functionalscript 0.46.0 → 0.47.0

package/fjs/djs/parser/module.f.mjs

@@ -8,257 +8,349 @@
  * @import { Fold } from '../../types/function/operator/types.ts'
  * @import { DjsToken, DjsTokenWithMetadata } from '../tokenizer/types.ts'
  * @import { OrderedMap } from '../../types/ordered_map/types.ts'
- * @import { AstArray, AstConst, AstModule, AstModuleRef } from '../ast/types.ts'
+ * @import { AstArray, AstConst, AstModule, AstModuleRef, AstObject } from '../ast/types.ts'
  * @import { TokenMetadata } from '../../js/tokenizer/types.ts'
- * @import { ParseError, _ValueToken } from './types.ts'
+ * @import { ParseError, _FramingKeyword, _OrdinaryTokenName, _ValueToken } from './types.ts'
+ * @import { Assert } from '../../asserts/types.ts'
+ * @import { Equal } from '../../types/ts/types.ts'
+ * @import { CodePointMeta } from '../../bnf/descent/types.ts'
+ * @import { Ast, AstSequence } from '../../bnf/matcher/types.ts'
+ * @import { Rule, TerminalRange } from '../../bnf/types.ts'
+ * @import { DescentMatch } from '../../bnf/descent/types.ts'
  */
 
 import { error, ok } from '../../types/result/module.f.mjs'
 import { fold, next, toArray, length, concat } from '../../types/list/module.f.mjs'
 import { setReplace, at } from '../../types/ordered_map/module.f.mjs'
 import { fromMap } from '../../types/object/module.f.mjs'
-import { assertEq } from '../../asserts/module.f.mjs'
+import { assert, assertEq, assertNotNullish } from '../../asserts/module.f.mjs'
+import { eof, oneEncode, option, rangeDecode, repeat0Plus, unicodeRange } from '../../bnf/module.f.mjs'
+import { encoding } from '../../bnf/token_symbol/module.f.mjs'
+import { toData } from '../../bnf/data/module.f.mjs'
+import { descentParserRuleSet } from '../../bnf/descent/module.f.mjs'
 
-/** @typedef {['array', List<AstConst>]} _DjsStackArray */
+/**
+ * The ordinary token stream a BNF parser layer consumes, with the tokenizer's
+ * one physical end-of-input token split off.
+ *
+ * @typedef {{
+ *   readonly tokens: readonly DjsTokenWithMetadata[]
+ *   readonly eofMetadata: TokenMetadata
+ * }} _TokenStream
+ */
+
+/**
+ * Splits the tokenizer's single final physical `eof` token off a token list.
+ *
+ * A BNF parser backend synthesizes its own logical end-of-input, so passing the
+ * tokenizer's physical `eof` through as an ordinary symbol would create a second
+ * end marker. Dropping it outright would instead lose the source position that a
+ * failure *at* physical end has to be reported from, so its metadata is kept
+ * aside as `eofMetadata` rather than discarded or refabricated.
+ *
+ * The tokenizer's contract is exactly one `eof`, in final position; a stream
+ * carrying one anywhere else is rejected here rather than parsed.
+ *
+ * A stream with no `eof` at all has two causes, and they are not reported the
+ * same way. A lexical failure — an unterminated string or comment — ends the
+ * stream at an `error` token and emits no `eof`, which is the tokenizer working
+ * correctly on bad input; that error is passed through with its own position.
+ * Anything else missing an `eof` is a genuine contract violation and has no
+ * position to report.
+ *
+ * @type {(tokenList: List<DjsTokenWithMetadata>) => Result<_TokenStream, ParseError>}
+ */
+const splitEof = tokenList => {
+    const a = toArray(tokenList)
+    const eofIdx = a.findIndex(({ token }) => token.kind === 'eof')
+    if (eofIdx === -1) {
+        // A lexical failure ends the stream at its `error` token and emits no
+        // `eof`, so the absence of one is not always a broken contract. Rejecting
+        // it as one would answer "unterminated string at 1:11" with "missing
+        // end-of-input token" and no position at all, so the error is reported
+        // where it happened — the same place the hand-written parser reports it.
+        const lastToken = a[a.length - 1]
+        return lastToken !== undefined && lastToken.token.kind === 'error'
+            ? error({ message: 'unexpected token', metadata: lastToken.metadata })
+            : error({ message: 'missing end-of-input token', metadata: null })
+    }
+    const last = a.length - 1
+    if (eofIdx !== last) {
+        return error({ message: 'end-of-input token is not final', metadata: a[eofIdx].metadata })
+    }
+    return ok({ tokens: a.slice(0, last), eofMetadata: a[last].metadata })
+}
 
-/** @typedef {['object', OrderedMap<AstConst>, string]} _DjsStackObject */
+/**
+ * The parser layer's complete finite alphabet: every token name its grammar may
+ * name as a terminal, and the exact set a token-name-to-symbol mapping has to be
+ * validated over before parsing.
+ *
+ * `eof` is not a member — {@link splitEof} removes the tokenizer's physical
+ * end-of-input token before any name is mapped, and the backend synthesizes its
+ * own logical one.
+ *
+ * The names are the *token* vocabulary, not the tokenizer grammar's tag
+ * vocabulary: only eight punctuators survive into `DjsToken`, so the JS operator
+ * set the tokenizer recognizes is far larger than what reaches this layer.
+ *
+ * A name is not always a kind. The framing keywords arrive as `id` tokens and
+ * need terminals of their own, or the grammar could not tell `export default`
+ * from two arbitrary identifiers — see {@link framingKeywords}.
+ *
+ * The `_…AreComplete` assertions below check both halves against `DjsToken` and
+ * `_FramingKeyword` at compile time, so a kind or keyword added there breaks the
+ * build rather than going unrepresented.
+ */
+const tokenKindNames = /** @type {const} */ ([
+    'true', 'false', 'null', 'undefined',
+    '{', '}', ':', ',', '[', ']', '.', '=',
+    'string', 'number', 'error', 'id', 'bigint',
+    'ws', 'nl', '//', '/*',
+])
 
-/** @typedef {_DjsStackArray | _DjsStackObject} _DjsStackElement */
+/**
+ * The framing keywords, which the tokenizer emits as `id` tokens carrying the
+ * word in `value`. Kept as its own list because the mapping has to recognize
+ * exactly these values, not merely encode them.
+ *
+ * **A grammar over this alphabet owes them an identifier rule.** None of the
+ * five is reserved: outside the framing positions the parser accepts them as
+ * ordinary identifiers, so `const export = 1`, `export default export`, and
+ * `{ from: 2, default: 3 }` all parse today. Once each carries its own symbol, a
+ * rule whose identifier terminal is the bare `id` symbol rejects every one of
+ * them. Wherever an identifier is accepted — binding names, references, object
+ * keys, import names — the terminal has to be the union of `id` and these five.
+ *
+ * Giving a word its own symbol narrows where it is *required*, never where it is
+ * *allowed*.
+ */
+const framingKeywords = /** @type {const} */ (['import', 'const', 'export', 'default', 'from'])
 
-/** @typedef {List<_DjsStackElement>} _DjsStack */
+/**
+ * The complete alphabet: one name per `DjsToken` kind except `eof`, plus one per
+ * framing keyword. No keyword collides with a kind, so the two lists concatenate
+ * without a name being registered twice — which `encoding` would reject anyway.
+ */
+const ordinaryTokenNames = [...tokenKindNames, ...framingKeywords]
 
-/** @typedef {_InitialState | _NewLineRequiredState | _ImportState | _ConstState | _ExportState | _ParseValueState | _ResultState | _ErrorState} _ParserState */
+/** @typedef {Assert<Equal<(typeof tokenKindNames)[number], Exclude<DjsToken['kind'], 'eof'>>>} _KindsAreComplete */
 
-/** @typedef {{
- *   readonly refs: OrderedMap<AstModuleRef>
- *   readonly modules: List<string>
- *   readonly consts: List<AstConst>
- * }} _ModuleState */
-
-/** @typedef {{
- *   readonly state: ''
- *   readonly module: _ModuleState
- * }} _InitialState */
-
-/** @typedef {{
- *   readonly state: 'nl'
- *   readonly module: _ModuleState
- * }} _NewLineRequiredState */
-
-/** @typedef {{
- *   readonly state: 'import' | 'import+name' | 'import+from'
- *   readonly module: _ModuleState
- * }} _ImportState */
-
-/** @typedef {{
- *   readonly state: 'const' | 'const+name'
- *   readonly module: _ModuleState
- * }} _ConstState */
-
-/** @typedef {{
- *   readonly state: 'export'
- *   readonly module: _ModuleState
- * }} _ExportState */
+/** @typedef {Assert<Equal<(typeof framingKeywords)[number], _FramingKeyword>>} _KeywordsAreComplete */
+
+/** @typedef {Assert<Equal<(typeof ordinaryTokenNames)[number], _OrdinaryTokenName>>} _AlphabetIsComplete */
 
 /**
- * Where the value parser stands inside the value it is reading.
+ * `eof` is not a member of the alphabet, so a second end marker cannot be
+ * encoded rather than merely going unused — and `encode` would reject the name
+ * outright. Checked at the type level because that is where it is decidable:
+ * `includes('eof')` does not even compile against this element type.
  *
- * The object states spell a property out left to right: `'{'` expects a key,
- * `'{k'` the `:` after one, `'{:'` the value, `'{v'` the `,` or `}` after it,
- * and `'{,'` the next key. A computed key `["a"]` takes the two extra steps
- * `'{['` (the string inside the brackets) and `'{[k'` (the closing `]`), then
- * rejoins the plain path at `'{k'`.
+ * @typedef {Assert<Equal<Extract<_OrdinaryTokenName, 'eof'>, never>>} _EofIsNotAName
+ */
+
+/**
+ * The alphabet's encoding, built once for the module rather than per parse.
  *
- * @typedef {'' | '[' | '[v' | '[,' | '{' | '{[' | '{[k' | '{k' | '{:' | '{v' | '{,'} _ValueState
+ * `encoding` asserts what the mapping needs — capacity, and no repeated name —
+ * so an alphabet that could not produce distinct symbols fails here at load
+ * rather than midway through a parse. Symbols start at `0x110000`, one past the
+ * last Unicode scalar value, so a token symbol can never be mistaken for a code
+ * point of the layer below.
  */
+const tokenEncoding = encoding(ordinaryTokenNames)
 
-/** @typedef {{
- *   readonly state: 'constValue' | 'exportValue'
- *   readonly module: _ModuleState
- *   readonly valueState: _ValueState
- *   readonly top: _DjsStackElement | null
- *   readonly stack: _DjsStack
- * }} _ParseValueState */
+/**
+ * One ordinary token as a descent input leaf: the symbol standing for its kind,
+ * paired with the whole token as metadata.
+ *
+ * The grammar above sees only the symbol — one per token, which is what makes a
+ * token stream an alphabet — while the token's value and source position ride
+ * along untouched, so nothing a diagnostic or an AST fold needs is lost.
+ *
+ * `eof` is not in the alphabet and `encode` would reject it. Reaching it here
+ * means {@link splitEof} was skipped, which is a caller bug rather than bad
+ * input, so this asserts instead of widening the result to a `Result`.
+ *
+ * @type {(t: DjsTokenWithMetadata) => CodePointMeta<DjsTokenWithMetadata>}
+ */
+const tokenToSymbol = t => {
+    const { token } = t
+    // A framing keyword arrives as an `id` carrying the word, so the name comes
+    // from the value there and from the kind everywhere else. `find` rather than
+    // a set membership test because it also narrows the result to the keyword
+    // union, which is what lets `encode` be called without a cast.
+    const keyword = token.kind === 'id'
+        ? framingKeywords.find(k => k === token.value)
+        : undefined
+    const name = keyword ?? token.kind
+    assert(name !== 'eof', ['eof token reached the parser alphabet', t])
+    return [tokenEncoding.encode(name), t]
+}
 
-/** @typedef {{
- *   readonly state: 'result'
- *   readonly module: _ModuleState
- * }} _ResultState */
+/**
+ * One token name as a grammar terminal.
+ *
+ * A symbol and a `TerminalRange` are both plain numbers, so `oneEncode` is what
+ * says which one is meant — `encode` returns the bare symbol a stream carries,
+ * and a rule needs the singleton range containing it.
+ *
+ * @type {(name: _OrdinaryTokenName) => TerminalRange}
+ */
+const sym = name => oneEncode(tokenEncoding.encode(name))
 
-/** @typedef {{
- *   readonly state: 'error'
- *   readonly error: ParseError
- * }} _ErrorState */
+/**
+ * Trivia is skipped between every pair of tokens, so almost every rule below is
+ * interleaved with it.
+ */
+const trivia = repeat0Plus({
+    ws: sym('ws'),
+    nl: sym('nl'),
+    lineComment: sym('//'),
+    blockComment: sym('/*'),
+})
 
 /**
- * A statement begins with `import`, `const`, or `export` and with nothing
- * else. The three are a whitelist that grows as the language does, not a
- * closed set; what a statement may never begin with is a **value**. A text
- * that does — `42`, `[1,2]`, `{"a":1}` — is a JSON document, and reading it
- * as a module would give it a value no JavaScript engine gives it: as
- * JavaScript `{"a":1}` does not parse at all, and `[1,2]` is an expression
- * statement exporting nothing.
- *
- * The statements are also ordered: every `import` precedes every `const`, and
- * `export default` ends the module. `eof` here is a module with no
- * `export default`, which is what "unexpected end" reports.
- *
- * @type {(token: DjsTokenWithMetadata) => (state: _InitialState) => _ParserState}
+ * Trivia that stops at a newline, for the one place a newline is not trivia:
+ * the statement separator. `import`/`const` statements must be newline-separated
+ * — the `'nl'` state in the hand-written parser — so a rule that swallowed
+ * newlines as trivia everywhere could not express it.
  */
-const parseInitialOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'id': {
-            switch (token.value) {
-                case 'import': return length(state.module.consts) === 0
-                    ? { ...state, state: 'import' }
-                    : { state: 'error', error: { message: 'import must come before const', metadata } }
-                case 'const': return { ...state, state: 'const' }
-                case 'export': return { ...state, state: 'export' }
-            }
-            break
-        }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-    }
-    return { state: 'error', error: { message: 'unexpected token', metadata } }
-}
+const statementEnd = () => [
+    repeat0Plus({ ws: sym('ws'), lineComment: sym('//'), blockComment: sym('/*') }),
+    sym('nl'),
+    trivia,
+]
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _NewLineRequiredState) => _ParserState} */
-const parseNewLineRequiredOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case '//':
-        case '/*': return state
-        case 'nl': return { ...state, state: '' }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * Every word that may stand where an identifier is expected: a plain `id` and
+ * each framing keyword, since none of them is reserved.
+ *
+ * This is the union {@link framingKeywords} obliges the grammar to provide.
+ */
+const identifier = {
+    id: sym('id'),
+    import: sym('import'),
+    const: sym('const'),
+    export: sym('export'),
+    default: sym('default'),
+    from: sym('from'),
+}
+
+/** A value that is one token. */
+const primitive = {
+    null: sym('null'),
+    true: sym('true'),
+    false: sym('false'),
+    undefined: sym('undefined'),
+    number: sym('number'),
+    string: sym('string'),
+    bigint: sym('bigint'),
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ExportState) => _ParserState} */
-const parseExportOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        case 'id': {
-            if (token.value === 'default') return { ...state, state: 'exportValue', valueState: '', top: null, stack: null }
-        }
-    }
-    return { state: 'error', error: { message: 'unexpected token', metadata } }
+/**
+ * `open item, item, ... ,? close` with trivia everywhere and an optional
+ * trailing comma, which both arrays and objects allow.
+ *
+ * The trailing comma works because a failed repetition round rewinds rather than
+ * failing the match ([`bnf/descent`](../../bnf/descent/README.md)): on the final
+ * `,` the round consumes the comma, finds `]` where an item belongs, and ends the
+ * repetition back at the comma for the optional tail to take.
+ *
+ * @type {(open: TerminalRange, close: TerminalRange, item: Rule) => Rule}
+ */
+const delimited = (open, close, item) => () => {
+    // Each element is wrapped in a one-branch variant so it carries the tag
+    // `item`. The branch is a *sequence* rather than the rule itself, because a
+    // variant used directly as another variant's branch loses its tag to
+    // whichever inner branch matches — and every element here is a variant.
+    // The tag is what lets the fold find elements by name instead of by
+    // position in the delimiter scaffolding.
+    const element = { item: [item] }
+    return [
+        open,
+        trivia,
+        option([
+            element,
+            trivia,
+            repeat0Plus([sym(','), trivia, element, trivia]),
+            option([sym(','), trivia]),
+        ]),
+        close,
+    ]
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ResultState) => _ParserState} */
-const parseResultOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*':
-        case 'eof': return state
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
-}
+/** @type {Rule} */
+const value = () => ({ primitive, ref: identifier, array, object })
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ConstState) => _ParserState} */
-const parseConstOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'id': {
-            if (at(token.value)(state.module.refs) !== null)
-                return { state: 'error', error: { message: 'duplicate id', metadata } }
-            /** @type {AstModuleRef} */
-            const cref = ['cref', length(state.module.consts)]
-            const refs = setReplace(token.value)(cref)(state.module.refs)
-            return { ...state, state: 'const+name', module: { ...state.module, refs: refs } }
-        }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
-}
+const array = delimited(sym('['), sym(']'), value)
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ConstState) => _ParserState} */
-const parseConstNameOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case '=': return { ...state, state: 'constValue', valueState: '', top: null, stack: null }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/** A property name: bare identifier, string literal, or a computed `["a"]`. */
+const key = {
+    plain: identifier,
+    string: sym('string'),
+    computed: () => [sym('['), trivia, { name: [sym('string')] }, trivia, sym(']')],
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ImportState) => _ParserState} */
-const parseImportOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'id': {
-            if (at(token.value)(state.module.refs) !== null) {
-                return { state: 'error', error: { message: 'duplicate id', metadata } }
-            }
-            /** @type {AstModuleRef} */
-            const aref = ['aref', length(state.module.modules)]
-            const refs = setReplace(token.value)(aref)(state.module.refs)
-            return { ...state, state: 'import+name', module: { ...state.module, refs: refs } }
-        }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/** @type {Rule} */
+const member = { member: () => [{ key: [key] }, trivia, sym(':'), trivia, { value: [value] }] }
+
+const object = delimited(sym('{'), sym('}'), member)
+
+// Each statement is tagged for the same reason an element is: the fold reads
+// the module by finding `import`/`const`/`export` nodes, not by counting past
+// the trivia and separators between them.
+const importStatement = {
+    import: () => [
+        sym('import'), trivia, { name: [identifier] },
+        trivia, sym('from'), trivia, { module: [sym('string')] },
+    ],
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ImportState) => _ParserState} */
-const parseImportNameOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        case 'id': {
-            if (token.value === 'from') return { ...state, state: 'import+from' }
-        }
-    }
-    return { state: 'error', error: { message: 'unexpected token', metadata } }
+const constStatement = {
+    const: () => [
+        sym('const'), trivia, { name: [identifier] },
+        trivia, sym('='), trivia, { value: [value] },
+    ],
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ImportState) => _ParserState} */
-const parseImportFromOp = ({ token, metadata }) => state => {
-    switch (token.kind) {
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'string': {
-            const modules = concat(state.module.modules)([token.value])
-            return { ...state, state: 'nl', module: { ...state.module, modules: modules } }
-        }
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+const exportStatement = {
+    export: () => [sym('export'), trivia, sym('default'), trivia, { value: [value] }],
 }
 
-/** @type {(obj: _DjsStackObject) => (key: string) => _DjsStackObject} */
-const addKeyToObject = obj => key => (['object', obj[1], key])
+/**
+ * The whole module: every `import` before every `const`, one `export default`
+ * last, and nothing but trivia after it.
+ *
+ * The ordering the hand-written parser enforces with a `consts.length === 0`
+ * check is just the shape of this rule, which is the point of writing the
+ * grammar down: `import* const* export`.
+ *
+ * Ending on `eof` is what makes a trailing stray token a failure rather than a
+ * short match — the backend synthesizes that symbol after the physical input.
+ *
+ * @type {Rule}
+ */
+const djsModule = () => [
+    trivia,
+    repeat0Plus([importStatement, statementEnd]),
+    repeat0Plus([constStatement, statementEnd]),
+    exportStatement,
+    trivia,
+    eof,
+]
 
-/** @type {(obj: _DjsStackObject) => (value: AstConst) => _DjsStackObject} */
-const addValueToObject = obj => value => (['object', setReplace(obj[2])(value)(obj[1]), ''])
+/**
+ * The module matcher and the name of the rule to start it at.
+ *
+ * `toData` generates rule names, so the entry name belongs to the conversion and
+ * is read back from it rather than spelled here. Built once: converting the
+ * grammar and computing its nullability is per-grammar work, not per-parse.
+ */
+const [moduleRuleSet, moduleEntry] = toData(djsModule)
 
-/** @type {(array: _DjsStackArray) => (value: AstConst) => _DjsStackArray} */
-const addToArray = array => value => (['array', concat(array[1])([value])])
+/** @type {DescentMatch<DjsTokenWithMetadata>} */
+const moduleMatcher = descentParserRuleSet(moduleRuleSet)
 
 /**
  * The key of `{ __proto__: v }` and `{ "__proto__": v }`. JavaScript reads
@@ -269,91 +361,6 @@ const addToArray = array => value => (['array', concat(array[1])([value])])
  */
 const protoKey = '__proto__'
 
-/** @type {(valueState: _ValueState) => (state: _ParseValueState) => (key: string) => (metadata: TokenMetadata) => _ParserState} */
-const pushKey = valueState => state => key => metadata => {
-    if (state.top?.[0] === 'object') { return { ...state, valueState, top: addKeyToObject(state.top)(key), stack: state.stack } }
-    return { state: 'error', error: { message: 'error', metadata } }
-}
-
-/**
- * A key written as an identifier or a string literal, which is every key but
- * the computed one — so this is where `__proto__` is refused. A JSON document
- * spells that key the same way and means an ordinary property by it, but a
- * JSON document is not a module and this parser does not read one
- * ([spec: the `__proto__` key](../../../spec/README.md#the-__proto__-key)).
- *
- * @type {(state: _ParseValueState) => (key: string) => (metadata: TokenMetadata) => _ParserState}
- */
-const pushPlainKey = state => key => metadata => key === protoKey
-    ? { state: 'error', error: { message: '__proto__ requires the computed key form', metadata } }
-    : pushKey('{k')(state)(key)(metadata)
-
-/** @type {(state: _ParseValueState) => (value: AstConst) => _ParserState} */
-const pushValue = state => value => {
-    if (state.top === null) {
-        const consts = concat(state.module.consts)([value])
-        switch (state.state)
-        {
-            case 'exportValue': return { ...state, state: 'result', module: { ...state.module, consts: consts } }
-            case 'constValue': return { ...state, state: 'nl', module: { ...state.module, consts: consts } }
-        }
-    }
-    if (state.top?.[0] === 'array') { return { ...state, valueState: '[v', top: addToArray(state.top)(value), stack: state.stack } }
-    return { ...state, valueState: '{v', top: addValueToObject(state.top)(value), stack: state.stack }
-}
-
-/** @type {(state: _ParseValueState) => (name: string) => (metadata: TokenMetadata) => _ParserState} */
-const pushRef = state => name => metadata => {
-    const ref = at(name)(state.module.refs)
-    if (ref === null)
-        return { state: 'error', error: { message: 'const not found', metadata } }
-    return pushValue(state)(ref)
-}
-
-/** @type {(state: _ParseValueState) => _ParserState} */
-const startArray = state => {
-    const newStack = state.top === null ? null : { first: state.top, tail: state.stack }
-    return { ...state, valueState: '[', top: ['array', null], stack: newStack }
-}
-
-// Pops the enclosing container off `stack`. `next` is forced here rather than
-// left as a `drop(1)` thunk: the stack is written only by startArray/startObject,
-// always as a literal cons, and a lazy pop leaves one unforced thunk per closed
-// container — a chain that overflows the call stack when it is finally forced.
-/** @type {(state: _ParseValueState) => _ParseValueState} */
-const popStack = state => {
-    const ne = next(state.stack)
-    return ne === null
-        ? { ...state, valueState: '', top: null, stack: null }
-        : { ...state, valueState: '', top: ne.first, stack: ne.tail }
-}
-
-/** @type {(state: _ParseValueState) => _ParserState} */
-const endArray = state => {
-    const top = state.top
-    const newState = popStack(state)
-    if (top !== null && top[0] === 'array')
-    {
-        /** @type {AstArray} */
-        const array = ['array', toArray(top[1])]
-        return pushValue(newState)(array)
-    }
-    return pushValue(newState)(null)
-}
-
-/** @type {(state: _ParseValueState) => _ParserState} */
-const startObject = state => {
-    const newStack = state.top === null ? null : { first: state.top, tail: state.stack }
-    return { ...state, valueState: '{', top: ['object', null, ''], stack: newStack }
-}
-
-/** @type {(state: _ParseValueState) => _ParserState} */
-const endObject = state => {
-    const obj = state?.top !== null && state?.top[0] === 'object' ? fromMap(state.top[1]) : null
-    const newState = popStack(state)
-    return pushValue(newState)(obj)
-}
-
 /**
  * Only ever called on a token `isValueToken` has already confirmed carries a
  * value, so the switch covers every `_ValueToken` case with no fallback arm.
@@ -371,7 +378,6 @@ const tokenToValue = token => {
         case 'undefined': return undefined
     }
 }
-
 /**
  * @param {DjsToken} token
  * @returns {token is _ValueToken}
@@ -388,205 +394,338 @@ const isValueToken = token => {
         default: return false
     }
 }
+// -- folding the match into an `AstModule` ----------------------------------
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseValueOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case ']':
-            if (state.valueState === '[,') { return endArray(state) }
-            return { state: 'error', error: { message: 'unexpected token', metadata } }
-        case 'id': return pushRef(state)(token.value)(metadata)
-        case '[': return startArray(state)
-        case '{': return startObject(state)
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default:
-            if (isValueToken(token)) { return pushValue(state)(tokenToValue(token)) }
-            return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
-}
+/** @typedef {Ast<CodePointMeta<DjsTokenWithMetadata>>} _Node */
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseArrayStartOp = ({ token, metadata }) => state => {
-    if (isValueToken(token)) { return pushValue(state)(tokenToValue(token)) }
-    switch (token.kind)
-    {
-        case 'id': return pushRef(state)(token.value)(metadata)
-        case '[': return startArray(state)
-        case ']': return endArray(state)
-        case '{': return startObject(state)
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * The token a slot holds.
+ *
+ * Every slot the fold reads holds exactly one token, and it is always the
+ * leftmost leaf — a name, a module specifier, a primitive. Walking first
+ * children rather than searching keeps this total: there is no "not found" case
+ * to branch on.
+ *
+ * @type {(node: _Node) => DjsTokenWithMetadata}
+ */
+const tokenOf = node => {
+    const first = node.sequence[0]
+    return first instanceof Array ? first[1] : tokenOf(first)
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseArrayValueOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case ']': return endArray(state)
-        case ',': return { ...state, valueState: '[,', top: state.top, stack: state.stack }
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * A node's direct child carrying `tag`.
+ *
+ * Direct rather than recursive on purpose: a statement's `name` slot holds an
+ * identifier whose own tag may be `const` or `import`, so a search through the
+ * subtree would confuse a *word* with the statement spelling it.
+ *
+ * @type {(tag: string) => (node: _Node) => _Node}
+ */
+const slot = tag => node => {
+    const found = node.sequence.find(c => !(c instanceof Array) && c.tag === tag)
+    assert(found !== undefined && !(found instanceof Array), ['grammar slot missing', tag])
+    return found
 }
 
-// allow identifier property names (#2410)
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectStartOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case 'string':
-        case 'id':
-            return pushPlainKey(state)(token.value)(metadata)
-        case '[': return { ...state, valueState: '{[' }
-        case '}': return endObject(state)
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
+/**
+ * Every node tagged `tag` under `node`, in document order.
+ *
+ * An array's elements are not its direct children — they sit inside the option
+ * and repetition scaffolding `delimited` builds — so finding them takes a
+ * search rather than a lookup.
+ *
+ * The search cannot stray into a nested value, and needs no guard saying so:
+ * every element, member and statement is wrapped in a node carrying its own
+ * tag, so the wrapper matches and the search stops there, before it could
+ * descend into the array or object inside it. That is what the wrappers are
+ * for.
+ *
+ * Iterative, over an explicit stack, for the same reason {@link foldValue} is.
+ * A repetition is only *flat* in the AST when `toData` recognizes the
+ * right-recursive shape and emits a `Repeat`; nested inside this grammar's
+ * option scaffolding it does not, so a thousand siblings are a thousand levels
+ * of tree, and recursing over them overflows exactly as deep nesting would.
+ *
+ * @type {(tag: string) => (node: _Node) => readonly _Node[]}
+ */
+const descendantsTagged = tag => root => {
+    /** Pushes a node's children so the leftmost is visited first. */
+    /** @type {(rest: List<_Node>, sequence: AstSequence<CodePointMeta<DjsTokenWithMetadata>>) => List<_Node>} */
+    const pushChildren = (rest, sequence) => {
+        let stack = rest
+        let i = sequence.length
+        while (i !== 0) {
+            i = i - 1
+            const child = sequence[i]
+            if (!(child instanceof Array)) { stack = { first: child, tail: stack } }
+        }
+        return stack
     }
-}
-
-// computed property keys with a constant string key (#2470)
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectComputedKeyOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case 'string': return pushKey('{[k')(state)(token.value)(metadata)
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
+    /** @type {List<_Node>} */
+    let found = null
+    let stack = pushChildren(null, root.sequence)
+    for (;;) {
+        const top = next(stack)
+        if (top === null) { return toArray(found) }
+        const node = top.first
+        if (node.tag === tag) {
+            found = concat(found)([node])
+            stack = top.tail
+        } else {
+            stack = pushChildren(top.tail, node.sequence)
+        }
     }
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectComputedKeyEndOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case ']': return { ...state, valueState: '{k' }
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+const valueSlot = slot('value')
+
+const nameSlot = slot('name')
+
+const moduleSlot = slot('module')
+
+const keySlot = slot('key')
+
+const itemsOf = descendantsTagged('item')
+
+const membersOf = descendantsTagged('member')
+
+/**
+ * The property name a key spells, and whether it was the computed spelling.
+ *
+ * The distinction exists for `__proto__` alone: JavaScript reads a bare or
+ * string `__proto__` as an instruction to replace the prototype, while
+ * `{ ["__proto__"]: v }` denotes an ordinary property — so only the spelling
+ * separates a rejected key from an accepted one.
+ *
+ * @type {(node: _Node) => readonly[string, boolean]}
+ */
+const keyOf = node => {
+    const spelling = node.sequence[0]
+    assert(!(spelling instanceof Array), 'a key held no spelling')
+    const computed = spelling.tag === 'computed'
+    const { token } = tokenOf(computed ? nameSlot(spelling) : spelling)
+    assert('value' in token && typeof token.value === 'string', 'a key token carried no name')
+    return [token.value, computed]
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectKeyOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case ':': return { ...state, valueState: '{:', top: state.top, stack: state.stack }
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * A fold in progress: the names bound so far, the module specifiers and the
+ * body collected so far, and the first error if one has been met.
+ *
+ * The error rides in the state rather than wrapping every step in a `Result`,
+ * so a step reads as one expression instead of a nested match. Once set it is
+ * never replaced, which is what makes the reported error the *first* one.
+ *
+ * @typedef {{
+ *   readonly refs: OrderedMap<AstModuleRef>
+ *   readonly modules: readonly string[]
+ *   readonly consts: readonly AstConst[]
+ *   readonly error: ParseError | null
+ * }} _FoldState
+ */
+
+/** @type {(message: string) => (token: DjsTokenWithMetadata) => ParseError} */
+const foldError = message => ({ metadata }) => ({ message, metadata })
+
+/**
+ * Binds a name to a reference, rejecting one already bound.
+ *
+ * `import` and `const` share one map, so a name taken by either is taken for
+ * both — the same rule the state machine gets from consulting one `refs`.
+ *
+ * @type {(state: _FoldState) => (node: _Node) => (ref: AstModuleRef) => _FoldState}
+ */
+const bind = state => node => ref => {
+    const withMetadata = tokenOf(nameSlot(node))
+    const { token } = withMetadata
+    assert('value' in token && typeof token.value === 'string', 'a name token carried no name')
+    return at(token.value)(state.refs) !== null
+        ? { ...state, error: foldError('duplicate id')(withMetadata) }
+        : { ...state, refs: setReplace(token.value)(ref)(state.refs) }
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectColonOp = ({ token, metadata }) => state => {
-    if (isValueToken(token)) { return pushValue(state)(tokenToValue(token)) }
-    switch (token.kind)
-    {
-        case 'id': return pushRef(state)(token.value)(metadata)
-        case '[': return startArray(state)
-        case '{': return startObject(state)
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * A frame of {@link foldValue}'s explicit stack: the container being built, the
+ * element nodes still to read, and what has been built so far.
+ *
+ * `done` is a `List` rather than an array because a frame gains one element at a
+ * time: appending to an array per element would copy the whole prefix each time,
+ * which is what makes the obvious spelling quadratic in an array's length.
+ *
+ * @typedef {{
+ *   readonly items: readonly _Node[]
+ *   readonly index: number
+ *   readonly array: List<AstConst>
+ *   readonly object: OrderedMap<AstConst>
+ *   readonly keys: readonly(readonly[string, boolean])[]
+ *   readonly isArray: boolean
+ * }} _FoldFrame
+ */
+
+/**
+ * The error a frame's current key earns, or `null`.
+ *
+ * Checked as each member is reached rather than by scanning every key first, so
+ * that an earlier member's failure is reported before a later key's. Scanning
+ * ahead reported `__proto__` in `{a: missing, __proto__: 1}`, where the parser
+ * this replaces reports the unresolved `missing` — errors are first-to-last, and
+ * a key is not special enough to jump the queue.
+ *
+ * @type {(frame: _FoldFrame) => ParseError | null}
+ */
+const badKey = frame => {
+    if (frame.isArray) { return null }
+    const [name, computed] = frame.keys[frame.index]
+    return name === protoKey && !computed
+        // at the key itself, not at the object's `{`
+        ? foldError('__proto__ requires the computed key form')(tokenOf(keySlot(frame.items[frame.index])))
+        : null
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectNextOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case '}': return endObject(state)
-        case ',': return { ...state, valueState: '{,', top: state.top, stack: state.stack }
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
+/**
+ * A value, resolved against the names bound so far.
+ *
+ * Iterative, over an explicit stack, because a value nests arbitrarily and the
+ * call stack does not: recursion here overflows at a few thousand containers,
+ * which is the defect `containerStackCost` was written to catch when the parser
+ * this replaced had its own version of it.
+ *
+ * Returns the error channel alongside the value because a reference can fail to
+ * resolve at any depth, and a container has to stop building when one does. A
+ * failed fold yields `null` for the value, which is never mistaken for a
+ * successful `null` — the caller reads the error, not the value.
+ *
+ * @type {(state: _FoldState) => (node: _Node) => readonly[AstConst, ParseError | null]}
+ */
+const foldValue = state => root => {
+    /** @type {List<_FoldFrame>} */
+    let stack = null
+    let node = root
+    /** @type {AstConst} */
+    let value = null
+    // `true` while descending into `node`; `false` while handing `value` back
+    // to the frame that asked for it.
+    let descending = true
+    for (;;) {
+        if (descending) {
+            const child = node.sequence[0]
+            assert(!(child instanceof Array), 'a value slot held no value')
+            if (child.tag === 'array' || child.tag === 'object') {
+                const isArray = child.tag === 'array'
+                const items = isArray ? itemsOf(child) : membersOf(child)
+                const keys = isArray ? [] : items.map(member => keyOf(keySlot(member)))
+                /** @type {_FoldFrame} */
+                const frame = { items, index: 0, array: null, object: null, keys, isArray }
+                stack = { first: frame, tail: stack }
+                if (items.length === 0) {
+                    value = isArray ? ['array', []] : fromMap(null)
+                    stack = assertNotNullish(next(stack)).tail
+                    descending = false
+                } else {
+                    const rejected = badKey(frame)
+                    if (rejected !== null) { return [null, rejected] }
+                    node = isArray ? items[0] : valueSlot(items[0])
+                }
+            } else {
+                const withMetadata = tokenOf(child)
+                const { token } = withMetadata
+                if (isValueToken(token)) {
+                    value = tokenToValue(token)
+                } else {
+                    // anything else the value rule admits is an identifier, so
+                    // it names a `const` or an `import` — or nothing, which is
+                    // the error.
+                    assert('value' in token && typeof token.value === 'string', 'a reference carried no name')
+                    const ref = at(token.value)(state.refs)
+                    if (ref === null) { return [null, foldError('const not found')(withMetadata)] }
+                    value = ref
+                }
+                descending = false
+            }
+        } else {
+            /** @type {{ readonly first: _FoldFrame, readonly tail: List<_FoldFrame> } | null} */
+            const top = next(stack)
+            if (top === null) { return [value, null] }
+            /** @type {_FoldFrame} */
+            const frame = top.first
+            const index = frame.index + 1
+            const array = frame.isArray ? concat(frame.array)([value]) : frame.array
+            const object = frame.isArray
+                ? frame.object
+                : setReplace(frame.keys[frame.index][0])(value)(frame.object)
+            if (index === frame.items.length) {
+                /** @type {AstArray} */
+                const asArray = ['array', toArray(array)]
+                /** @type {AstObject} */
+                const asObject = fromMap(object)
+                value = frame.isArray ? asArray : asObject
+                stack = top.tail
+            } else {
+                const moved = { ...frame, index, array, object }
+                const rejected = badKey(moved)
+                if (rejected !== null) { return [null, rejected] }
+                stack = { first: moved, tail: top.tail }
+                node = frame.isArray ? frame.items[index] : valueSlot(frame.items[index])
+                descending = true
+            }
+        }
     }
 }
 
-/** @type {(token: DjsTokenWithMetadata) => (state: _ParseValueState) => _ParserState} */
-const parseObjectCommaOp = ({ token, metadata }) => state => {
-    switch (token.kind)
-    {
-        case '}': return endObject(state)
-        case 'string':
-        case 'id':
-            return pushPlainKey(state)(token.value)(metadata)
-        case '[': return { ...state, valueState: '{[' }
-        case 'ws':
-        case 'nl':
-        case '//':
-        case '/*': return state
-        case 'eof': return { state: 'error', error: { message: 'unexpected end', metadata } }
-        default: return { state: 'error', error: { message: 'unexpected token', metadata } }
-    }
+/**
+ * Adds one statement's value to the body, if nothing has failed yet.
+ *
+ * @type {(state: _FoldState) => (node: _Node) => _FoldState}
+ */
+const addValue = state => node => {
+    if (state.error !== null) { return state }
+    const [value, valueError] = foldValue(state)(valueSlot(node))
+    return valueError !== null
+        ? { ...state, error: valueError }
+        : { ...state, consts: [...state.consts, value] }
 }
 
-/** @type {Fold<DjsTokenWithMetadata, _ParserState>} */
-const foldOp = token => state => {
-    switch (state.state) {
-        case '': return parseInitialOp(token)(state)
-        case 'nl': return parseNewLineRequiredOp(token)(state)
-        case 'import': return parseImportOp(token)(state)
-        case 'import+name': return parseImportNameOp(token)(state)
-        case 'import+from': return parseImportFromOp(token)(state)
-        case 'const': return parseConstOp(token)(state)
-        case 'const+name': return parseConstNameOp(token)(state)
-        case 'export': return parseExportOp(token)(state)
-        case 'result': return parseResultOp(token)(state)
-        case 'error': return { state: 'error', error: state.error }
-        case 'constValue':
-        case 'exportValue':
-        {
-            switch (state.valueState)
-            {
-                case '': return parseValueOp(token)(state)
-                case '[': return parseArrayStartOp(token)(state)
-                case '[v': return parseArrayValueOp(token)(state)
-                case '[,': return parseValueOp(token)(state)
-                case '{': return parseObjectStartOp(token)(state)
-                case '{[': return parseObjectComputedKeyOp(token)(state)
-                case '{[k': return parseObjectComputedKeyEndOp(token)(state)
-                case '{k': return parseObjectKeyOp(token)(state)
-                case '{:': return parseObjectColonOp(token)(state)
-                case '{v': return parseObjectNextOp(token)(state)
-                case '{,': return parseObjectCommaOp(token)(state)
-            }
-        }
+/**
+ * Folds a matched module into an `AstModule`.
+ *
+ * The statements are read positionally from the root — trivia, imports,
+ * consts, the export — because the module rule is one sequence and its parts
+ * cannot move. Everything below that is read by slot name instead.
+ *
+ * Each name is bound *before* the value that follows it is folded, which is
+ * what makes `const a = a` resolve to the constant being defined rather than
+ * fail. That is the state machine's order too, and the reason it is not one of
+ * the divergences.
+ *
+ * @type {(root: _Node) => Result<AstModule, ParseError>}
+ */
+const foldModule = root => {
+    const [, imports, consts, exported] = root.sequence
+    assert(!(imports instanceof Array) && !(consts instanceof Array) && !(exported instanceof Array),
+        'the module rule did not produce its statement groups')
+    /** @type {_FoldState} */
+    let state = { refs: null, modules: [], consts: [], error: null }
+    for (const statement of descendantsTagged('import')(imports)) {
+        state = bind(state)(statement)(['aref', state.modules.length])
+        if (state.error !== null) { break }
+        const specifier = tokenOf(moduleSlot(statement))
+        assert('value' in specifier.token && typeof specifier.token.value === 'string',
+            'an import specifier carried no text')
+        state = { ...state, modules: [...state.modules, specifier.token.value] }
+    }
+    for (const statement of descendantsTagged('const')(consts)) {
+        if (state.error !== null) { break }
+        state = bind(state)(statement)(['cref', state.consts.length])
+        state = addValue(state)(statement)
     }
+    state = addValue(state)(exported)
+    if (state.error !== null) { return error(state.error) }
+    // annotated rather than inferred: a bare `[modules, consts]` widens to an
+    // array, because `readonly string[]` is itself assignable to `AstBody`.
+    /** @type {AstModule} */
+    const astModule = [state.modules, state.consts]
+    return ok(astModule)
 }
 
 /**
@@ -597,70 +736,138 @@ const foldOp = token => state => {
  * module, and `fjs/media/json` is its reader
  * ([spec: JSON input](../../../spec/README.md#json-input)).
  *
+ * The grammar it accepts is written down, in the rules above, rather than
+ * implied by the control flow of a state machine — which is what this replaced.
+ *
  * @type {(tokenList: List<DjsTokenWithMetadata>) => Result<AstModule, ParseError>}
  */
 export const parseFromTokens = tokenList => {
-    const state = fold(foldOp)({ state: '', module: { refs: null, modules: null, consts: null } })(tokenList)
-    switch (state.state) {
-        case 'result': return ok(/** @satisfies {AstModule} */ ([toArray(state.module.modules), toArray(state.module.consts)]))
-        case 'error': return error(state.error)
-        default: return error({ message: 'unexpected end', metadata: null })
+    const [tag, stream] = splitEof(toArray(tokenList))
+    if (tag === 'error') { return error(stream) }
+    const { tokens, eofMetadata } = stream
+    const { ast, success, failure } = moduleMatcher(moduleEntry, tokens.map(tokenToSymbol))
+    if (!success) {
+        const { idx } = assertNotNullish(failure)
+        // A failure past the last token is the end of input rather than a token
+        // the reader can point at, and the hand-written parser words it that
+        // way; matching it costs one comparison already being made.
+        const atEnd = idx >= tokens.length
+        return error({
+            message: atEnd ? 'unexpected end' : 'unexpected token',
+            metadata: atEnd ? eofMetadata : tokens[idx].metadata,
+        })
     }
+    return foldModule(ast)
 }
 
+/** @type {(kind: 'eof' | ',') => (line: number) => DjsTokenWithMetadata} */
+const proofToken = kind => line => ({ token: { kind }, metadata: { path: 'a.js', line, column: 1 } })
+
+const proofEof = proofToken('eof')
+
+const proofComma = proofToken(',')
+
 export const proof = {
-    pushKey: {
-        // `pushKey` is only ever invoked while `state.top` is an object (the
-        // `'{'`/`'{,'` value-states guarantee it), so its non-object guard is
-        // a defensive branch unreachable through `parseFromTokens`. Call it
-        // directly to cover that branch.
-        nonObjectTop: () => {
-            /** @type {_ParseValueState} */
-            const state = {
-                state: 'exportValue',
-                module: { refs: null, modules: null, consts: null },
-                valueState: '[',
-                top: null,
-                stack: null,
-            }
-            const result = pushKey('{k')(state)('key')({ path: 'test', line: 0, column: 0 })
-            assertEq(result.state, 'error')
+    ordinaryTokenNames: {
+        // `_AlphabetIsComplete` pins membership at compile time, but a repeated
+        // name widens to the same union and so is invisible to it. The check
+        // matters because the token-symbol mapping this alphabet feeds has to be
+        // injective over it — two entries for one name would break that.
+        noDuplicates: () => {
+            assertEq(new Set(ordinaryTokenNames).size, ordinaryTokenNames.length)
         },
     },
-    endArray: {
-        // `endArray` is only ever invoked while `state.top` is an array (the
-        // `'['`/`'[v'`/`'[,'` value-states guarantee it), so its non-array
-        // guard is a defensive branch unreachable through `parseFromTokens`.
-        // Call it directly to cover that branch.
-        nonArrayTop: () => {
-            /** @type {_ParseValueState} */
-            const state = {
-                state: 'exportValue',
-                module: { refs: null, modules: null, consts: null },
-                valueState: '[,',
-                top: null,
-                stack: null,
-            }
-            const result = endArray(state)
-            assertEq(result.state, 'result')
+    tokenToSymbol: {
+        // One symbol per token, distinct across the alphabet, and above every
+        // code point — the three properties that let a token stream be the
+        // alphabet of the layer above.
+        distinctAndAboveUnicode: () => {
+            const symbols = ordinaryTokenNames.map(n => tokenEncoding.encode(n))
+            assertEq(new Set(symbols).size, ordinaryTokenNames.length)
+            const [, unicodeLast] = rangeDecode(unicodeRange)
+            assert(symbols.every(s => s > unicodeLast), JSON.stringify(symbols))
+        },
+        // The distinction the hand-written parser makes by comparing
+        // `token.value`: a framing keyword and an ordinary identifier arrive as
+        // the same `id` kind, and the grammar can only tell them apart if they
+        // get different symbols here.
+        framingKeywordsAreNotIdentifiers: () => {
+            /** @type {(value: string) => number} */
+            const symbolOf = value =>
+                tokenToSymbol({ token: { kind: 'id', value }, metadata: { path: 'a.js', line: 1, column: 1 } })[0]
+            const id = symbolOf('foo')
+            const keywords = framingKeywords.map(symbolOf)
+            assert(keywords.every(s => s !== id), JSON.stringify([id, keywords]))
+            assertEq(new Set(keywords).size, framingKeywords.length)
+            assertEq(tokenEncoding.decode(symbolOf('export')), 'export')
+            assertEq(tokenEncoding.decode(id), 'id')
+        },
+        // The token rides along untouched, so a fold or a diagnostic above still
+        // has its value and its position.
+        carriesTheToken: () => {
+            /** @type {DjsTokenWithMetadata} */
+            const t = { token: { kind: 'string', value: 'v' }, metadata: { path: 'a.js', line: 7, column: 1 } }
+            const [symbol, meta] = tokenToSymbol(t)
+            assertEq(meta, t)
+            assertEq(meta.metadata.line, 7)
+            assertEq(tokenEncoding.decode(symbol), 'string')
+        },
+        throw: {
+            eofRejected: () => tokenToSymbol(proofEof(1)),
         },
     },
-    endObject: {
-        // `endObject` is only ever invoked while `state.top` is an object
-        // (the `'{'`/`'{k'`/`'{:'`/`'{v'`/`'{,'` value-states guarantee it),
-        // so its non-object guard is a defensive branch unreachable through
-        // `parseFromTokens`. Call it directly to cover that branch.
-        nonObjectTop: () => {
-            /** @type {_ParseValueState} */
-            const state = {
-                state: 'exportValue',
-                module: { refs: null, modules: null, consts: null },
-                valueState: '{,',
-                top: null,
-                stack: null,
+    splitEof: {
+        // The tokenizer always ends its stream with one `eof`, so every branch
+        // but this one is reachable only from a hand-built token list.
+        final: () => {
+            const [tag, value] = splitEof([proofComma(1), proofEof(2)])
+            assert(tag === 'ok', tag)
+            assertEq(value.tokens.length, 1)
+            assertEq(value.eofMetadata.line, 2)
+        },
+        onlyEof: () => {
+            const [tag, value] = splitEof([proofEof(1)])
+            assert(tag === 'ok', tag)
+            assertEq(value.tokens.length, 0)
+            assertEq(value.eofMetadata.line, 1)
+        },
+        missing: () => {
+            const [tag, value] = splitEof([proofComma(1)])
+            assert(tag === 'error', tag)
+            assertEq(value.metadata, null)
+        },
+        // An empty stream is also missing its `eof`, and has not even a last
+        // token to blame it on. The tokenizer never produces one — an empty
+        // source still yields `eof` — so only a hand-built list reaches here.
+        empty: () => {
+            const [tag, value] = splitEof(null)
+            assert(tag === 'error', tag)
+            assertEq(value.metadata, null)
+        },
+        // The one stream with no `eof` that is not a broken contract: a lexical
+        // failure stops the tokenizer at an `error` token. The position has to
+        // survive, or "unterminated string at 1:11" would be reported as
+        // "missing end-of-input token" with nowhere to point.
+        lexicalError: () => {
+            /** @type {DjsTokenWithMetadata} */
+            const errorToken = {
+                token: { kind: 'error', message: 'unterminated string literal' },
+                metadata: { path: 'a.js', line: 3, column: 7 },
             }
-            const result = endObject(state)
-            assertEq(result.state, 'result')
+            const [tag, value] = splitEof([errorToken])
+            assert(tag === 'error', tag)
+            assertEq(value.metadata?.line, 3)
+            assertEq(value.metadata?.column, 7)
+        },
+        notFinal: () => {
+            const [tag, value] = splitEof([proofEof(1), proofComma(2)])
+            assert(tag === 'error', tag)
+            assertEq(value.metadata?.line, 1)
+        },
+        duplicate: () => {
+            const [tag, value] = splitEof([proofEof(1), proofEof(2)])
+            assert(tag === 'error', tag)
+            assertEq(value.metadata?.line, 1)
         },
     },
 }