Phase 22.1 WIP: SSOT resolver + TLV infrastructure + Hako MIR builder setup

Setup infrastructure for Phase 22.1 (TLV C shim & Resolver SSOT):

Core changes:
- Add nyash_tlv, nyash_c_core, nyash_kernel_min_c crates (opt-in)
- Implement SSOT resolver bridge (src/using/ssot_bridge.rs)
- Add HAKO_USING_SSOT=1 / HAKO_USING_SSOT_HAKO=1 env support
- Add HAKO_TLV_SHIM=1 infrastructure (requires --features tlv-shim)

MIR builder improvements:
- Fix using/alias consistency in Hako MIR builder
- Add hako.mir.builder.internal.{prog_scan,pattern_util} to nyash.toml
- Normalize LLVM extern calls: nyash.console.* → nyash_console_*

Smoke tests:
- Add phase2211 tests (using_ssot_hako_parity_canary_vm.sh)
- Add phase2220, phase2230, phase2231 test structure
- Add phase2100 S3 backend selector tests
- Improve test_runner.sh with quiet/timeout controls

Documentation:
- Add docs/ENV_VARS.md (Phase 22.1 env vars reference)
- Add docs/development/runtime/C_CORE_ABI.md
- Update de-rust-roadmap.md with Phase 22.x details

Tools:
- Add tools/hakorune_emit_mir.sh (Hako-first MIR emission wrapper)
- Add tools/tlv_roundtrip_smoke.sh placeholder
- Improve ny_mir_builder.sh with better backend selection

Known issues (to be fixed):
- Parser infinite loop in static method parameter parsing
- Stage-B output contamination with "RC: 0" (needs NYASH_JSON_ONLY=1)
- phase2211/using_ssot_hako_parity_canary_vm.sh fork bomb (needs recursion guard)

Next steps: Fix parser infinite loop + Stage-B quiet mode for green tests
This commit is contained in:
nyash-codex
2025-11-09 15:11:18 +09:00
parent 5d2cd5bad0
commit 981ddd890c
62 changed files with 1981 additions and 103 deletions

View File

@ -0,0 +1,37 @@
# C Core ABI (Design Stage) — Phase 22.2
Status: design-stage shim; defaults OFF; behavior unchanged.
Purpose
- Define a minimal, stable C ABI boundary to enable future replacement of selected Rust runtime paths when Hakorune is compiled to native EXE via LLVM.
- Keep Rust path authoritative while stubbing C calls behind env+feature gates; ON/OFF parity must hold.
Conventions
- Encoding: UTF8, null-terminated (const char*). No ownership transfer.
- Return: `int` (0 = success; negative values reserved for future detailed errors).
- Threading: functions must be reentrant; stateful access goes via (type_id, instance_id).
Functions (initial)
- `int ny_core_probe_invoke(const char* target, const char* method, int32_t argc)`
- No-op probe for diagnostics; safe to call for any target/method pair.
- `int ny_core_map_set(int32_t type_id, uint32_t instance_id, const char* key, const char* val)`
- Design stub for MapBox.set. Current implementation is no-op; Rust path performs the actual mutation.
Gates & Features
- Build: `cargo build --release -p nyash-rust --features c-core`
- Env:
- `HAKO_C_CORE_ENABLE=1` — enable c-core probe routing
- `HAKO_C_CORE_TARGETS=MapBox.set,ArrayBox.push` — limit targets (default: MapBox.set)
- Tags: `[c-core:invoke:<Box>.<method>]`
Call Sites (Rust)
- PluginLoaderV2 (enabled): `src/runtime/plugin_loader_v2/enabled/ffi_bridge.rs`
- When gated ON + targeted, call C shim then continue with the original Rust path (parity preserved).
Validation
- Parity canaries compare ON/OFF outputs (and rc) for MapBox.set; later for ArrayBox.push/get/size.
- Failure/unavailable paths must fall back immediately to the Rust path.
Roadmap
- Expand ArrayBox (push → get → size) with the same staged approach.
- Formalize error codes and minimal state API only after parity is stable.

View File

@ -10,41 +10,56 @@ Purpose: reduce Rust surface (nonplugin) while keeping correctness and revers
- Acceptance: `cargo build --release` (default features) remains green; quick smokes green.
- Revert: `git mv archive/rust-llvm-backend/llvm src/backend/`.
### Phase 1 — Quick Wins (12 months)
- hv1_inline → Hako parity (already functionally covered; keep as perf path).
- MIR interpreter parity check; route Primary to Hako VM.
- TLV codec → C shim (FFI) with thin Rust wrapper.
- LLVM wrapper → Hako/C harness (Python stays primary until Hako IR is ready).
### Phase 1 — Parser/MIR Hakofirst22.0
- Make Hako the primary for Parser/MIR; Rust builder becomes fallback.
- Verify quick canaries green under registry; keep hv1 inline parity.
### Phase 2 — TLV C shim & Resolver SSOT22.1
- TLV codec to C shim (+ Rust FFI); Resolver/Using SSOT in Hako shared by runner/analyzer.
Deliverables & Tests
- hv1_inline: parity canaries = phase2037/flow + phase2170/state; add `HAKO_VERIFY_DISABLE_INLINE` opt-out (optional).
- TLV C shim: roundtrip tests (encode→decode) on representative payloads; FFI error mapping spec.
- MIR interpreter: keep as diagnostic path; parity sample set (const/binop/compare/branch/jump/ret/phi/mir_call minimal).
- LLVM wrapper: unify CLI entry; ensure `NYASH_LLVM_USE_HARNESS=1` path stays green.
- 22.0: registry builder default ON; hv1直列 green; Core/Interpreter is diagnostic.
- 22.1: TLV roundtrip smokes; Using SSOT parity between runner/analyzer.
### Phase 2 — Core Thinning (24 months)
- Plugin loader thin C wrapper (dlopen/dlsym), unify host ABI.
- Basic boxes (Array/Map/File) small C core (handlebased), keep Rust shim.
- Resolver/Using: SSOT in Hako; runner uses shared policy.
### Phase 3 — Core Thinning I22.2, 24 months
- Plugin loader thin C wrapper (dlopen/dlsym) and basic boxes C core; Rust shim remains.
SSOT for Using/Resolver (summary)
- Resolution order: modules (nyash.toml) → relative path inference → not found (warn) with verbose details.
- Analyzer/HakoCheck follows the same order; runner shares policy helpers. No pathliteral using in strict profiles.
### Phase 3 — Longhaul (36 months)
### Phase 4 — Longhaul22.3, 36 months
- Python llvmlite → Hako IR builder + C ABI.
- Parser/MIR builder fully Hakofirst; Rust becomes fallback.
- NyKernel minimal C runtime (BoxCall dispatcher + collections + file).
### Phase 21.10 — LLVM Line Unification (SSOT + crate probe)
- SSOT builder (`tools/ny_mir_builder.sh`) selects backend by env; crate path optin.
- Add crate S3 canaries (ternary/map/print); defaults unchanged.
### Phase 21.11 — Flip default to crate (ny-llvmc)
- Make crate default when available; llvmlite becomes optin.
- S3 reps run via crate in quick; legacy remains available.
### Phase 21.12 — Hako Native LLVM Builder (bootstrap)
- Experimental native (Hako→LLVM C API) path for minimal EXE.
- Behind `NYASH_LLVM_BACKEND=native` toggle; no default impact.
### Phase 21.13 — llvmlite deprecation (default off)
- Remove llvmlite from auto paths; keep explicit toggle + optional CI job.
### Phase 21.14 — Optimization & Perf Harness
- Perf harness + PHI invariants; optimize hot paths; publish numbers.
## Principles
- Guard everything by env/features; defaults unchanged.
- Keep changes reversible (small diffs, RESTORE docs, fallbacks).
- Test gates: quick smokes + representative hv1/hakovm parity.
## Today (suggested)
1) Stage Phase0 move as a script (not autorun) + RESTORE.md.
2) Add phase docs (docs/private/roadmap/phases/phase-21.9/PLAN.md).
3) Keep CI/dev instructions intact (no build break when features=none).
1) Lock 22.0 (Parser/MIR Hakofirst) — builder registryを既定ON、quickが緑。
2) Prepare 22.1 (TLV C shim & Resolver SSOT) — I/F草案と最小スモーク。
3) LLVM統一21.1021.14は並行で準備、切替は22.x完了後に本格実施。
## Test Strategy (gates)
- Quick: tools/smokes/v2/profiles/quick/core/* (phase2037 flow, phase2170 state) — green.