Skip to content

CAPABILITY-MATRIX.adoc

Latest commit

 

History

History
197 lines (151 loc) · 8.07 KB

CAPABILITY-MATRIX.adoc

File metadata and controls

197 lines (151 loc) · 8.07 KB

AffineScript Capability Matrix

Important

This document is the single authoritative source for AffineScript feature readiness. Where any other document (README.adoc, BACKEND-IMPLEMENTATION.md, COMPILER-CAPABILITIES.md, ALPHA-1-RELEASE-NOTES.md, ROADMAP.adoc, .machine_readable/6a2/STATE.a2ml) disagrees with this matrix, this matrix wins and the other document is drift to be corrected. Reconstructed and verified 2026-05-19 against origin/main@dc5b817 with the compiler as oracle. Tracks issue #175 / #176 (DOC-01..09).
Table of Contents

How to read this

Status vocabulary, used precisely:

Term Meaning

enforced

Wired into the CLI pipeline and gates real user programs today.

works

Implemented and exercised by the test gate; not necessarily complete.

partial

Implemented for common cases; known gaps documented in the row.

declared-but-unwired

Surface syntax and/or an internal module exists, but the feature is not enforced on user programs through the CLI.

parse-only

Surface syntax parses; no semantics behind it yet.

absent

Does not exist, regardless of what prose elsewhere claims.

Gate baseline at reconstruction: opam exec --switch=. — dune runtest --force = 260 tests, 0 failures (2026-05-19).

Compiler core

Component Status Notes (truthed)

Lexer

works

Full Unicode. Quantity-literal ZERO/ONE tokens are not emitted (QTT numeric surface uses :1/:0/:ω sugar parsed elsewhere).

Parser

works

Menhir grammar, ADR-009 conformance. Stage B CLOSED: #215 closed with the owner-ratified ADR-012 disposition — the residual ~68 S/R + state-401 R/R conflicts are an intentional won’t-fix (Menhir resolves them correctly; the gate proves every parse is intended; eliminating them is the contortion ADR-012 forbids). just build masks + proves them, just build-loud/dune build reveal all. Module-qualified type/effect paths now parse — Pkg.Type / Pkg::Type (mixed, deep), :: canonical (ADR-014, PR #241, Refs #228; zero Menhir conflict delta). Cleared the dominant estate parse blocker (was 525/1177 .affine); estate re-audit + ReScript residue (#229) follow separately.

AST

works

All current language constructs represented.

Name resolution

works

Module loader, use/:: value paths, scope analysis.

Type checker

works

Bidirectional inference, effect rows, subsumption, Never/bottom, block divergence. Enforced on every check/compile/eval path.

Effect system

partial

Pure/impure separation + effect polymorphism enforced. Handlers: interpreter-complete; WasmGC dispatch is UnsupportedFeature pending the EH proposal or the CPS transform (the #225 CPS line is closing the async slice of this).

Borrow checker

partial — graph validation landed (CORE-01 pt 1)

Place/borrow/move infra + use-after-move, conflicting-borrow, move-while-borrowed, lambda-capture. CORE-01 part 1 (PR #240, Refs #177, gate 263/263): borrow-graph validation wired — BorrowOutlivesOwner now emitted (&local escaping its block); shared-XOR-exclusive enforced at use sites (UseWhileExclusivelyBorrowed); ownership now derived from the parameter type (TyOwn/TyRef/TyMut) so the owned/ref/mut discipline is actually enforced from real parsed source (closed a latent hole); call-arg borrows made temporary. Deferred (CORE-01 pt 2+): NLL/region inference, flow-sensitive escape via outer = &x, tighter quantity integration.

Quantity / affine types

enforced

QTT semiring in lib/quantity.ml, invoked inside the standard CLI pipeline. @linear/@erased/@unrestricted (canonical) + :1/:0/:ω sugar. Scaled-Let (ADR-002) closes BUG-001/002. Linear arrows enforced.

Interpreter

works

Pattern matching, control flow, effect handler dispatch, resume. let mut mutation path has known baseline gaps.

Traits

partial

Registry, TopTrait/TopImpl, unification-based find_impl, impl-satisfies-trait. Missing: associated-type substitution in method bodies, where-clause supertraits, coherence checking.

Dependent / refinement types

parse-only

TRefined parses; predicates do not reduce; no SMT/decision procedure.

Row polymorphism

partial

Records + effect rows in typecheck/unify; not fully exercised end-to-end.

Formatter / linter

works

AST pretty-printer; 4 lint rules. affinescript fmt is the canonical-form authority (will own the ADR-014 .→:: normalisation once #228 lands).

LSP

works

Phases A–D: hover, goto-def, completion, JSON-RPC server --stdio.

Backends (truthed — breadth is real but uneven)

The compiler emits to many targets. Maturity is not uniform; do not claim "N production backends". Order is by enforcement depth.

Target Status Notes

WASM (core)

works

Cross-module imports + extern fn → WASM import emission. The reference target.

WASM-GC

partial

Variant-with-args + same-arity PatCon destructuring; silent-bad-codegen fallbacks eliminated. Mixed-arity matches need uniform variant rep; effects/try-catch/call_ref deferred (EH/CPS — #225 line).

typed-wasm

contract, narrow

Not a backend flag — a discipline-bearing shape over the WASM output: the affinescript.ownership custom section
lib/tw_interface.ml / lib/tw_verify.ml boundary verifier. Emitted-wasm enforcement covers typed-wasm Level 7 (aliasing) + Level 10 (linearity) only. See ECOSYSTEM.adoc for the actual contract. Producer-side; the target spec is the separate hyperpolymath/typed-wasm repo.

Deno-ESM

works

Direct AST→ES-module transpiler (lib/codegen_deno.ml), --deno-esm / .deno.js. Shipped + consumer-verified (ubicity).

Node-CJS

works

lib/codegen_node.ml; CJS shim, handle table, the .affine VS Code extension compiles through it.

Julia, JS, Rust, Lua

works (transpile)

Module_loader.flatten_imports inlines imported public fns; smoke-tested.

C, WGSL, Faust, ONNX, OCaml, Bash, Nickel, ReScript, LLVM, Verilog, Gleam, CUDA, Metal, OpenCL, MLIR, Why3, Lean, SPIR-V

partial / experimental

Code-generators exist and build. Coverage varies widely; treat as experimental unless a row above promotes them. Any non-Wasm backend with no FnExtern story raises Match_failure at runtime by design (correct "this target has no host-impl story" signal).

Standard library

stdlib ~95%; 19/19 stdlib files compile resolve→typecheck→codegen with the AOT CI gate (test/test_stdlib_aot.ml) since Stage C closed 2026-05-18. Portable primitives Http (#160) / Json (#161) / Dict-Map (#162) are the current authoring frontier. A few primitives remain extern builtins.

What AffineScript is NOT (anti-over-claim)

  • Not "production-ready". Alpha. The borrow checker has a known soundness gap (CORE-01).

  • Not "N production backends". One reference target (WASM), two solid JS-host targets (Deno-ESM, Node-CJS), the rest experimental.

  • typed-wasm is not an AffineScript subsystem. It is a separate, language-agnostic target with its own repo, proofs, and other producers (ephapax). AffineScript is one producer among several.

  • The satellite ecosystem is mostly scaffold/skeleton — see the satellite registry in ECOSYSTEM.adoc. affinescript-tea/-dom-loader/-cadre were imaginary until the #175 scaffolds; -dom/-pixijs are skeletons.

Change control

Any PR that re-introduces backend-breadth, "production-ready", or stdlib-percentage over-claims must be rejected (DOC-01..09 / issue #176; Hypatia/gitbot DOC-FORMAT/DOC-DEDUP rules). Update this file in the same PR as any capability change; STATE.a2ml mirrors, it does not lead.

See also

  • ECOSYSTEM.adoc — the spine, the real AS↔typed-wasm contract, the satellite registry, the Stage A–E definitions, INT-01..12.

  • TECH-DEBT.adoc — the coordination ledger.

  • hyperpolymath/typed-wasm — the target spec (separate repo).