parallel workspace run
5 Markdown/power/control-flow corpus Green on semantics v2 160 AST + 73 semantic + 49 compile-negative + 2 warning + 21 runtime-negative fixtures; conformance_ast/semantics/negative pass
6 Payload/control source spans Green on semantics v2 UTF-8 spans, control-expression offsets and heading boundaries covered by fixtures and CLI tests; old condition headers were removed
7 Independent Tool consumers Done locally sample-view/sample-knowledge/sample-proof E2E test
8 lock v4 reproducibility + extension identities Implemented; tamper/migration tests added package_env, extension_selection PASS
9 Analysis/Trace/Tool/ABI/Foreign Implemented; supported-matrix bounded protocol tests, plugin/worker/foreign tests
10 Linux strict L1 Local VM PASS at 5fcc111; remote G3 runner still missing 7/7 + positive CLI + strict JSON/inert-@knowledge; strict-gc-5fcc111
11 Linux strict L2 NOT IMPLEMENTED / UNSUPPORTED strict rejects Worker/Foreign/native ABI
12 Windows Native/Worker ABI UNSUPPORTED no Windows ABI artifact/claim
13 RC artifact Linux candidate only; no release clean-install + checksums; macOS not built here
14 Release approval NOT STARTED requires owner approval

3. Verified capabilities

  • Frozen Markdown profile marqdo.md.v1 with 160 AST goldens and explicit CRLF byte semantics (unchanged by the semantic rewrite).
  • Native Rust interpreter with the unified semantics: 73 semantic fixtures, 49 compile-error fixtures, 2 warning fixtures and 21 runtime-error fixtures all green at 9c1723d; AST profile unchanged (160 goldens).
  • Native ** power is pinned end-to-end: right associativity, unary-minus precedence, parentheses, float negative exponents, branch/loop conditions, Strong+CodeSpan payloads, and inert prose/code fences.
  • lock v4 full transitive package-instance graph plus extension identity records, verified Git cache, strict sync --locked, atomic .marqdo/env materialization; run/check never fetch or rewrite the lock. Locked path content tamper is rejected without rewriting the lock or environment; lock v3 requires an explicit marqdo lock migration.
  • Analysis v2 (logical identity, resources, byte digest) and Trace v2 (module ownership, bounds, truncation, stale detection) with independent Python consumers and JSON Schema validation.
  • Tool Protocol v1 with arbitrary command names (no CLI hard-coding), SemVer/protocol negotiation, whole-bundle digest verification, atomic install/remove, and three independent Analysis/Trace consumers.
  • Plugin ABI v1: native + worker transports, capability broker executing fs.read through ProjectFs, per-invocation budgets, host-owned handle identity, crash/timeout isolation. Not an OS sandbox.
  • Foreign kernels: one-shot Python/Node, typed wire, timeout/output caps, provider/environment digests, environment-bound provider, interpreter version verification.
  • Linux strict L1: Bubblewrap namespace supervisor, runtime probe, fail-closed launch, child self-verification, 7/7 adversarial matrix.
  • Addressable memory (RFC-CM P1, ADR-0022): opaque Memory values shared by handle identity, region isolation, zero-filled 64 KiB page growth, bounded byte read/write, memory.new/size/grow, m[i] = byte, run and per-region budgets, host/wire rejection, Analysis v2 target projection.
  • Recoverable Result convention (RFC-CM CM-A05, ADR-0024): success is {ok:true,value:...}, failure is {ok:false,code,message[,details]} as an ordinary Map; expected conditions return it, invariant violations stay fatal. A .mq.md first-fit/free-list allocator reuses blocks, splits free blocks, coalesces address-adjacent blocks and returns alloc.no_space without aborting.
  • CM-A07 Compat-A (ADR-0027): explicit .mq.md io.写值 chain (format.显示 → UTF-8 encode → Memory → resource.write(stdout)) matches builtin 打印 for the covered value set; builtin remains default and no alias is routed.
  • Compat-C print completion (ADR-0031): HostOp::Print/op_print removed; the print builtin only routes to bundled io.写值 or fails with kernel.unavailable; trace has resource effects only.
  • Compat-B print routing (ADR-0030): Program.print_override routes 打印/print to the bundled io.写值 when the program uses it; stdlib modules are injected lazily; resource operations are trace effect events and HostOp::Print no longer runs on that path. Strict permits only stdio/stdout; stdin/stderr/file/network resources remain refused.
  • Stdio streams and console productization (ADR-0040): stdio/stdin is read-only over one shared run cursor; stdio/stdout/stdio/stderr are write-only with separate host buffers and output-budget accounting; the CLI copies non-terminal process stdin at composition and embeds program stderr as a separate JSON string (or writes it to CLI stderr). stdlib/io/stream.mq.md checks short writes and close errors; stdlib/io/console.mq.md implements line input and print through the resource ABI; stdlib/text/utf8.mq.md provides a validating UTF-8 decoder.
  • Full-value stdlib formatting (ADR-0029): FunctionRef module/name and PluginHandle plugin/export are read-only observation members (host id private); stdlib/format.mq.md reproduces builtin 打印/repr for all 15 value types, including Formula/Code/FunctionRef and Code UTF-8 byte length.
  • C ABI provider worker binding (CM-A09, ADR-0038): the separate marqdo-provider-worker helper loads a trusted C ABI library and serves the framed-JSON transport; process isolation but not a sandbox.
  • Provider package manifest v1 (CM-A09, ADR-0037): a local marqdo.provider manifest resolves a trusted C ABI library and registers it; no registry, signatures or CLI install yet.
  • C ABI ResourceProvider v1 (CM-A09, ADR-0036): trusted shared libraries expose open/read/write/close through the frozen header; version/load failures are fail-closed; project-mode + name grant only. Plugin-Worker binding remains open.
  • Worker resource provider transport (CM-A09, ADR-0035): host-managed helper process with framed JSON, ping/open/read/write/close; project-mode + name grant; trusted, not a sandbox; C ABI binding still open.
  • HTTP framing and client (P5/C10): stdlib/net/http.mq.md parses status, headers and Transfer-Encoding: chunked, builds requests, and an end-to-end test parses a loopback TCP response; fixture 71 pins the same logic. No host HTTP/TLS implementation.
  • Blocking TCP endpoint (P5 first slice, ADR-0034): project-mode resource.open("tcp", ip:port, ["read","write"]) with exact IP grant and timeouts; no DNS/TLS/UDP/server sockets; strict/pure refuse. CLI --allow-net.
  • In-process ResourceProvider v1 (CM-A09 first slice, ADR-0033): trusted providers register additional resource kinds without parser/runtime/HostOp changes; project-mode + name grant required; pure/strict refuse provider kinds; C/Worker transport remains open.
  • P6 baseline profile: tools/perf/bench.py measures 7 .mq.md workloads with wall time and peak RSS; no optimizer/backend was added and P6 is not claimed complete.
  • Cross-storage table equivalence (P4, ADR-0032): fixture 72 compares T0 row Table results with a .mq.md columnar List/Map store for sort/filter.
  • Table stdlib algorithms (CM-A08, ADR-0032): stdlib/data/table.mq.md provides select/filter/sort/append/set/join/group/转GFM over the logical Value::Table; inputs are never mutated; GFM AST laziness and physical T0 storage are unchanged. std:fs/file provides bounded 读文本/写文本/复制文件 over the existing resource.* ABI plus 存在/元数据/列出目录/创建目录/ 删除/重命名/追加文本/读取片段/遍历 over resource.request (ADR-0041/0045); data/json/data/csv provide recursive JSON (128-depth bound, multi-value) and quoted single/multi-row CSV product parsers; path and a regex subset are pure .mq.md; datetime is pure UTC ISO-8601 and clock.现在 reads clock.now in project mode (ADR-0042); env.读取 reads only exact --allow-env NAME grants in project mode (ADR-0043); process.运行 runs an exact --allow-process PROGRAM grant with 5 s timeout and 1 MiB caps (ADR-0044, not an OS sandbox); random.字节 reads explicitly granted random.bytes entropy (ADR-0046). File watch, binary offset reads, streaming incremental JSON/CSV streaming, pure-mode regex groups/alternation and regex look-around/backreferences remain open; project mode has host-backed groups and captures (ADR-0050).
  • Bundled standard library (CM-A10, ADR-0028): stdlib/ ships format/text/io/fs/data/path/regex/datetime/clock/env/process/random/compress/net/http; explicit std: imports resolve only inside the bundle; a SHA-256 digest over *.mq.md is verified before loading; Analysis records project.stdlib; a frozen-binary test asserts the CLI hash is unchanged after running a std:io project and that tampering is rejected. 打印/print is routed to the bundled stdlib (Compat-C, ADR-0030/0031); there is no legacy host formatter. tools/pack-stdlib.py additionally emits a byte-deterministic offline archive + descriptor as an extraction-ready artifact (evidence). The stdlib intentionally stays in this repository for now (ADR-0039); extraction is a future option when size/release cadence requires it, not a pending gate.
  • Resource streams (RFC-CM CM-A06, ADR-0025/0026/0040): opaque host-issued Resource handles with resource.open/read/write/close; bytes move only through Memory; stdin is read-only over one shared run cursor; stdout and stderr are separate capability/output-budget checked buffers; project-file read/write is path/grant checked; file writes buffer until close and never flush on drop; a .mq.md chunked copy loop reproduces a file without a Rust copy_file; resource.request adds non-streaming fs.exists/ fs.metadata/fs.list/fs.mkdir/fs.remove/fs.rename (project mode, grant-checked, recoverable fs Result); strict permits only the stdio stdout console channel and refuses stdin/stderr/file/TCP/provider resources (CM-A11 lane PASS). Semantic fixture 66 and CLI/host tests.
  • Test and documentation migration (review RFC-CM P0/P2): crates/cli/tests/docs_consistency.rs pins one current overview, the status/index links and SHA-pinned evidence; crates/cli/tests/architecture_boundaries.rs pins the workspace graph, marqdo-runtime -> marqdo-host, the bundled stdlib digest and the stdio/file/TCP endpoints; crates/cli/tests/stdlib_product.rs runs every public stdlib module (clock/compress/console/csv/datetime/demo/env/format/fs/http/io/json/path/process/random/regex/stream/table/text/utf8) through std: imports over stdlib/tests/*.mq.md. tools/pack-stdlib.py now sorts .mq.md by the same relative path key as the CLI. Source/test baseline 0d40de7; later docs-only commits keep code/test bytes identical. Evidence: docs-and-test-migration.
  • Module-boundary refactor (review RFC-CM P1): CLI split into commands/{sandbox,run,check,inspect,init,package,tool}.rs; host split into host/{resource_ops,providers,effects,dispatch}.rs plus process.rs. commands.rs keeps shared project/runtime helpers and host.rs keeps the Host trait methods and capability checks; this is a behavior-preserving split, not a new crate or backend.
  • Minimal structure observation/construction (RFC-CM CM-A04, ADR-0023): value-semantics list.append/concat, map.keys/has/get/set/remove, Unicode scalar text.codepoints/text.from_codepoint, and non-negative table.columns/row/cell. Pure builtins; no mutation, no new syntax.

4. Open blockers and limits

  1. Native matrix is green; remote G3/G4 are not. GitHub run 30 on 5fcc111 PASSes Linux, Windows and macOS, and Woodpecker #25 is green. The remaining release blockers are a remote G3 security artifact and a MEASURED G4 report on the fixed node.
  2. G-C is VM-verified but not CI-signed. The Woodpecker Docker pipeline is not an OS-isolation lane; no ephemeral self-hosted security runner has run security-linux.yml for this SHA.
  3. Windows Native/Worker ABI is explicitly UNSUPPORTED (no Windows cc fixture lane, Unix-gated worker tests).
  4. fs.write, net.request, process, clock, random are not executed broker services; unsupported requests must return capability.unavailable.
  5. Resource streams cover project-file read/write (including std:fs/file read/write/copy/append/read_at/walk), non-streaming resource.request fs.* metadata/directory operations, stdout, compatible print, trusted in-process providers and TCP/UDP/listener/TLS endpoints with DNS resolution after an exact host grant. File watch, device and provider packaging remain open. Strict permits only stdio and refuses the rest (CM-A11 lane PASS).
  6. uv/npm/venv provisioning is not implemented. Only a pre-provisioned, digest-pinned directory with a regular-file executable is supported.
  7. No tagged Windows/macOS RC archive exists. The native build/test lanes PASS at 5fcc111; the release-candidate job is tag/dispatch gated and has not produced those archives.
  8. No release tag/binary publication. The release decision is DO NOT RELEASE until the remote G3 artifact and the MEASURED G4 report exist.

5. RFC-CM compute-machine status

The two 2026-11-20 architecture RFCs are design input, not implemented behavior. This table records only what is in this tree now.

RFC phase State Evidence / limit
P0 freeze (ADRs, spec deltas, baseline) P1 + partial P2 slice done ADR-0022 freezes CM-A01/CM-A02/CM-A03; ADR-0023 freezes CM-A04 (minimal List/Map/Text/Table observation and construction); ADR-0024 freezes CM-A05 (recoverable Result map convention); spec/expression.ebnf, spec/evaluation.md, spec/value-wire.md, spec/analysis-v2.md, spec/security.md updated; existing AST/value goldens unchanged
P1 Memory + place assignment Implemented locally Value::Memory, run-scoped MemoryStore (64 KiB pages, 64 MiB total / 16 MiB region), memory.new/size/grow, m[i] read, m[i] = byte; semantic fixtures 45–47, runtime negatives 13–18, examples/memory-buffer.mq.md; evidence evidence/2026-11-20-memory-p1.md
P2 Marqdo-written allocator/data structures + minimal structure observation PASS (local pure lane); stdlib remains in-repo by design .mq.md-only evidence: bump allocator with u32 codec, linked list, failed-allocation atomicity (48–50); CM-A04 structure primitives and a table filter (51–54, ADR-0023); a first-fit free-list allocator with block reuse, splitting, address-adjacent coalescing and Result maps (55, ADR-0024), a Memory-backed binary tree with recursive insert/in-order traversal (56), a .mq.md CSV row splitter (57), an open-addressing hash table with linear probing, update and hash.full (58), a .mq.md UTF-8 encoder over Memory (59), a validating UTF-8 decoder rejecting truncated, bad-continuation and surrogate/overlong input (60), a Memory-backed graph adjacency list (61), a recursive JSON parser for nested arrays/objects, strings with escapes, integers, booleans, null and error Results (62), a Memory-backed stack bytecode VM with PUSH/ADD/MUL/JMP/JZ/HALT (63), big-endian 16/32/64-bit plus LEB128 integer codecs (64), a .mq.md table sort rebuilding a new table from table.row/columns/cell (65), a table join (67), a quoted/escaped CSV row parser (68), table group (69), an Int matrix multiply (70), and a sparse-matrix multiply (73); the first-fit allocator also detects double free (alloc.double_free). Evidence: memory-p2-closure, structure-api-p2, result-and-free-list-p2. Local gate PASS: crates/cli/tests/p2_closure.rs runs 21 C01–C06/C11/C12 fixtures plus the stdlib table self-test in --mode pure (no foreign/plugin/import grants), scans sources for foreign/plugin/import shortcuts and compares value/stdout to goldens; evidence p2-closure-gate. The bundled stdlib/ tree, std: imports, digest verification, Analysis identity and frozen-CLI closure test are implemented (ADR-0028). The stdlib intentionally remains in-repo (ADR-0039); a separate repository is a future extraction option, not a gate. Linux/Windows/macOS native CI is green at 5fcc111; a remote G3 security runner and a fixed G4 performance node are still pending.
P3 Resource/Stream ABI PARTIAL ADR-0025/0026/0040 CM-A06 slices: opaque Resource, resource.open/read/write/close, stdin read, separate stdout/stderr write, granted project-file read and buffered project-file write in project, EOF/closed/handle errors, wire/host rejection; strict permits only stdio/stdout and refuses stdin/stderr/file/network resources. stdlib/io/stream.mq.md, stdlib/io/console.mq.md and stdlib/text/utf8.mq.md are product-level .mq.md APIs; fixture 66 + CLI/host tests. CM-A07 Compat-C is implemented: when a program calls 打印/print, the loader injects the bundled stdlib/io.mq.md and the runtime routes to io.写值 (all 15 value types, UTF-8 via Memory, resource.write(stdout)); HostOp::Print and the legacy Value::display() print path are removed, and a missing bundle fails with kernel.unavailable. Strict permits only the existing stdio console channel. Providers are registered through ADR-0033 (in-process) and ADR-0035 (worker helper). Still missing: directories, C ABI provider binding, provider packages, strict provider support.
P4 Table logical/physical migration PASS (local gate); T1/T2/T3 future ADR-0032/CM-A08: Value::Table semantics and GFM AST laziness unchanged; stdlib/data/table.mq.md provides select/filter/sort/append/set/join/group/转GFM as value-returning .mq.md functions with input immutability; zero Rust table algorithms were added. Cross-storage equivalence is pinned by fixture 72 (row Table vs .mq.md columnar store, identical headers/rows/cells). The optional T1/T2/T3 physical-storage/column-store migration was not required by the gate and is not implemented.
P5 third-party Provider / TCP / HTTP PASS (local gate); external network/CI and plugin-Worker resource binding open C10: stdlib/net/http.mq.md + fixture 71 parse HTTP/1.1 status/headers/chunked bodies and build requests; http_over_tcp.rs runs a loopback HTTP exchange wholly in .mq.md over resource.open("tcp",...). ADR-0034 IP-literal blocking TCP endpoint (exact IP grant, project-only, 5s timeouts) and ADR-0033 in-process ResourceProvider v1 (registration, protocol/kind checks, name grant, mock read/write provider) ADR-0035 host-managed worker provider transport (framed JSON helper, ping/open/read/write/close, trusted-only) and ADR-0036 trusted C ABI provider (shared-library symbols, version/load checks) ADR-0037 local provider-package manifest installation and ADR-0038 C ABI provider worker binding (separate trusted helper process) are tested. External DNS/TLS/UDP, host HTTP, provider packages/signatures and the plugin-Worker resource binding remain out of scope.
P6 performance / native backends BASELINE ONLY reproducible profile harness tools/perf/bench.py and p6-profile for 7 workloads; no bytecode/JIT/columnar numeric backend or equivalence/optimization gate yet

The P1 memory domain is pure computation: it needs no grant and is available in pure, project and strict (bounded by the memory budgets). It adds no HostOp, no Markdown role and no CLI command. Memory handles cannot cross the wire or a host/plugin/foreign boundary. Block-level safety of a user-written allocator is explicitly not claimed.

6. Evidence files

7. Reproduce

export PATH="$HOME/.cargo/bin:$PATH"
# On hosts with a read-only $HOME, point CARGO_HOME at a writable directory.
cargo fmt --all -- --check
cargo clippy --workspace --all-targets -- -D warnings
cargo test --workspace --locked
cargo build --release --locked -p marqdo-cli

Consumer evidence:

./target/debug/marqdo check examples/multi-module/index.mq.md --refs --emit /tmp/a.json
./target/debug/marqdo run examples/multi-module/index.mq.md --trace-file /tmp/t.json
python3 tests/consumers/analysis_consumer.py /tmp/a.json
python3 tests/consumers/trace_consumer.py /tmp/a.json /tmp/t.json

G-C (ordinary user, disposable VM only; bind the source archive and write the evidence document):

export MARQDO_TESTED_SHA="$(git rev-parse HEAD)"
export MARQDO_SOURCE_ARCHIVE="/path/to/marqdo-${MARQDO_TESTED_SHA}.tar"
export MARQDO_SOURCE_ARCHIVE_SHA256="sha256:<host-computed digest>"
export MARQDO_STRICT_EVIDENCE="/tmp/evidence-${MARQDO_TESTED_SHA}/strict-evidence.json"
export MARQDO_REQUIRE_STRICT_SECURITY=1
./tests/security/run-lane.sh

A root-only PASS, a container PASS or a missing/shorter archive is not evidence. See security-vm.md.