@thomasfosterau/effect-parse 0.0.0-stage → 0.1.0

package/src/internal/compiler/entries.ts

@@ -0,0 +1,36 @@
+/**
+ * The shapes of `map(p, Object.fromEntries)` that the compilers build
+ * directly (see `internal/entries.ts`).
+ */
+import type { Node, NodeOf, ParserLike } from "../node.js"
+import { nodeOf } from "../node.js"
+
+/** A repeated pair for `Object.fromEntries`: the sequence and the indices of its key and value. */
+export const pairOf = (p: ParserLike): { readonly parsers: ReadonlyArray<ParserLike>; readonly key: number; readonly value: number } | undefined => {
+  let node = nodeOf(p)
+  while (isDiagnostic(node)) node = nodeOf(node.self)
+  if (node._tag !== "Seq") return undefined
+  const output = node.output
+  if (output === "tuple") return node.parsers.length === 2 ? { parsers: node.parsers, key: 0, value: 1 } : undefined
+  if (typeof output === "number" || output[0] === output[1]) return undefined
+  return { parsers: node.parsers, key: output[0], value: output[1] }
+}
+
+const isDiagnostic = (
+  node: Node
+): node is NodeOf<"Context" | "Expected" | "MapError" | "OrFail" | "Trace" | "RetryAfter"> =>
+  node._tag === "Context" || node._tag === "Expected" || node._tag === "MapError" || node._tag === "OrFail" ||
+  node._tag === "Trace" || node._tag === "RetryAfter"
+
+/**
+ * Whether `map(p, Object.fromEntries)` can build its object directly: `p`
+ * is a `repeat` or `separated` of a pair, possibly inside `cut`, the kept
+ * element of a sequence (`preceded`, `terminated`, `delimited`) and
+ * diagnostics.
+ */
+export const entriesShape = (p: ParserLike): boolean => {
+  const node = nodeOf(p)
+  if (isDiagnostic(node) || node._tag === "Mode") return entriesShape(node.self)
+  if (node._tag === "Seq") return typeof node.output === "number" && entriesShape(node.parsers[node.output]!)
+  return (node._tag === "Repeat" || node._tag === "Separated") && pairOf(node.self) !== undefined
+}

package/src/internal/compiler/first.ts

@@ -0,0 +1,252 @@
+/**
+ * First sets: which input units can start a match of a parser.
+ *
+ * A first set has a mode:
+ *
+ * - `Strict`: when the next unit is not in `units` (or the input has
+ *   ended), the parser fails with `Backtrack`.
+ * - `Optional`: in that case, the parser either fails with `Backtrack` or
+ *   succeeds without consuming input.
+ * - `Unknown`: no guarantee.
+ *
+ * Units are UTF-16 code units for strings and bytes for byte inputs; units
+ * from `TableSize` up are summarized by `high`. Array inputs have no units.
+ *
+ * Skipping a parser that is certain to fail also skips the callbacks it
+ * would have run before failing, which is why callbacks must be pure.
+ */
+import type { Encoding } from "../../Encoding.js"
+import type { Node, ParserLike } from "../node.js"
+import { nodeOf } from "../node.js"
+import { fastLimit } from "../text.js"
+import type { CharSet } from "./sets.js"
+import { kindOfEncoding, reachesAbove, TableSize, textSet, tokenSet } from "./sets.js"
+
+export const Strict = 0
+export const Optional = 1
+export const Unknown = 2
+
+export interface First {
+  readonly mode: 0 | 1 | 2
+  readonly units: Uint8Array
+  readonly high: boolean
+}
+
+const none = new Uint8Array(TableSize)
+const all = new Uint8Array(TableSize).fill(1)
+
+const unknown: First = { mode: Unknown, units: all, high: true }
+const emptyOptional: First = { mode: Optional, units: none, high: false }
+const failing: First = { mode: Strict, units: none, high: false }
+const anyUnit: First = { mode: Strict, units: all, high: true }
+
+const union = (a: First, b: First, mode: 0 | 1 | 2): First => {
+  const units = new Uint8Array(TableSize)
+  for (let i = 0; i < TableSize; i++) units[i] = a.units[i]! | b.units[i]!
+  return { mode, units, high: a.high || b.high }
+}
+
+const withMode = (f: First, mode: 0 | 1 | 2): First => f.mode === mode ? f : { ...f, mode }
+
+const lowerAscii = (c: number): number => c >= 65 && c <= 90 ? c + 32 : c
+const upperAscii = (c: number): number => c >= 97 && c <= 122 ? c - 32 : c
+
+const ofUnits = (codes: ReadonlyArray<number>, mode: 0 | 1 | 2 = Strict): First => {
+  const units = new Uint8Array(TableSize)
+  let high = false
+  for (const c of codes) {
+    if (c < TableSize) units[c] = 1
+    else high = true
+  }
+  return { mode, units, high }
+}
+
+export class FirstSets {
+  private readonly memo = new Map<ParserLike, First | undefined>()
+  constructor(readonly kind: 0 | 1 | 2) {}
+
+  of(p: ParserLike): First {
+    if (this.kind === 2) return unknown
+    if (this.memo.has(p)) return this.memo.get(p) ?? unknown // in progress: a cycle
+    this.memo.set(p, undefined)
+    const f = this.compute(nodeOf(p))
+    this.memo.set(p, f)
+    return f
+  }
+
+  /** Units of the code points of a set, as an encoding (or `Token`) reads them. */
+  private setUnits(set: CharSet, enc: Encoding<any> | undefined, mode: 0 | 1): First | undefined {
+    if (enc === undefined || kindOfEncoding(enc) === 0) {
+      if (this.kind === 1) return { mode, units: set.table, high: false }
+      return { mode, units: set.table, high: reachesAbove(set, TableSize) }
+    }
+    if (enc.unit !== 1) return undefined
+    const fast = fastLimit(enc)
+    if (fast === 0) return undefined
+    const units = set.table.slice()
+    // Multi-unit sequences start with a unit at or above `fast`
+    if (fast < TableSize) units.fill(reachesAbove(set, fast) ? 1 : 0, fast)
+    return { mode, units, high: false }
+  }
+
+  private compute(node: Node): First {
+    const kind = this.kind as 0 | 1
+    switch (node._tag) {
+      case "Succeed":
+      case "Sync":
+      case "Location":
+      case "RestLen":
+      case "Eof":
+        return emptyOptional
+      case "Fail":
+        return node.mode === "Backtrack" ? failing : unknown
+      case "Any":
+        return node.encoding !== undefined && kindOfEncoding(node.encoding) !== kind ? unknown : anyUnit
+      case "Take":
+        return node.count === 0 ? emptyOptional : anyUnit
+      case "Literal": {
+        const lit = node.literal
+        if (node.encoding !== undefined) {
+          if (kindOfEncoding(node.encoding) !== kind) return unknown
+          if (node.caseless) return lit.length === 0 ? unknown : anyUnit
+          const encoded = node.encoding.encode(lit as string)
+          if (encoded.length === 0) return emptyOptional
+          return ofUnits([typeof encoded === "string" ? encoded.charCodeAt(0) : encoded[0]!])
+        }
+        if (lit.length === 0) return emptyOptional
+        let first: number
+        if (typeof lit === "string") {
+          first = kind === 0 ? lit.charCodeAt(0) : new TextEncoder().encode(lit)[0]!
+        } else if (lit instanceof Uint8Array && kind === 1) {
+          first = lit[0]!
+        } else {
+          return unknown
+        }
+        return ofUnits(node.caseless ? [first, lowerAscii(first), upperAscii(first)] : [first])
+      }
+      case "OneOf":
+      case "TakeWhile": {
+        if (node.encoding !== undefined && kindOfEncoding(node.encoding) !== kind) return unknown
+        const min = node._tag === "OneOf" ? 1 : node.min
+        if (node._tag === "TakeWhile" && node.max === 0) return emptyOptional
+        const mode = min > 0 ? Strict : Optional
+        if (node.negate) return mode === Strict ? anyUnit : withMode(anyUnit, Optional)
+        const set = node.encoding === undefined ? tokenSet(node.set, kind) : textSet(node.set)
+        return this.setUnits(set, node.encoding, mode) ?? (mode === Strict ? anyUnit : unknown)
+      }
+      case "Map":
+      case "As":
+      case "Tap":
+      case "TryMap":
+      case "FilterMap":
+      case "Filter":
+      case "Taken":
+      case "WithTaken":
+      case "Span":
+      case "WithSpan":
+      case "Context":
+      case "Expected":
+      case "MapError":
+      case "OrFail":
+      case "Peek":
+      case "Trace":
+      case "ResumeAfter":
+      case "RetryAfter":
+        return this.of(node.self)
+      case "FlatMap":
+      case "SeparatedFold":
+      case "Dispatch": {
+        const f = this.of(node._tag === "Dispatch" ? node.selector : node.self)
+        return f.mode === Strict ? f : unknown
+      }
+      case "Mode": {
+        const f = this.of(node.self)
+        return node.from === "Backtrack" ? (f.mode === Unknown ? f : unknown) : f
+      }
+      case "Opt":
+      case "OrElseSucceed": {
+        const f = this.of(node.self)
+        return f.mode === Unknown ? f : withMode(f, Optional)
+      }
+      case "Seq":
+      case "Struct":
+        return this.sequence(node.parsers)
+      case "ZipWith":
+        return this.sequence([node.self, node.that])
+      case "Alt": {
+        if (node.parsers.length === 0) return failing
+        let acc: First | undefined
+        let mode: 0 | 1 | 2 = Strict
+        for (const p of node.parsers) {
+          const f = this.of(p)
+          if (f.mode === Unknown) return unknown
+          if (f.mode === Optional) mode = Optional
+          acc = acc === undefined ? f : union(acc, f, mode)
+        }
+        return withMode(acc!, mode)
+      }
+      case "Repeat":
+      case "Fold":
+      case "Separated": {
+        if (node.max === 0) return emptyOptional
+        const f = this.of(node.self)
+        if (f.mode !== Strict) return unknown
+        return node.min === 0 ? withMode(f, Optional) : f
+      }
+      case "Suspend":
+        return this.of(node.resolve())
+      default:
+        return unknown
+    }
+  }
+
+  private sequence(parsers: ReadonlyArray<ParserLike>): First {
+    let acc: First | undefined
+    for (const p of parsers) {
+      const f = this.of(p)
+      if (f.mode === Unknown) return unknown
+      acc = acc === undefined ? f : union(acc, f, f.mode)
+      if (f.mode === Strict) return acc
+    }
+    return acc ?? emptyOptional
+  }
+}
+
+/**
+ * For an `Alt`: the branches worth trying for each next unit, each next
+ * high unit, and at the end of the input, or `undefined` when pruning
+ * would never skip a branch.
+ */
+export interface AltPlan {
+  /** Indexed by unit: indexes into `lists`. */
+  readonly byUnit: Uint8Array
+  readonly high: number
+  readonly end: number
+  readonly lists: ReadonlyArray<ReadonlyArray<number>>
+}
+
+export const planAlt = (firsts: ReadonlyArray<First>): AltPlan | undefined => {
+  const n = firsts.length
+  if (n < 2 || n > 64) return undefined
+  const lists: Array<ReadonlyArray<number>> = []
+  const index = new Map<string, number>()
+  const listOf = (keep: (f: First) => boolean): number => {
+    const list: Array<number> = []
+    for (let i = 0; i < n; i++) if (keep(firsts[i]!)) list.push(i)
+    const key = list.join(",")
+    let k = index.get(key)
+    if (k === undefined) {
+      k = lists.length
+      lists.push(list)
+      index.set(key, k)
+    }
+    return k
+  }
+  const byUnit = new Uint8Array(TableSize)
+  for (let u = 0; u < TableSize; u++) byUnit[u] = listOf((f) => f.mode !== Strict || f.units[u] === 1)
+  const high = listOf((f) => f.mode !== Strict || f.high)
+  const end = listOf((f) => f.mode !== Strict)
+  if (lists.length === 1 && lists[0]!.length === n) return undefined
+  if (lists.length > 255) return undefined
+  return { byUnit, high, end, lists }
+}