feat(res-to-affine): Phase-1 migration assistant skeleton (Refs #57)

docs/MIGRATION-ASSISTANT.adoc

@@ -0,0 +1,127 @@
+// SPDX-License-Identifier: MPL-2.0
+// SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell
+= ReScript → AffineScript migration assistant — architectural decision
+:toc: macro
+:toclevels: 2
+
+Tracks issue #57 (parser + metaparser). Companion to
+link:RESCRIPT-ELIMINATION.adoc[RESCRIPT-ELIMINATION.adoc], which is the
+authoritative ledger for the broader estate ReScript-surface retirement.
+
+toc::[]
+
+== Context
+
+Estate language policy retires ReScript in favour of AffineScript →
+typed-wasm. Per the inventories captured in
+link:RESCRIPT-ELIMINATION.adoc[RESCRIPT-ELIMINATION.adoc] and the
+upstream tracker `hyperpolymath/gitbot-fleet#148`, ~5k LOC of ReScript
+remains in `gitbot-fleet/bots/sustainabot/bot-integration/src/` alone;
+the idaptik tail is ~542 `.res` files plus ~80 `.ts`. By-hand
+translation is impractical, *and* a literal transliterator misses the
+point — the AffineScript answer to ReScript's anti-patterns is
+*re-decomposition*, not a token-level rewrite.
+
+Issue #57 proposes a *migration assistant*: a tool that reads `.res`,
+recognises the anti-patterns surfaced by the idaptik Wave 3 pilot, and
+emits a `.affine` *skeleton* that surfaces the work the human migrator
+still owes.
+
+== Decision
+
+* The migration assistant lives at `tools/res-to-affine/` as an OCaml
+  CLI built by the repo's existing `dune` toolchain.
+* The canonical source-of-truth grammar for `.res` parsing is
+  https://github.com/rescript-lang/tree-sitter-rescript[`rescript-lang/tree-sitter-rescript`],
+  vendored manifest-only at `editors/tree-sitter-rescript/` (pinned to
+  commit `990214a83f25801dfe0226bd7e92bb71bba1970f`, version 6.0.0,
+  MIT-licensed and compatible with this repo's MPL-2.0).
+* The tool ships in three phases:
++
+[cols="1,3,2"]
+|===
+| Phase | What it does | Status
+
+| 1
+| Text-scan emitter detecting 4 of the 6 anti-patterns, emitting a
+  `.affine` skeleton with migration markers and the quoted original
+  for reference.
+| this PR
+
+| 2
+| Replaces the text scanner with a tree-sitter AST walker reading the
+  vendored grammar. Adds the two deferred patterns. Same emitter
+  interface.
+| follow-up
+
+| 3
+| Partial *translation* of pure-structural forms (type aliases, sum
+  decls, simple `let` bindings, `switch` → `match`). Effect-laden /
+  exception-bearing / globally-mutating regions remain TODO islands.
+| follow-up
+|===
+* The Phase-1 deliverable is **deliberately small and useful in
+  isolation**. It gates the architectural commitment to tree-sitter
+  behind something that already pays its way against real estate
+  `.res` files.
+
+== Alternatives considered
+
+=== Use the ReScript compiler's own AST (`bs-tools` / `rescript ast`)
+
+The richest signal is in the ReScript compiler's typed AST. Rejected
+because:
+
+* Adding `rescript` as a build-time dependency contradicts the estate
+  language policy (which bans new ReScript code and treats ReScript as
+  the artefact to be retired).
+* The ReScript compiler's AST changes across versions in
+  non-backwards-compatible ways; pinning would create an ongoing
+  compatibility burden in the wrong direction.
+
+=== Write a hand-rolled `.res` lexer/parser in OCaml
+
+We already have `lib/rescript_codegen.ml` going *affinescript → .res*,
+so the grammar is partly understood. Rejected because:
+
+* ReScript's surface syntax is large; recreating it for a one-way
+  migration tool is days-to-weeks of work that the canonical
+  tree-sitter grammar has already done and maintains.
+* The community grammar is MIT-licensed and version-pinned; the cost
+  of consuming it is a one-line manifest plus an install script.
+
+=== Pattern-detector only (no AST in any phase)
+
+Phase 1 *is* this — but committing to it permanently would leave the
+two structural anti-patterns (callback records, oversized functions)
+undetected forever, and would block Phase 3 (partial translation),
+which is what makes the tool earn its keep on idaptik's 542 files.
+
+== Consequences
+
+* `editors/tree-sitter-rescript/` exists for the migration pipeline,
+  not as an editor binding. The editor binding for AffineScript itself
+  remains `editors/tree-sitter-affinescript/`.
+* `tools/res-to-affine/` is the first OCaml tool under `tools/`
+  (existing tools are shell scripts or Rust). The `dune` integration
+  is local to the tool's own `dune` file; no workspace changes.
+* Phase 2 introduces `tree-sitter` CLI as a runtime dependency for the
+  migration assistant. It is *not* a build-time dependency for the
+  AffineScript compiler itself. CI for the migration tool's Phase-2
+  tests will need to install `tree-sitter-cli`.
+* The Phase plan is recorded in
+  link:../tools/res-to-affine/README.md[`tools/res-to-affine/README.md`];
+  this document is the architectural decision, the README is the
+  user/contributor surface.
+
+== References
+
+* `tools/res-to-affine/README.md` — tool usage, Phase plan, design rationale.
+* `editors/tree-sitter-rescript/README.md` — vendoring manifest details.
+* `affinescript#57` — parser + metaparser proposal.
+* `gitbot-fleet#148` — downstream tracker for the consumed ReScript subtree.
+* link:RESCRIPT-ELIMINATION.adoc[`RESCRIPT-ELIMINATION.adoc`] — estate-wide ledger.
+* https://github.com/hyperpolymath/idaptik/blob/main/migration/main/LESSONS.md[idaptik LESSONS.md]
+  — six anti-patterns the assistant targets.
+* https://github.com/hyperpolymath/idaptik/blob/main/migration/main/PILOT.md[idaptik PILOT.md]
+  — original Wave-3 pilot that surfaced the six patterns.

editors/tree-sitter-rescript/README.md

@@ -0,0 +1,44 @@
+<!-- SPDX-License-Identifier: MPL-2.0 -->
+<!-- SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell -->
+
+# tree-sitter-rescript (vendoring manifest)
+
+This directory is a **manifest-only vendoring** of the canonical
+[`rescript-lang/tree-sitter-rescript`][upstream] grammar. The grammar
+itself is not copied into this repository — `package.json` declares it
+as a dependency, and `scripts/install.sh` fetches and builds it at the
+pinned commit.
+
+The grammar is consumed by `tools/res-to-affine/`, the `.res → .affine`
+migration assistant (`affinescript#57`). It is **not** an editor binding
+for AffineScript; for that, see `editors/tree-sitter-affinescript/`.
+
+## Pinned upstream
+
+- **Repository:** <https://github.com/rescript-lang/tree-sitter-rescript>
+- **Commit:** `990214a83f25801dfe0226bd7e92bb71bba1970f`
+- **Version:** 6.0.0
+- **License:** MIT (preserved upstream; compatible with this repo's MPL-2.0)
+
+When updating the pin, regenerate `tools/res-to-affine/test/expected/`
+snapshots, since AST shapes may shift.
+
+## Install
+
+```sh
+./scripts/install.sh
+```
+
+This writes a `tree-sitter-rescript` directory under `tools/vendor/`
+(gitignored — same convention as the WASI adapter pinning), containing
+the generated parser. Requires `git` and `tree-sitter` CLI on PATH.
+
+## Why manifest, not copy
+
+The upstream grammar is ~10k lines of JS plus generated C. Copying it
+into this MPL-2.0 repo would (a) bloat the tree, (b) create an ongoing
+sync burden, and (c) duplicate MIT-licensed code we have no business
+modifying. The manifest+install approach keeps the dependency explicit
+and pinned without absorbing the source.
+
+[upstream]: https://github.com/rescript-lang/tree-sitter-rescript

editors/tree-sitter-rescript/package.json

@@ -0,0 +1,16 @@
+{
+  "name": "@affinescript/tree-sitter-rescript-vendoring",
+  "version": "0.1.0",
+  "private": true,
+  "description": "Manifest-only vendoring of rescript-lang/tree-sitter-rescript for the .res -> .affine migration assistant (affinescript#57).",
+  "license": "MPL-2.0",
+  "dependencies": {
+    "tree-sitter-rescript": "github:rescript-lang/tree-sitter-rescript#990214a83f25801dfe0226bd7e92bb71bba1970f"
+  },
+  "devDependencies": {
+    "tree-sitter-cli": "^0.25.0"
+  },
+  "scripts": {
+    "install-grammar": "./scripts/install.sh"
+  }
+}

editors/tree-sitter-rescript/scripts/install.sh

@@ -0,0 +1,41 @@
+#!/usr/bin/env bash
+# SPDX-License-Identifier: MPL-2.0
+# SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell
+#
+# Fetch and build the pinned tree-sitter-rescript grammar.
+# Output goes under ../../.build/tree-sitter-rescript/ (gitignored).
+
+set -euo pipefail
+
+UPSTREAM_COMMIT="990214a83f25801dfe0226bd7e92bb71bba1970f"
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../../.." && pwd)"
+# tools/vendor/ is the repo's convention for fetched-not-committed deps
+# (see .gitignore line 103, mirrors the WASI adapter provisioning).
+BUILD_DIR="${REPO_ROOT}/tools/vendor/tree-sitter-rescript"
+
+if ! command -v tree-sitter >/dev/null 2>&1; then
+  echo "error: tree-sitter CLI not found on PATH" >&2
+  echo "       install via: npm install -g tree-sitter-cli" >&2
+  exit 2
+fi
+
+if ! command -v git >/dev/null 2>&1; then
+  echo "error: git not found on PATH" >&2
+  exit 2
+fi
+
+mkdir -p "$(dirname "$BUILD_DIR")"
+
+if [ -d "$BUILD_DIR/.git" ]; then
+  git -C "$BUILD_DIR" fetch --quiet origin "$UPSTREAM_COMMIT" || true
+  git -C "$BUILD_DIR" checkout --quiet "$UPSTREAM_COMMIT"
+else
+  rm -rf "$BUILD_DIR"
+  git clone --quiet https://github.com/rescript-lang/tree-sitter-rescript.git "$BUILD_DIR"
+  git -C "$BUILD_DIR" checkout --quiet "$UPSTREAM_COMMIT"
+fi
+
+cd "$BUILD_DIR"
+tree-sitter generate
+
+echo "tree-sitter-rescript built at ${BUILD_DIR} (commit ${UPSTREAM_COMMIT})"

tools/res-to-affine/README.md

@@ -0,0 +1,138 @@
+<!-- SPDX-License-Identifier: MPL-2.0 -->
+<!-- SPDX-FileCopyrightText: 2026 Jonathan D.A. Jewell -->
+
+# `res-to-affine` — ReScript-to-AffineScript migration assistant
+
+A small OCaml CLI that reads a `.res` file and emits a `.affine` skeleton
+with **migration markers** — comments that name each anti-pattern the
+scanner found, point at the source line, and propose the AffineScript
+answer the human migrator should consider before porting.
+
+Tracks: [`affinescript#57`](https://github.com/hyperpolymath/affinescript/issues/57)
+(parser + metaparser).
+Consumed by: [`hyperpolymath/gitbot-fleet#148`](https://github.com/hyperpolymath/gitbot-fleet/issues/148)
+and the broader `idaptik` migration.
+
+## Usage
+
+```sh
+# print skeleton to stdout
+dune exec tools/res-to-affine/main.exe -- path/to/Foo.res
+
+# or write to a file
+dune exec tools/res-to-affine/main.exe -- path/to/Foo.res -o Foo.affine
+```
+
+The output is **not compilable**. It is a starting point for the human:
+a quoted copy of the original sits at the bottom; the top carries a
+migration-considerations block; the middle is a `module` stub with
+`TODO`s. The human picks the decomposition; the tool surfaces what
+needs re-decomposing.
+
+## What gets flagged (Phase 1)
+
+The six anti-patterns surfaced in the
+[idaptik Wave 3 pilot](https://github.com/hyperpolymath/idaptik/blob/main/migration/main/LESSONS.md),
+of which the line-based scanner reliably detects four:
+
+| Tag | Detection | AffineScript answer |
+|---|---|---|
+| `side-effect-import` | `let _ = Mod.foo` at top level | Explicit registration call |
+| `raw-js` | `%raw(...)` or `[%bs.raw ...]` | Typed extern (`ABI-FFI-README.md`) |
+| `untyped-exception` | `Promise.catch`, `Js.Exn`, `raise`, `try` | `Result[E, A]` / `Validation[E, A]` |
+| `mutable-global` | `:=` operator | Affine record threaded through |
+
+Deferred to Phase 2 (need real AST):
+
+- **inline lambda callback record** — N ≥ 3 `~handler: (...) =>` lambdas
+  inside one record literal (collapse to a row-polymorphic record).
+- **oversized function** — function body > ~50 LOC (decompose).
+
+## Why a skeleton and not a transliteration
+
+The Frontier Programming Guides' standing rule is **re-decompose, not
+transliterate**. A line-for-line port preserves the source's anti-patterns
+into the target language and produces `.affine` files that are technically
+parseable but architecturally still ReScript. The migration assistant's
+job is to *make the re-decomposition tractable*, not to skip it. So:
+
+- The skeleton is **honest about being incomplete** — it does not
+  compile, on purpose.
+- The original source is **quoted at the bottom** so the migrator
+  doesn't tab between files while writing the port.
+- Each marker links a source line to the AffineScript pattern that
+  replaces it, so the migrator's next action is clear.
+
+## Phase plan
+
+### Phase 1 — text-scan emitter (this PR)
+
+- OCaml binary builds with the repo's existing `dune` toolchain.
+- `Scanner` walks lines with `str` regexes; cheap and dependency-free.
+- `Emitter` writes the migration-considerations block, a `module` stub,
+  and the quoted source.
+- Snapshot tests under `test/` ensure stable output.
+
+This phase is **deliberately small**. It is useful immediately — runs
+against any `.res` file, surfaces 4 of 6 anti-patterns, gives the
+migrator a starting document — and it gates the architectural commitment
+to tree-sitter in Phase 2 behind something that already pays its way.
+
+### Phase 2 — tree-sitter AST walker
+
+- Install the pinned grammar from
+  `editors/tree-sitter-rescript/` (manifest-only vendoring of
+  `rescript-lang/tree-sitter-rescript@990214a`).
+- Replace `Scanner` with a walker over the s-expression output of
+  `tree-sitter parse --quiet`, parsed by the existing `sexplib0`
+  dependency.
+- Adds the two deferred patterns (callback records, oversized
+  functions) and unlocks **structural** translation of trivial forms
+  (e.g. `option<X>` → `Option[X]`, `result<X, Y>` → `Result[Y, X]`,
+  `switch x { | A => ... }` → `match x { A => ... }`).
+- The `Emitter` interface does not change: same skeleton shape, same
+  marker schema, richer body.
+
+### Phase 3 — partial translation
+
+Once the AST walker exists, the emitter can do more than mark — it can
+**translate** the pure-structural parts (type aliases, sum decls,
+simple `let` bindings, switch-to-match) and leave only effect-laden,
+exception-bearing, or globally-mutating regions as TODO. The skeleton
+becomes a working port of ~60–80% of the input, with TODO islands
+where re-decomposition is genuinely required.
+
+Phase 3 is when the tool earns its keep on idaptik's 542 files.
+
+## Testing
+
+```sh
+dune test tools/res-to-affine/
+```
+
+To regenerate snapshots after an intentional emitter change:
+
+```sh
+cd tools/res-to-affine/test
+../../../_build/default/tools/res-to-affine/main.exe \
+    fixtures/sample.res > expected/sample.affine
+```
+
+The fixture under `test/fixtures/sample.res` is synthetic and exercises
+every Phase-1 anti-pattern. Real `.res` files from the estate (e.g.
+`gitbot-fleet/bots/sustainabot/bot-integration/src/*.res`) can be run
+ad hoc through the CLI without changes to the test suite.
+
+## Non-goals
+
+- **Not a ReScript compiler.** The scanner does not parse ReScript;
+  even Phase 2 only walks the tree-sitter CST, not the ReScript
+  type-checker's AST. If a `.res` file is syntactically invalid the
+  tool may still emit a (less useful) skeleton.
+- **Not a build-time dependency on ReScript.** The pinned grammar is a
+  parser, not the ReScript compiler. The estate's language policy
+  (CLAUDE.md) bans new ReScript code; this tool exists to **help retire
+  the existing ReScript surface**, not to bring more in.
+- **Not for editor integration.** Editor tree-sitter bindings for
+  AffineScript live at `editors/tree-sitter-affinescript/`; this tool's
+  vendored grammar is for the migration pipeline only.

tools/res-to-affine/dune

@@ -0,0 +1,13 @@
+; SPDX-License-Identifier: MPL-2.0
+
+(executable
+ (name main)
+ (modules main)
+ (public_name res-to-affine)
+ (package affinescript)
+ (libraries res_to_affine cmdliner fmt fmt.tty))
+
+(library
+ (name res_to_affine)
+ (modules scanner emitter)
+ (libraries str))