Migrate fjs/bnf/data to .f.mjs - #1487

Approving. Reviewed at 28199f7 against main (03126e4). Five modules — bnf/data, bnf/ll1, bnf/descent, effects, effects/list — plus ~45 dependents updated.

Two modules genuinely restructure, and both check out

bnf/data, bnf/descent, and effects/list are equivalent to main's compiled output. The other two each carry one real code change, both the same idiom: JSDoc has no call-site type-argument syntax, so an inline argument gets extracted into a named const that can be annotated.

// ll1:  rangeMap<DispatchResult>({…})   ->
/** @type {Properties<_DispatchResult>} */
const dispatchProps = { … }
const dispatchOp = rangeMap(dispatchProps)

// effects:  fold<T, Effect<O|Q, S>>(item => acc => step(acc, f(item)))   ->
/** @type {Fold<T, Effect<O | Q, S>>} */
const op = item => acc => step(acc, f(item))
return step(items, fold(op)(pure(init)))

I enumerated every normalized difference in each module to confirm these are the only ones — one hunk in ll1, one in effects — then checked behaviour rather than arguing from the shape. effects is the core combinator module, so it got a direct comparison against main:

foldStep sum 1..5    main: [15]   pr: [15]
foldStep empty list  main: [42]   pr: [42]      (init passes through)
forEachStep          main: [null] pr: [null]
okStep ok / error    main: [42] [["error","boom"]]
                     pr  : [42] [["error","boom"]]

Both okStep branches, the empty-list edge, and the folding path all match. For ll1 I leaned on the repo's own coverage instead — my ad-hoc parser harness was calling the API wrong (it failed identically against main, so it was my mistake, not a finding), and the three migrated modules' proofs contribute 164 passing cases, with another 222 across effects.

effects/list shows up as differing only because my normalizer rewrites '.f.js' in single quotes and that file uses double quotes — the sole difference is the import specifier, and empty/nonEmpty produce identical results.

Everything else

• npx tsc clean; 2356 pass / 0 fail.
• Public API: 0 removed, 0 non-_ additions. All 10 new types are _-prefixed. The 3 changed types and 3 changed const signatures are entirely those renames (AstRule→_AstRule, DispatchRule→_DispatchRule, DispatchMap→_DispatchMap) plus one effects/module.f.ts→effects/types.ts import path. I confirmed all three renamed types were non-exported on main, so nothing public was renamed.
• No any as a type in any emitted declaration.
• Markdown links: 144 broken on main, 144 here, none added, 0 label/target mismatches. No bare .f.ts code-span references to migrated files either — the shape that slipped through on #1486.
• All four new types.ts files keep the @module header through emit.
• CHANGELOG entries for all five, including the explicit "proof.f.ts stays TypeScript for now" note.

The proof.f.ts decision is right and I verified the reason. All three bnf proofs import runtime values from ../testlib.f.ts, and data's also from media/json/module.f.ts — both still .f.ts on disk, so §2 does not yet permit proof.f.mjs. Stating that in the PR body and the CHANGELOG rather than leaving a reviewer to work it out is the right call.

CI still finishing on 28199f7; verified locally at that commit.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 28199f733e