Proposal
Successor to #57 (now closed — its goal, the .res→.affine declaration translator, is delivered). This issue tracks the next translation model for tools/res-to-affine/: a partial-port mode that translates expression- and function-level constructs, plus module-qualified references.
Why a separate issue / model
--translate today guarantees that every emitted form type-checks standalone (main.exe check → Type checking passed), which is exactly why it's bounded to self-contained top-level declarations: type aliases, sums, structs, generics, and literal let→const (slices delivered in #477 / #481 / #484). Two remaining forms cannot meet that guarantee by construction and need a different model:
1. switch→match (+ function translation) — partial-port mode
A match is an expression, only meaningful inside a function body. Producing a type-checkable result means translating the whole enclosing function — but ReScript bindings are usually un-annotated (let f = x => …), while AffineScript fn requires parameter and return types, so the output won't compile.
The model this needs is translate-with-TODO-holes: render the structural skeleton of a function (signature stub + match arms with translated patterns) and leave un-inferable bodies / types as explicit // TODO islands. This drops the standalone-type-check guarantee — output is a partial port a human finishes, matching the README's original "~60–80% translation" vision. Scope to design:
- function signature stubs from un-annotated params (TODO-typed params, inferred-where-possible)
switch e { | P => body, … } → match e { P => body, … } (pattern translation; => kept; leading | dropped; commas added)- a small expression translator for the common, safe cases (literals, identifiers, constructor application, simple calls/operators), TODO for the rest
2. Module-qualified references
Belt.Map.t now parses (the #228 grammar gap closed — parser.mly:515 wires qualified_type_name into type_expr_primary) but Belt::Map::T won't resolve against a target module that doesn't exist yet. Needs a module-mapping story (ReScript stdlib module → AffineScript stdlib module, or an explicit "unresolved import" marker) before it can be emitted without breaking a build.
Acceptance / starting points
- Decide the partial-port output contract (it is explicitly not required to type-check; document that distinction from the declaration translator).
- Reuse the existing walker AST (
tools/res-to-affine/walker.ml) and the conservative skip-don't-guess discipline.
- Gate tests like the existing
walker-phase3* suites (skip without the tree-sitter grammar).
Context
Filed as the successor to #57 at the owner's request.
Proposal
Successor to #57 (now closed — its goal, the
.res→.affinedeclaration translator, is delivered). This issue tracks the next translation model fortools/res-to-affine/: a partial-port mode that translates expression- and function-level constructs, plus module-qualified references.Why a separate issue / model
--translatetoday guarantees that every emitted form type-checks standalone (main.exe check→ Type checking passed), which is exactly why it's bounded to self-contained top-level declarations: type aliases, sums, structs, generics, and literallet→const(slices delivered in #477 / #481 / #484). Two remaining forms cannot meet that guarantee by construction and need a different model:1.
switch→match(+ function translation) — partial-port modeA
matchis an expression, only meaningful inside a function body. Producing a type-checkable result means translating the whole enclosing function — but ReScript bindings are usually un-annotated (let f = x => …), while AffineScriptfnrequires parameter and return types, so the output won't compile.The model this needs is translate-with-TODO-holes: render the structural skeleton of a function (signature stub +
matcharms with translated patterns) and leave un-inferable bodies / types as explicit// TODOislands. This drops the standalone-type-check guarantee — output is a partial port a human finishes, matching the README's original "~60–80% translation" vision. Scope to design:switch e { | P => body, … }→match e { P => body, … }(pattern translation;=>kept; leading|dropped; commas added)2. Module-qualified references
Belt.Map.tnow parses (the #228 grammar gap closed —parser.mly:515wiresqualified_type_nameintotype_expr_primary) butBelt::Map::Twon't resolve against a target module that doesn't exist yet. Needs a module-mapping story (ReScript stdlib module → AffineScript stdlib module, or an explicit "unresolved import" marker) before it can be emitted without breaking a build.Acceptance / starting points
tools/res-to-affine/walker.ml) and the conservative skip-don't-guess discipline.walker-phase3*suites (skip without the tree-sitter grammar).Context
Pkg.Type/Pkg.Effectunrepresentable (estate-wide port blocker; ADR-014) #228.tools/res-to-affine/README.md§"Scope boundary — what--translatewill not do".Filed as the successor to #57 at the owner's request.