polytypo 1.2.0 → 1.3.0

package/README.md

@@ -57,6 +57,7 @@ the parsers that mode needs:
 | `polytypo/text`     | none                                                          |
 | `polytypo/html`     | parse5                                                        |
 | `polytypo/markdown` | parse5, micromark and its GFM, frontmatter and MDX extensions |
+| `polytypo/yaml`     | none                                                          |
 | `polytypo`          | all of the above                                              |
 
 HTML and Markdown are first-class modes, not an afterthought — tags, attributes and fenced code
@@ -69,6 +70,53 @@ transform(`<a title="test... wait">Wait... she said "go on."</a>`, { locale: "en
 // <a title="test... wait">Wait… she said “go on.”</a>
 ```
 
+`yaml` mode is the one that asks something of you, and it asks for a reason. YAML is a data format
+with prose in some of it, so you name the keys whose values are prose; there is no default and no
+guess:
+
+```ts
+import { transform } from "polytypo/yaml";
+
+transform("summary: Rates -- all of them...\nrun: git diff -- a--b\n", {
+  locale: "en-US",
+  keys: ["summary"],
+});
+// summary: Rates—all of them…
+// run: git diff -- a--b
+```
+
+Nothing in YAML's syntax separates a sentence from a shell script: `description` holds one and
+`run` holds the other, spelled identically. Without `keys` the same document comes back with
+`if !` rewritten as `if!`. Quoting, indentation, anchors and a block scalar's chomping indicator
+are never decoded and rewritten — the file is located, not re-emitted — so the trailing newlines
+of a `|+` block come back exactly as you wrote them.
+
+In `markdown` mode the skipped regions are the dialect's own structural ones, not whatever looks
+like code. CommonMark counts an indented block as code at **four** spaces; at two it is an ordinary
+paragraph, so a JSON sample indented by two comes back with curly quotes — valid JSON in, invalid
+JSON out, and nothing in the return value says so. MDX has no indented code blocks at all, so there
+even four spaces is prose and a brace in it is a JSX expression: the same sample throws
+`POLYTYPO_MALFORMED_INPUT` instead. A fence or a code span is the one marker that means _code_ in
+both dialects.
+
+`analyze()` runs the same pipeline and reports what it would do instead of doing it — one record
+per edit, each with the rule that made it and code-point offsets into the input you passed (into
+the **document**, in `html`, `markdown` and `yaml` mode, not into a span):
+
+```ts
+import { analyze } from "polytypo";
+
+analyze(`Wait... "really"?`, { locale: "en-US" });
+// [ { ruleId: "ellipsis", start: 4, end: 7, before: "...", after: "…" },
+//   { ruleId: "quotes",   start: 8, end: 9, before: `"`,   after: "“" }, … ]
+```
+
+It is a report, not a patch. The list is empty exactly when `transform` would return the input
+unchanged, and every `ruleId` is a rule that was enabled for that call — but two rules may touch
+the same original range (French `spaces` deletes the space before `:` and `nbsp` puts a no-break
+one back), so replaying the list is not guaranteed to reproduce the output. Call `transform` for
+the text. Full contract: `spec/rules/analyze.md`.
+
 The aggregate `polytypo` entry defaults `mode` to `"text"` and is the one to use when the mode is
 chosen at runtime. `locale` has no default anywhere and must always be passed explicitly — there is
 no silent fallback to English.

package/dist/chunk-442V6TKI.cjs

@@ -0,0 +1,31 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});
+
+
+
+
+
+
+
+var _chunkYX6QTWOVcjs = require('./chunk-YX6QTWOV.cjs');
+
+// src/engine/text-pipeline.ts
+function runTextPipeline(input, options) {
+  const narrowTarget = _chunkYX6QTWOVcjs.resolveNarrowTarget.call(void 0, options.narrowNbsp);
+  const planned = _chunkYX6QTWOVcjs.planRules.call(void 0, options.rules);
+  const locale = _chunkYX6QTWOVcjs.getLocaleData.call(void 0, options.locale);
+  return _chunkYX6QTWOVcjs.fromCodePoints.call(void 0, _chunkYX6QTWOVcjs.runRules.call(void 0, _chunkYX6QTWOVcjs.toCodePoints.call(void 0, input), planned, locale, "text", narrowTarget));
+}
+function analyzeTextPipeline(input, options) {
+  const narrowTarget = _chunkYX6QTWOVcjs.resolveNarrowTarget.call(void 0, options.narrowNbsp);
+  const planned = _chunkYX6QTWOVcjs.planRules.call(void 0, options.rules);
+  const locale = _chunkYX6QTWOVcjs.getLocaleData.call(void 0, options.locale);
+  const cp = _chunkYX6QTWOVcjs.toCodePoints.call(void 0, input);
+  const origin = cp.map((_value, index) => index);
+  return _chunkYX6QTWOVcjs.runRulesRecording.call(void 0, cp, planned, locale, "text", narrowTarget, origin, cp.length);
+}
+
+
+
+
+exports.runTextPipeline = runTextPipeline; exports.analyzeTextPipeline = analyzeTextPipeline;
+//# sourceMappingURL=chunk-442V6TKI.cjs.map

package/dist/chunk-442V6TKI.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-442V6TKI.cjs","../src/engine/text-pipeline.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACA;ACQO,SAAS,eAAA,CAAgB,KAAA,EAAe,OAAA,EAAmC;AAChF,EAAA,MAAM,aAAA,EAAe,mDAAA,OAAoB,CAAQ,UAAU,CAAA;AAC3D,EAAA,MAAM,QAAA,EAAU,yCAAA,OAAU,CAAQ,KAAK,CAAA;AACvC,EAAA,MAAM,OAAA,EAAS,6CAAA,OAAc,CAAQ,MAAM,CAAA;AAC3C,EAAA,OAAO,8CAAA,wCAAe,4CAAS,KAAkB,CAAA,EAAG,OAAA,EAAS,MAAA,EAAQ,MAAA,EAAQ,YAAY,CAAC,CAAA;AAC5F;AAGO,SAAS,mBAAA,CAAoB,KAAA,EAAe,OAAA,EAAqC;AACtF,EAAA,MAAM,aAAA,EAAe,mDAAA,OAAoB,CAAQ,UAAU,CAAA;AAC3D,EAAA,MAAM,QAAA,EAAU,yCAAA,OAAU,CAAQ,KAAK,CAAA;AACvC,EAAA,MAAM,OAAA,EAAS,6CAAA,OAAc,CAAQ,MAAM,CAAA;AAC3C,EAAA,MAAM,GAAA,EAAK,4CAAA,KAAkB,CAAA;AAC7B,EAAA,MAAM,OAAA,EAAS,EAAA,CAAG,GAAA,CAAI,CAAC,MAAA,EAAQ,KAAA,EAAA,GAAU,KAAK,CAAA;AAC9C,EAAA,OAAO,iDAAA,EAAkB,EAAI,OAAA,EAAS,MAAA,EAAQ,MAAA,EAAQ,YAAA,EAAc,MAAA,EAAQ,EAAA,CAAG,MAAM,CAAA;AACvF;ADRA;AACA;AACE;AACA;AACF,6FAAC","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-442V6TKI.cjs","sourcesContent":[null,"import type { Options } from \"../types.js\";\nimport { fromCodePoints, toCodePoints } from \"./codepoints.js\";\nimport type { Change } from \"./origin.js\";\nimport { getLocaleData } from \"./locale.js\";\nimport { resolveNarrowTarget } from \"./narrow-target.js\";\nimport { planRules, runRules, runRulesRecording } from \"./rule-runner.js\";\n\n/**\n * `text` mode only. Deliberately imports nothing from `../modes/html.js` or\n * `../modes/markdown.js` (or their parser dependencies) — this is what makes `polytypo/text`'s\n * module graph exclude `parse5` and the Micromark stack (AUDIT_REMEDIATION_AND_RELEASE_PLAN.md\n * 5.1).\n *\n * Validation order — `planRules` before `getLocaleData` — mirrors the pre-Stage-5 aggregate\n * `runPipeline` exactly (an unknown-rule error must win over an unknown-locale error when both\n * are present, since that is public, tested behaviour, not an implementation detail this\n * refactor was authorised to change).\n */\nexport function runTextPipeline(input: string, options: Partial<Options>): string {\n  const narrowTarget = resolveNarrowTarget(options.narrowNbsp);\n  const planned = planRules(options.rules);\n  const locale = getLocaleData(options.locale);\n  return fromCodePoints(runRules(toCodePoints(input), planned, locale, \"text\", narrowTarget));\n}\n\n/** analyze.md §1: the same pipeline as `runTextPipeline`, reporting instead of applying. */\nexport function analyzeTextPipeline(input: string, options: Partial<Options>): Change[] {\n  const narrowTarget = resolveNarrowTarget(options.narrowNbsp);\n  const planned = planRules(options.rules);\n  const locale = getLocaleData(options.locale);\n  const cp = toCodePoints(input);\n  const origin = cp.map((_value, index) => index);\n  return runRulesRecording(cp, planned, locale, \"text\", narrowTarget, origin, cp.length);\n}\n"]}

package/dist/chunk-72GQS2N7.cjs

@@ -0,0 +1,147 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});
+
+var _chunkYX6QTWOVcjs = require('./chunk-YX6QTWOV.cjs');
+
+// src/modes/html.ts
+var _parse5 = require('parse5');
+
+// src/modes/parse-error.ts
+function describe(error) {
+  if (typeof error !== "object" || error === null) return String(error);
+  const positioned = error;
+  const reason = typeof positioned.reason === "string" ? positioned.reason : error.message;
+  const place = positioned.place;
+  if (place !== null && place !== void 0 && typeof place.line === "number" && typeof place.column === "number") {
+    return `${reason} (line ${place.line}, column ${place.column})`;
+  }
+  return reason;
+}
+function wrapParserErrors(what, run) {
+  try {
+    return run();
+  } catch (error) {
+    if (error instanceof _chunkYX6QTWOVcjs.PolytypoError) throw error;
+    throw new (0, _chunkYX6QTWOVcjs.PolytypoError)(
+      "POLYTYPO_MALFORMED_INPUT",
+      `Input does not parse as ${what}: ${describe(error)}`
+    );
+  }
+}
+
+// src/modes/html.ts
+var SKIPPED_ELEMENTS = /* @__PURE__ */ new Set([
+  "code",
+  "pre",
+  "kbd",
+  "samp",
+  "var",
+  "script",
+  "style",
+  "textarea",
+  "svg",
+  "math"
+]);
+function isSkippedElement(tagName) {
+  return SKIPPED_ELEMENTS.has(tagName);
+}
+var AMPERSAND = 38;
+var SEMICOLON = 59;
+var HASH = 35;
+var LOWER_X = 120;
+var UPPER_X = 88;
+var MAX_REFERENCE_BODY = 34;
+function isAsciiDigit(unit) {
+  return unit >= 48 && unit <= 57;
+}
+function isAsciiHexDigit(unit) {
+  return isAsciiDigit(unit) || unit >= 97 && unit <= 102 || unit >= 65 && unit <= 70;
+}
+function isAsciiAlphanumeric(unit) {
+  return isAsciiDigit(unit) || unit >= 97 && unit <= 122 || unit >= 65 && unit <= 90;
+}
+function characterReferenceEnd(source, at, limit) {
+  if (source.charCodeAt(at) !== AMPERSAND) return -1;
+  let i = at + 1;
+  if (i >= limit) return -1;
+  if (source.charCodeAt(i) === HASH) {
+    i += 1;
+    const unit = i < limit ? source.charCodeAt(i) : -1;
+    const hex = unit === LOWER_X || unit === UPPER_X;
+    if (hex) i += 1;
+    const digitsAt = i;
+    while (i < limit && (hex ? isAsciiHexDigit(source.charCodeAt(i)) : isAsciiDigit(source.charCodeAt(i)))) {
+      i += 1;
+    }
+    if (i === digitsAt) return -1;
+    return i < limit && source.charCodeAt(i) === SEMICOLON ? i + 1 : -1;
+  }
+  const nameAt = i;
+  while (i < limit && i - nameAt < MAX_REFERENCE_BODY && isAsciiAlphanumeric(source.charCodeAt(i))) {
+    i += 1;
+  }
+  if (i === nameAt) return -1;
+  return i < limit && source.charCodeAt(i) === SEMICOLON ? i + 1 : -1;
+}
+function splitCharacterReferences(source, start, end) {
+  const spans = [];
+  let cursor = start;
+  let i = start;
+  while (i < end) {
+    if (source.charCodeAt(i) === AMPERSAND) {
+      const stop = characterReferenceEnd(source, i, end);
+      if (stop > 0) {
+        if (i > cursor) spans.push({ start: cursor, end: i });
+        cursor = stop;
+        i = stop;
+        continue;
+      }
+    }
+    i += 1;
+  }
+  if (end > cursor) spans.push({ start: cursor, end });
+  return spans;
+}
+function isParent(node) {
+  return "childNodes" in node;
+}
+function collect(node, source, spans) {
+  if (node.nodeName === "#text") {
+    const location = node.sourceCodeLocation;
+    if (location === void 0 || location === null) return;
+    for (const span of splitCharacterReferences(source, location.startOffset, location.endOffset)) {
+      spans.push(span);
+    }
+    return;
+  }
+  if (node.nodeName === "#comment" || node.nodeName === "#documentType") return;
+  if ("tagName" in node && isSkippedElement(node.tagName)) return;
+  if (node.nodeName === "template" && "content" in node) {
+    collect(node.content, source, spans);
+    return;
+  }
+  if (!isParent(node)) return;
+  for (const child of node.childNodes) collect(child, source, spans);
+}
+function htmlSpans(source) {
+  const document = wrapParserErrors("HTML", () => _parse5.parse.call(void 0, source, { sourceCodeLocationInfo: true }));
+  const spans = [];
+  collect(document, source, spans);
+  return spans;
+}
+function htmlFragmentSpans(source, offset) {
+  const fragment = wrapParserErrors(
+    "HTML",
+    () => _parse5.parseFragment.call(void 0, source, { sourceCodeLocationInfo: true })
+  );
+  const spans = [];
+  collect(fragment, source, spans);
+  return offset === 0 ? spans : spans.map((span) => ({ start: span.start + offset, end: span.end + offset }));
+}
+
+
+
+
+
+
+exports.wrapParserErrors = wrapParserErrors; exports.isSkippedElement = isSkippedElement; exports.htmlSpans = htmlSpans; exports.htmlFragmentSpans = htmlFragmentSpans;
+//# sourceMappingURL=chunk-72GQS2N7.cjs.map

package/dist/chunk-72GQS2N7.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-72GQS2N7.cjs","../src/modes/html.ts","../src/modes/parse-error.ts"],"names":[],"mappings":"AAAA;AACE;AACF,wDAA6B;AAC7B;AACA;ACJA,gCAAqC;ADMrC;AACA;AEAA,SAAS,QAAA,CAAS,KAAA,EAAwB;AACxC,EAAA,GAAA,CAAI,OAAO,MAAA,IAAU,SAAA,GAAY,MAAA,IAAU,IAAA,EAAM,OAAO,MAAA,CAAO,KAAK,CAAA;AACpE,EAAA,MAAM,WAAA,EAAa,KAAA;AACnB,EAAA,MAAM,OAAA,EACJ,OAAO,UAAA,CAAW,OAAA,IAAW,SAAA,EAAW,UAAA,CAAW,OAAA,EAAU,KAAA,CAAgB,OAAA;AAC/E,EAAA,MAAM,MAAA,EAAQ,UAAA,CAAW,KAAA;AACzB,EAAA,GAAA,CACE,MAAA,IAAU,KAAA,GACV,MAAA,IAAU,KAAA,EAAA,GACV,OAAO,KAAA,CAAM,KAAA,IAAS,SAAA,GACtB,OAAO,KAAA,CAAM,OAAA,IAAW,QAAA,EACxB;AACA,IAAA,OAAO,CAAA,EAAA;AACT,EAAA;AACO,EAAA;AACT;AAcgB;AACV,EAAA;AACK,IAAA;AACA,EAAA;AACH,IAAA;AACE,IAAA;AACJ,MAAA;AACA,MAAA;AACF,IAAA;AACF,EAAA;AACF;AFjBY;AACA;ACfN;AACJ,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACD;AAEe;AACP,EAAA;AACT;AAEM;AACA;AACO;AACP;AACA;AAGA;AAEG;AACA,EAAA;AACT;AAES;AACA,EAAA;AACT;AAES;AACA,EAAA;AACT;AAQS;AACH,EAAA;AACI,EAAA;AACC,EAAA;AAEL,EAAA;AACG,IAAA;AACC,IAAA;AACA,IAAA;AACG,IAAA;AACH,IAAA;AAEJ,IAAA;AAGK,MAAA;AACP,IAAA;AACI,IAAA;AACG,IAAA;AACT,EAAA;AAEM,EAAA;AAEJ,EAAA;AAIK,IAAA;AACP,EAAA;AACU,EAAA;AACH,EAAA;AACT;AAOgB;AACR,EAAA;AACF,EAAA;AACI,EAAA;AACD,EAAA;AACD,IAAA;AACI,MAAA;AACF,MAAA;AACE,QAAA;AACJ,QAAA;AACI,QAAA;AACJ,QAAA;AACF,MAAA;AACF,IAAA;AACK,IAAA;AACP,EAAA;AACU,EAAA;AACH,EAAA;AACT;AAES;AACA,EAAA;AACT;AAES;AACE,EAAA;AACD,IAAA;AACF,IAAA;AACJ,IAAA;AACQ,MAAA;AACR,IAAA;AACA,IAAA;AACF,EAAA;AAGS,EAAA;AAEL,EAAA;AAGK,EAAA;AACC,IAAA;AACR,IAAA;AACF,EAAA;AAEK,EAAA;AACL,EAAA;AACF;AAQgB;AACR,EAAA;AACA,EAAA;AACE,EAAA;AACD,EAAA;AACT;AAGgB;AACR,EAAA;AAA4B,IAAA;AAChC,IAAA;AACF,EAAA;AACM,EAAA;AACE,EAAA;AACD,EAAA;AAGT;AD9BY;AACA;AACA;AACA;AACA;AACA;AACA","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-72GQS2N7.cjs","sourcesContent":[null,"import { parse, parseFragment } from \"parse5\";\nimport type { DefaultTreeAdapterMap } from \"parse5\";\nimport { wrapParserErrors } from \"./parse-error.js\";\nimport type { Span } from \"./spans.js\";\n\ntype Node = DefaultTreeAdapterMap[\"node\"];\ntype ParentNode = DefaultTreeAdapterMap[\"parentNode\"];\n\n/**\n * spec/rules/modes.md 3.6, exhaustive and **closed**: extending it is a spec change, not an\n * implementation decision. `svg` and `math` are here because in MathML a quotation mark, a\n * hyphen and a prime are operators and identifiers — substituting a curly glyph changes what\n * the expression means. Everything not listed is processable, **including unknown and custom\n * elements**: guessing from a tag name is exactly the heuristic that diverges across runtimes.\n */\nconst SKIPPED_ELEMENTS: ReadonlySet<string> = new Set([\n  \"code\",\n  \"pre\",\n  \"kbd\",\n  \"samp\",\n  \"var\",\n  \"script\",\n  \"style\",\n  \"textarea\",\n  \"svg\",\n  \"math\",\n]);\n\nexport function isSkippedElement(tagName: string): boolean {\n  return SKIPPED_ELEMENTS.has(tagName);\n}\n\nconst AMPERSAND = 0x26;\nconst SEMICOLON = 0x3b;\nconst HASH = 0x23;\nconst LOWER_X = 0x78;\nconst UPPER_X = 0x58;\n\n/** Longest named reference in the WHATWG table is 32 characters; the guard is a bound, not a rule. */\nconst MAX_REFERENCE_BODY = 34;\n\nfunction isAsciiDigit(unit: number): boolean {\n  return unit >= 0x30 && unit <= 0x39;\n}\n\nfunction isAsciiHexDigit(unit: number): boolean {\n  return isAsciiDigit(unit) || (unit >= 0x61 && unit <= 0x66) || (unit >= 0x41 && unit <= 0x46);\n}\n\nfunction isAsciiAlphanumeric(unit: number): boolean {\n  return isAsciiDigit(unit) || (unit >= 0x61 && unit <= 0x7a) || (unit >= 0x41 && unit <= 0x5a);\n}\n\n/**\n * The end offset of a well-formed character reference starting at `at`, or -1. A bare `&` that a\n * parser would repair is deliberately **not** treated as a reference: it stays inside the span,\n * where no rule can emit or delete it, rather than manufacturing a boundary that suppresses\n * conversions around it.\n */\nfunction characterReferenceEnd(source: string, at: number, limit: number): number {\n  if (source.charCodeAt(at) !== AMPERSAND) return -1;\n  let i = at + 1;\n  if (i >= limit) return -1;\n\n  if (source.charCodeAt(i) === HASH) {\n    i += 1;\n    const unit = i < limit ? source.charCodeAt(i) : -1;\n    const hex = unit === LOWER_X || unit === UPPER_X;\n    if (hex) i += 1;\n    const digitsAt = i;\n    while (\n      i < limit &&\n      (hex ? isAsciiHexDigit(source.charCodeAt(i)) : isAsciiDigit(source.charCodeAt(i)))\n    ) {\n      i += 1;\n    }\n    if (i === digitsAt) return -1;\n    return i < limit && source.charCodeAt(i) === SEMICOLON ? i + 1 : -1;\n  }\n\n  const nameAt = i;\n  while (\n    i < limit &&\n    i - nameAt < MAX_REFERENCE_BODY &&\n    isAsciiAlphanumeric(source.charCodeAt(i))\n  ) {\n    i += 1;\n  }\n  if (i === nameAt) return -1;\n  return i < limit && source.charCodeAt(i) === SEMICOLON ? i + 1 : -1;\n}\n\n/**\n * modes.md 3.6: every character reference is skipped as an **opaque unit**, and a text node\n * containing one is split into spans around it. That is the only way `&nbsp;` survives as\n * `&nbsp;` instead of becoming a literal U+00A0 on the way out.\n */\nexport function splitCharacterReferences(source: string, start: number, end: number): Span[] {\n  const spans: Span[] = [];\n  let cursor = start;\n  let i = start;\n  while (i < end) {\n    if (source.charCodeAt(i) === AMPERSAND) {\n      const stop = characterReferenceEnd(source, i, end);\n      if (stop > 0) {\n        if (i > cursor) spans.push({ start: cursor, end: i });\n        cursor = stop;\n        i = stop;\n        continue;\n      }\n    }\n    i += 1;\n  }\n  if (end > cursor) spans.push({ start: cursor, end });\n  return spans;\n}\n\nfunction isParent(node: Node): node is ParentNode {\n  return \"childNodes\" in node;\n}\n\nfunction collect(node: Node, source: string, spans: Span[]): void {\n  if (node.nodeName === \"#text\") {\n    const location = node.sourceCodeLocation;\n    if (location === undefined || location === null) return;\n    for (const span of splitCharacterReferences(source, location.startOffset, location.endOffset)) {\n      spans.push(span);\n    }\n    return;\n  }\n\n  // Comments, doctype and processing instructions never contribute a span (modes.md 3.6).\n  if (node.nodeName === \"#comment\" || node.nodeName === \"#documentType\") return;\n\n  if (\"tagName\" in node && isSkippedElement(node.tagName)) return;\n\n  // A `template`'s children live in a separate document fragment; `template` is not skipped.\n  if (node.nodeName === \"template\" && \"content\" in node) {\n    collect(node.content, source, spans);\n    return;\n  }\n\n  if (!isParent(node)) return;\n  for (const child of node.childNodes) collect(child, source, spans);\n}\n\n/**\n * Locate the processable spans of an HTML document. The tree is used only to find offsets and is\n * then discarded — modes.md 4 forbids reserialisation, which is what makes attribute quoting,\n * self-closing forms, entity spelling, tag case and malformed-input recovery preserved by\n * construction rather than by effort.\n */\nexport function htmlSpans(source: string): Span[] {\n  const document = wrapParserErrors(\"HTML\", () => parse(source, { sourceCodeLocationInfo: true }));\n  const spans: Span[] = [];\n  collect(document, source, spans);\n  return spans;\n}\n\n/** The same, for a fragment: used for HTML blocks embedded in markdown (modes.md 3.7). */\nexport function htmlFragmentSpans(source: string, offset: number): Span[] {\n  const fragment = wrapParserErrors(\"HTML\", () =>\n    parseFragment(source, { sourceCodeLocationInfo: true }),\n  );\n  const spans: Span[] = [];\n  collect(fragment, source, spans);\n  return offset === 0\n    ? spans\n    : spans.map((span) => ({ start: span.start + offset, end: span.end + offset }));\n}\n","import { PolytypoError } from \"../errors.js\";\n\ninterface PositionedError {\n  readonly reason?: unknown;\n  readonly place?: { readonly line?: unknown; readonly column?: unknown } | null;\n}\n\nfunction describe(error: unknown): string {\n  if (typeof error !== \"object\" || error === null) return String(error);\n  const positioned = error as PositionedError;\n  const reason =\n    typeof positioned.reason === \"string\" ? positioned.reason : (error as Error).message;\n  const place = positioned.place;\n  if (\n    place !== null &&\n    place !== undefined &&\n    typeof place.line === \"number\" &&\n    typeof place.column === \"number\"\n  ) {\n    return `${reason} (line ${place.line}, column ${place.column})`;\n  }\n  return reason;\n}\n\n/**\n * spec/rules/modes.md 3.7.2. **The parser's own error type must never escape.** A\n * `VFileMessage`, a `Nokogiri::SyntaxError` or a Python exception on the public surface puts a\n * dependency's type into the contract and is unreproducible in the other four runtimes; only the\n * code is contractual (ARCHITECTURE.md 4.6), while the message and the source position are\n * useful and are not.\n *\n * Reachable in exactly one place: `dialect: \"mdx\"`, because MDX embeds JavaScript. HTML parsing\n * is specified with total error recovery and every byte sequence is valid CommonMark, so neither\n * of the other two languages can fail. The wrapper is applied to all of them anyway — the\n * guarantee is that no parser type escapes, not that this particular parser is trusted.\n */\nexport function wrapParserErrors<T>(what: string, run: () => T): T {\n  try {\n    return run();\n  } catch (error) {\n    if (error instanceof PolytypoError) throw error;\n    throw new PolytypoError(\n      \"POLYTYPO_MALFORMED_INPUT\",\n      `Input does not parse as ${what}: ${describe(error)}`,\n    );\n  }\n}\n"]}

package/dist/chunk-7S5PZ7T5.cjs

@@ -0,0 +1,205 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});
+
+
+
+
+
+
+
+
+
+
+var _chunkYX6QTWOVcjs = require('./chunk-YX6QTWOV.cjs');
+
+// src/modes/spans.ts
+var SPACE = 32;
+var LINE_TERMINATORS = /* @__PURE__ */ new Set([
+  10,
+  13,
+  11,
+  12,
+  133,
+  8232,
+  8233
+]);
+function gapIsLineBoundary(source, from, to) {
+  for (let i = from; i < to; i += 1) {
+    if (LINE_TERMINATORS.has(source.charCodeAt(i))) return true;
+  }
+  return false;
+}
+function normalizeSpans(spans) {
+  const sorted = spans.filter((s) => s.end > s.start).sort((a, b) => a.start - b.start);
+  const out = [];
+  for (const span of sorted) {
+    const last = out[out.length - 1];
+    if (last === void 0) {
+      out.push(span);
+      continue;
+    }
+    if (span.start < last.end) {
+      throw new (0, _chunkYX6QTWOVcjs.PolytypoError)(
+        "POLYTYPO_RULE_CONTRACT",
+        `mode extractor produced overlapping spans (${last.start}, ${last.end}) and (${span.start}, ${span.end})`
+      );
+    }
+    if (span.start === last.end) {
+      out[out.length - 1] = { start: last.start, end: span.end };
+      continue;
+    }
+    out.push(span);
+  }
+  return out;
+}
+function concatenateSpans(source, spans) {
+  const cp = [];
+  let previous;
+  for (const span of spans) {
+    if (previous !== void 0) {
+      cp.push(gapIsLineBoundary(source, previous.end, span.start) ? _chunkYX6QTWOVcjs.LINE_MARKER : _chunkYX6QTWOVcjs.MARKER);
+    }
+    for (const value of _chunkYX6QTWOVcjs.toCodePoints.call(void 0, source.slice(span.start, span.end))) cp.push(value);
+    previous = span;
+  }
+  return cp;
+}
+function spanRangesOf(cp) {
+  const ranges = [];
+  let first = 0;
+  for (let i = 0; i < cp.length; i += 1) {
+    if (_chunkYX6QTWOVcjs.isMarker.call(void 0, cp[i])) {
+      ranges.push({ first, last: i - 1 });
+      first = i + 1;
+    }
+  }
+  ranges.push({ first, last: cp.length - 1 });
+  return ranges;
+}
+function spanContaining(ranges, p) {
+  for (const range of ranges) {
+    if (p >= range.first && p <= range.last + 1) return range;
+  }
+  return void 0;
+}
+function filterBoundaryEdits(cp, edits, ranges) {
+  const out = [];
+  for (const edit of edits) {
+    let containsMarker = false;
+    for (let i = edit.start; i < edit.end; i += 1) {
+      if (_chunkYX6QTWOVcjs.isMarker.call(void 0, cp[i])) {
+        containsMarker = true;
+        break;
+      }
+    }
+    if (containsMarker) continue;
+    const p = edit.start;
+    const q = edit.end - 1;
+    const d = edit.end - edit.start;
+    const r = edit.replacement.length;
+    const span = spanContaining(ranges, p);
+    if (span !== void 0 && (p === span.first || q === span.last)) {
+      if (r > d) continue;
+      const first = edit.replacement[0];
+      const last = edit.replacement[r - 1];
+      if (r > 0 && p === span.first && first === SPACE && cp[p] !== SPACE) continue;
+      if (r > 0 && q === span.last && last === SPACE && cp[q] !== SPACE) continue;
+    }
+    out.push(edit);
+  }
+  return out;
+}
+function splitOnMarker(cp, expected) {
+  const pieces = [[]];
+  for (const value of cp) {
+    if (_chunkYX6QTWOVcjs.isMarker.call(void 0, value)) {
+      pieces.push([]);
+      continue;
+    }
+    pieces[pieces.length - 1].push(value);
+  }
+  if (pieces.length !== expected) {
+    throw new (0, _chunkYX6QTWOVcjs.PolytypoError)(
+      "POLYTYPO_RULE_CONTRACT",
+      `boundary markers did not survive the pipeline: expected ${expected} spans, found ${pieces.length}`
+    );
+  }
+  return pieces;
+}
+function originOfSpans(source, spans) {
+  const codePointIndexOf = new Array(source.length + 1);
+  let cpIndex = 0;
+  for (let i = 0; i < source.length; ) {
+    codePointIndexOf[i] = cpIndex;
+    const code = source.codePointAt(i);
+    const width = code > 65535 ? 2 : 1;
+    if (width === 2) codePointIndexOf[i + 1] = cpIndex;
+    i += width;
+    cpIndex += 1;
+  }
+  codePointIndexOf[source.length] = cpIndex;
+  const origin = [];
+  let previous;
+  for (const span of spans) {
+    if (previous !== void 0) origin.push(_chunkYX6QTWOVcjs.NO_ORIGIN);
+    const base = codePointIndexOf[span.start];
+    const length = _chunkYX6QTWOVcjs.toCodePoints.call(void 0, source.slice(span.start, span.end)).length;
+    for (let k = 0; k < length; k += 1) origin.push(base + k);
+    previous = span;
+  }
+  return origin;
+}
+
+// src/engine/span-runner.ts
+function runRulesOverSpans(cp, planned, locale, mode, narrowTarget) {
+  let current = cp;
+  for (const id of planned) {
+    const rule = _chunkYX6QTWOVcjs.RULES[id];
+    if (rule === void 0) continue;
+    const edits = rule.apply({ cp: current, locale, mode, narrowTarget });
+    current = _chunkYX6QTWOVcjs.applyEdits.call(void 0, current, filterBoundaryEdits(current, edits, spanRangesOf(current)), id);
+  }
+  return current;
+}
+function runOverSpans(source, spans, planned, locale, mode, narrowTarget) {
+  const normalized = normalizeSpans(spans);
+  if (normalized.length === 0) return source;
+  const transformed = runRulesOverSpans(
+    concatenateSpans(source, normalized),
+    planned,
+    locale,
+    mode,
+    narrowTarget
+  );
+  const pieces = splitOnMarker(transformed, normalized.length);
+  let out = "";
+  let cursor = 0;
+  for (let i = 0; i < normalized.length; i += 1) {
+    const span = normalized[i];
+    const replacement = _chunkYX6QTWOVcjs.fromCodePoints.call(void 0, pieces[i]);
+    const original = source.slice(span.start, span.end);
+    out += source.slice(cursor, span.start);
+    out += replacement === original ? original : replacement;
+    cursor = span.end;
+  }
+  return out + source.slice(cursor);
+}
+function analyzeOverSpans(source, spans, planned, locale, mode, narrowTarget) {
+  const normalized = normalizeSpans(spans);
+  if (normalized.length === 0) return [];
+  return _chunkYX6QTWOVcjs.runRulesRecording.call(void 0, 
+    concatenateSpans(source, normalized),
+    planned,
+    locale,
+    mode,
+    narrowTarget,
+    originOfSpans(source, normalized),
+    _chunkYX6QTWOVcjs.toCodePoints.call(void 0, source).length,
+    (current, edits) => filterBoundaryEdits(current, edits, spanRangesOf(current))
+  );
+}
+
+
+
+
+exports.runOverSpans = runOverSpans; exports.analyzeOverSpans = analyzeOverSpans;
+//# sourceMappingURL=chunk-7S5PZ7T5.cjs.map

package/dist/chunk-7S5PZ7T5.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-7S5PZ7T5.cjs","../src/modes/spans.ts","../src/engine/span-runner.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACA;ACMA,IAAM,MAAA,EAAQ,EAAA;AAEd,IAAM,iBAAA,kBAAwC,IAAI,GAAA,CAAI;AAAA,EACpD,EAAA;AAAA,EAAM,EAAA;AAAA,EAAM,EAAA;AAAA,EAAM,EAAA;AAAA,EAAM,GAAA;AAAA,EAAM,IAAA;AAAA,EAAQ;AACxC,CAAC,CAAA;AAED,SAAS,iBAAA,CAAkB,MAAA,EAAgB,IAAA,EAAc,EAAA,EAAqB;AAC5E,EAAA,IAAA,CAAA,IAAS,EAAA,EAAI,IAAA,EAAM,EAAA,EAAI,EAAA,EAAI,EAAA,GAAK,CAAA,EAAG;AACjC,IAAA,GAAA,CAAI,gBAAA,CAAiB,GAAA,CAAI,MAAA,CAAO,UAAA,CAAW,CAAC,CAAC,CAAA,EAAG,OAAO,IAAA;AAAA,EACzD;AACA,EAAA,OAAO,KAAA;AACT;AAuBO,SAAS,cAAA,CAAe,KAAA,EAAgC;AAC7D,EAAA,MAAM,OAAA,EAAS,KAAA,CAAM,MAAA,CAAO,CAAC,CAAA,EAAA,GAAM,CAAA,CAAE,IAAA,EAAM,CAAA,CAAE,KAAK,CAAA,CAAE,IAAA,CAAK,CAAC,CAAA,EAAG,CAAA,EAAA,GAAM,CAAA,CAAE,MAAA,EAAQ,CAAA,CAAE,KAAK,CAAA;AACpF,EAAA,MAAM,IAAA,EAAc,CAAC,CAAA;AACrB,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,MAAA,EAAQ;AACzB,IAAA,MAAM,KAAA,EAAO,GAAA,CAAI,GAAA,CAAI,OAAA,EAAS,CAAC,CAAA;AAC/B,IAAA,GAAA,CAAI,KAAA,IAAS,KAAA,CAAA,EAAW;AACtB,MAAA,GAAA,CAAI,IAAA,CAAK,IAAI,CAAA;AACb,MAAA,QAAA;AAAA,IACF;AACA,IAAA,GAAA,CAAI,IAAA,CAAK,MAAA,EAAQ,IAAA,CAAK,GAAA,EAAK;AACzB,MAAA,MAAM,IAAI,oCAAA;AAAA,QACR,wBAAA;AAAA,QACA,CAAA,2CAAA,EAA8C,IAAA,CAAK,KAAK,CAAA,EAAA,EAAK,IAAA,CAAK,GAAG,CAAA,OAAA,EAAU,IAAA,CAAK,KAAK,CAAA,EAAA,EAAK,IAAA,CAAK,GAAG,CAAA,CAAA;AAAA,MACxG,CAAA;AAAA,IACF;AACA,IAAA,GAAA,CAAI,IAAA,CAAK,MAAA,IAAU,IAAA,CAAK,GAAA,EAAK;AAC3B,MAAA,GAAA,CAAI,GAAA,CAAI,OAAA,EAAS,CAAC,EAAA,EAAI,EAAE,KAAA,EAAO,IAAA,CAAK,KAAA,EAAO,GAAA,EAAK,IAAA,CAAK,IAAI,CAAA;AACzD,MAAA,QAAA;AAAA,IACF;AACA,IAAA,GAAA,CAAI,IAAA,CAAK,IAAI,CAAA;AAAA,EACf;AACA,EAAA,OAAO,GAAA;AACT;AAGO,SAAS,gBAAA,CAAiB,MAAA,EAAgB,KAAA,EAAkC;AACjF,EAAA,MAAM,GAAA,EAAe,CAAC,CAAA;AACtB,EAAA,IAAI,QAAA;AACJ,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,GAAA,CAAI,SAAA,IAAa,KAAA,CAAA,EAAW;AAC1B,MAAA,EAAA,CAAG,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,QAAA,CAAS,GAAA,EAAK,IAAA,CAAK,KAAK,EAAA,EAAI,8BAAA,EAAc,wBAAM,CAAA;AAAA,IACpF;AACA,IAAA,IAAA,CAAA,MAAW,MAAA,GAAS,4CAAA,MAAa,CAAO,KAAA,CAAM,IAAA,CAAK,KAAA,EAAO,IAAA,CAAK,GAAG,CAAC,CAAA,EAAG,EAAA,CAAG,IAAA,CAAK,KAAK,CAAA;AACnF,IAAA,SAAA,EAAW,IAAA;AAAA,EACb;AACA,EAAA,OAAO,EAAA;AACT;AAOO,SAAS,YAAA,CAAa,EAAA,EAAoC;AAC/D,EAAA,MAAM,OAAA,EAAsB,CAAC,CAAA;AAC7B,EAAA,IAAI,MAAA,EAAQ,CAAA;AACZ,EAAA,IAAA,CAAA,IAAS,EAAA,EAAI,CAAA,EAAG,EAAA,EAAI,EAAA,CAAG,MAAA,EAAQ,EAAA,GAAK,CAAA,EAAG;AACrC,IAAA,GAAA,CAAI,wCAAA,EAAS,CAAG,CAAC,CAAW,CAAA,EAAG;AAC7B,MAAA,MAAA,CAAO,IAAA,CAAK,EAAE,KAAA,EAAO,IAAA,EAAM,EAAA,EAAI,EAAE,CAAC,CAAA;AAClC,MAAA,MAAA,EAAQ,EAAA,EAAI,CAAA;AAAA,IACd;AAAA,EACF;AACA,EAAA,MAAA,CAAO,IAAA,CAAK,EAAE,KAAA,EAAO,IAAA,EAAM,EAAA,CAAG,OAAA,EAAS,EAAE,CAAC,CAAA;AAC1C,EAAA,OAAO,MAAA;AACT;AAEA,SAAS,cAAA,CAAe,MAAA,EAA8B,CAAA,EAAkC;AACtF,EAAA,IAAA,CAAA,MAAW,MAAA,GAAS,MAAA,EAAQ;AAC1B,IAAA,GAAA,CAAI,EAAA,GAAK,KAAA,CAAM,MAAA,GAAS,EAAA,GAAK,KAAA,CAAM,KAAA,EAAO,CAAA,EAAG,OAAO,KAAA;AAAA,EACtD;AACA,EAAA,OAAO,KAAA,CAAA;AACT;AA4BO,SAAS,mBAAA,CACd,EAAA,EACA,KAAA,EACA,MAAA,EACQ;AACR,EAAA,MAAM,IAAA,EAAc,CAAC,CAAA;AACrB,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,IAAI,eAAA,EAAiB,KAAA;AACrB,IAAA,IAAA,CAAA,IAAS,EAAA,EAAI,IAAA,CAAK,KAAA,EAAO,EAAA,EAAI,IAAA,CAAK,GAAA,EAAK,EAAA,GAAK,CAAA,EAAG;AAC7C,MAAA,GAAA,CAAI,wCAAA,EAAS,CAAG,CAAC,CAAW,CAAA,EAAG;AAC7B,QAAA,eAAA,EAAiB,IAAA;AACjB,QAAA,KAAA;AAAA,MACF;AAAA,IACF;AACA,IAAA,GAAA,CAAI,cAAA,EAAgB,QAAA;AAIpB,IAAA,MAAM,EAAA,EAAI,IAAA,CAAK,KAAA;AACf,IAAA,MAAM,EAAA,EAAI,IAAA,CAAK,IAAA,EAAM,CAAA;AACrB,IAAA,MAAM,EAAA,EAAI,IAAA,CAAK,IAAA,EAAM,IAAA,CAAK,KAAA;AAC1B,IAAA,MAAM,EAAA,EAAI,IAAA,CAAK,WAAA,CAAY,MAAA;AAC3B,IAAA,MAAM,KAAA,EAAO,cAAA,CAAe,MAAA,EAAQ,CAAC,CAAA;AACrC,IAAA,GAAA,CAAI,KAAA,IAAS,KAAA,EAAA,GAAA,CAAc,EAAA,IAAM,IAAA,CAAK,MAAA,GAAS,EAAA,IAAM,IAAA,CAAK,IAAA,CAAA,EAAO;AAC/D,MAAA,GAAA,CAAI,EAAA,EAAI,CAAA,EAAG,QAAA;AAMX,MAAA,MAAM,MAAA,EAAQ,IAAA,CAAK,WAAA,CAAY,CAAC,CAAA;AAChC,MAAA,MAAM,KAAA,EAAO,IAAA,CAAK,WAAA,CAAY,EAAA,EAAI,CAAC,CAAA;AACnC,MAAA,GAAA,CAAI,EAAA,EAAI,EAAA,GAAK,EAAA,IAAM,IAAA,CAAK,MAAA,GAAS,MAAA,IAAU,MAAA,GAAS,EAAA,CAAG,CAAC,EAAA,IAAM,KAAA,EAAO,QAAA;AACrE,MAAA,GAAA,CAAI,EAAA,EAAI,EAAA,GAAK,EAAA,IAAM,IAAA,CAAK,KAAA,GAAQ,KAAA,IAAS,MAAA,GAAS,EAAA,CAAG,CAAC,EAAA,IAAM,KAAA,EAAO,QAAA;AAAA,IACrE;AAEA,IAAA,GAAA,CAAI,IAAA,CAAK,IAAI,CAAA;AAAA,EACf;AACA,EAAA,OAAO,GAAA;AACT;AAGO,SAAS,aAAA,CAAc,EAAA,EAAuB,QAAA,EAA8B;AACjF,EAAA,MAAM,OAAA,EAAqB,CAAC,CAAC,CAAC,CAAA;AAC9B,EAAA,IAAA,CAAA,MAAW,MAAA,GAAS,EAAA,EAAI;AACtB,IAAA,GAAA,CAAI,wCAAA,KAAc,CAAA,EAAG;AACnB,MAAA,MAAA,CAAO,IAAA,CAAK,CAAC,CAAC,CAAA;AACd,MAAA,QAAA;AAAA,IACF;AACA,IAAC,MAAA,CAAO,MAAA,CAAO,OAAA,EAAS,CAAC,CAAA,CAAe,IAAA,CAAK,KAAK,CAAA;AAAA,EACpD;AACA,EAAA,GAAA,CAAI,MAAA,CAAO,OAAA,IAAW,QAAA,EAAU;AAC9B,IAAA,MAAM,IAAI,oCAAA;AAAA,MACR,wBAAA;AAAA,MACA,CAAA,wDAAA,EAA2D,QAAQ,CAAA,cAAA,EAAiB,MAAA,CAAO,MAAM,CAAA;AAAA,IAAA;AACnG,EAAA;AAEF,EAAA;AACF;AASO;AACL,EAAA;AACA,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAAW,EAAA;AAEb,EAAA;AAEA,EAAA;AACA,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAAW,EAAA;AAEb,EAAA;AACF;ADlFA;AACA;AEjIA;AAOE,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAA4F,EAAA;AAE9F,EAAA;AACF;AAYO;AAQL,EAAA;AACA,EAAA;AAEA,EAAA;AAAoB,IAAA;AACiB,IAAA;AACnC,IAAA;AACA,IAAA;AACA,IAAA;AACA,EAAA;AAEF,EAAA;AAEA,EAAA;AACA,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAAc,EAAA;AAEhB,EAAA;AACF;AAOO;AAQL,EAAA;AACA,EAAA;AACA,EAAA;AAAO,IAAA;AAC8B,IAAA;AACnC,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACgC,IAAA;AACX,IAAA;AACwD,EAAA;AAEjF;AF4FA;AACA;AACA;AACA;AACA","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-7S5PZ7T5.cjs","sourcesContent":[null,"import { toCodePoints } from \"../engine/codepoints.js\";\nimport { isMarker, LINE_MARKER, MARKER } from \"../engine/sentinels.js\";\nimport { PolytypoError } from \"../errors.js\";\nimport type { Edit } from \"../types.js\";\nimport { NO_ORIGIN } from \"../engine/origin.js\";\n\n/**\n * The boundary markers are defined in the engine (src/engine/sentinels.ts), which is where all\n * three non-code-point sentinels live and where their disjointness is maintained. They are\n * re-exported here because the mode layer is what writes them into the array.\n *\n * Which marker separates two spans is decided by the **raw source bytes of the gap between\n * them**, so it is decidable without asking the parser anything and is identical in five\n * runtimes: a gap containing a line terminator gives `LINE_MARKER`, anything else `MARKER`.\n */\nexport { LINE_MARKER, MARKER, isMarker } from \"../engine/sentinels.js\";\n\n/** `BREAK` as the rules define it (`spaces.md` 3.1), tested against the gap's raw source. */\n/** U+0020, the one emitted code point whose meaning is positional (modes.md 3.4, 5 item 2). */\nconst SPACE = 0x20;\n\nconst LINE_TERMINATORS: ReadonlySet<number> = new Set([\n  0x0a, 0x0d, 0x0b, 0x0c, 0x85, 0x2028, 0x2029,\n]);\n\nfunction gapIsLineBoundary(source: string, from: number, to: number): boolean {\n  for (let i = from; i < to; i += 1) {\n    if (LINE_TERMINATORS.has(source.charCodeAt(i))) return true;\n  }\n  return false;\n}\n\n/**\n * A processable span, identified by its offsets in the **original source**. Offsets are UTF-16\n * indices into the JS source string, which is what every JS parser reports; they are an\n * implementation detail of this runtime and never reach the rules, which index code points.\n */\nexport interface Span {\n  readonly start: number;\n  readonly end: number;\n}\n\n/** A span's extent in the concatenated code-point array: `s₀` and `s₁` of modes.md 3.4. */\nexport interface SpanRange {\n  readonly first: number;\n  readonly last: number;\n}\n\n/**\n * Sort, drop empties, and coalesce spans separated by nothing in the source (modes.md 7.5:\n * a parser that reports one text run as two adjacent nodes must not manufacture a boundary).\n * Overlapping spans are an extractor bug and are rejected rather than silently merged.\n */\nexport function normalizeSpans(spans: readonly Span[]): Span[] {\n  const sorted = spans.filter((s) => s.end > s.start).sort((a, b) => a.start - b.start);\n  const out: Span[] = [];\n  for (const span of sorted) {\n    const last = out[out.length - 1];\n    if (last === undefined) {\n      out.push(span);\n      continue;\n    }\n    if (span.start < last.end) {\n      throw new PolytypoError(\n        \"POLYTYPO_RULE_CONTRACT\",\n        `mode extractor produced overlapping spans (${last.start}, ${last.end}) and (${span.start}, ${span.end})`,\n      );\n    }\n    if (span.start === last.end) {\n      out[out.length - 1] = { start: last.start, end: span.end };\n      continue;\n    }\n    out.push(span);\n  }\n  return out;\n}\n\n/** `S₁ ⌢ [marker] ⌢ S₂ ⌢ … ⌢ Sₘ` (modes.md 3.5 step 2). */\nexport function concatenateSpans(source: string, spans: readonly Span[]): number[] {\n  const cp: number[] = [];\n  let previous: Span | undefined;\n  for (const span of spans) {\n    if (previous !== undefined) {\n      cp.push(gapIsLineBoundary(source, previous.end, span.start) ? LINE_MARKER : MARKER);\n    }\n    for (const value of toCodePoints(source.slice(span.start, span.end))) cp.push(value);\n    previous = span;\n  }\n  return cp;\n}\n\n/**\n * The span extents of the array as it stands. Recomputed after every rule, because applying\n * edits shifts every index after the first one — the markers themselves are guaranteed to\n * survive, since no edit may contain one.\n */\nexport function spanRangesOf(cp: readonly number[]): SpanRange[] {\n  const ranges: SpanRange[] = [];\n  let first = 0;\n  for (let i = 0; i < cp.length; i += 1) {\n    if (isMarker(cp[i] as number)) {\n      ranges.push({ first, last: i - 1 });\n      first = i + 1;\n    }\n  }\n  ranges.push({ first, last: cp.length - 1 });\n  return ranges;\n}\n\nfunction spanContaining(ranges: readonly SpanRange[], p: number): SpanRange | undefined {\n  for (const range of ranges) {\n    if (p >= range.first && p <= range.last + 1) return range;\n  }\n  return undefined;\n}\n\n/**\n * modes.md 3.4, two safety nets, both pure functions of `(p, q, r, s₀, s₁)` — so the verdict is\n * identical on every run, which is what makes redistribution deterministic (modes.md 5, point 3).\n *\n * 1. **No edit may contain a marker.** No rule can produce one; one that does is a bug, and the\n *    edit is discarded rather than treated as a redistribution question.\n * 2. **The edge-growth rule.** An edit is discarded if it would place code points at an\n *    extremity of its span that were not there before: with `d = q - p + 1` the replaced length\n *    and `r` the replacement length, discard when `p = s₀ and r > d`, or `q = s₁ and r > d`.\n *\n * The length test is the whole rule and needs no knowledge of Markdown or HTML syntax. It\n * separates exactly the cases that matter: `\"` → `“` at an edge is 1 → 1 and applies; `--` →\n * `␣–␣` at an edge is 2 → 3 and is discarded, while the same edit interior to a span applies;\n * `(c)` → `©` is 3 → 1 and applies, because shrinking is always safe. An insertion has `d = 0`,\n * so it is discarded exactly when its position coincides with a span edge — the rule this one\n * generalises.\n *\n * **Deletion at an edge is not restricted here**, and must not be: `r > d` is false for a\n * deletion, so this filter never sees one. Deletion at an edge is governed by modes.md 3.3's\n * *Edge tests* clause instead — where a rule asks \"am I at the edge of the text I am allowed to\n * modify\" rather than \"what character is here\", the marker behaves as `NONE`. That clause is the\n * one place a rule may treat a span edge as the end of the text, it exists because `spaces` is\n * the only rule that deletes, and 3.3 states the division: **a rule that deletes must treat a\n * span edge as the end of the text; a rule that replaces or inserts must not, and is governed by\n * this filter.** The single test that claims the clause is `spaces.md` 3.2 step 4.\n */\nexport function filterBoundaryEdits(\n  cp: readonly number[],\n  edits: readonly Edit[],\n  ranges: readonly SpanRange[],\n): Edit[] {\n  const out: Edit[] = [];\n  for (const edit of edits) {\n    let containsMarker = false;\n    for (let i = edit.start; i < edit.end; i += 1) {\n      if (isMarker(cp[i] as number)) {\n        containsMarker = true;\n        break;\n      }\n    }\n    if (containsMarker) continue;\n\n    // `[start, end)` in the engine's half-open form is `cp[p … q]` with p = start, q = end - 1;\n    // an insertion is `q = p - 1`, which falls out of the same expression.\n    const p = edit.start;\n    const q = edit.end - 1;\n    const d = edit.end - edit.start;\n    const r = edit.replacement.length;\n    const span = spanContaining(ranges, p);\n    if (span !== undefined && (p === span.first || q === span.last)) {\n      if (r > d) continue;\n      // The character clause. `r > d` is not the rule, only a formalisation of it that misses\n      // `r === d`: `dashes` P3 admits a run of THREE dashes, so `---` -> `␣–␣` is 3 -> 3 and the\n      // length test sees nothing while U+0020 lands on both extremities anyway. Testing the\n      // character catches it, and only U+0020 needs testing — it is the one code point any rule\n      // emits whose meaning comes from its position rather than from itself (modes.md 5 item 2).\n      const first = edit.replacement[0];\n      const last = edit.replacement[r - 1];\n      if (r > 0 && p === span.first && first === SPACE && cp[p] !== SPACE) continue;\n      if (r > 0 && q === span.last && last === SPACE && cp[q] !== SPACE) continue;\n    }\n\n    out.push(edit);\n  }\n  return out;\n}\n\n/** Redistribute the transformed array back to one piece per span (modes.md 3.5 step 4). */\nexport function splitOnMarker(cp: readonly number[], expected: number): number[][] {\n  const pieces: number[][] = [[]];\n  for (const value of cp) {\n    if (isMarker(value)) {\n      pieces.push([]);\n      continue;\n    }\n    (pieces[pieces.length - 1] as number[]).push(value);\n  }\n  if (pieces.length !== expected) {\n    throw new PolytypoError(\n      \"POLYTYPO_RULE_CONTRACT\",\n      `boundary markers did not survive the pipeline: expected ${expected} spans, found ${pieces.length}`,\n    );\n  }\n  return pieces;\n}\n\n/**\n * The origin map for `concatenateSpans` (analyze.md §2): for every code point of the joined\n * array, the code-point offset of the character it came from IN THE DOCUMENT, and NO_ORIGIN for\n * the markers, which came from nowhere. `Span` bounds are native string indices, so the\n * conversion to code-point offsets happens here and not in the caller — this is the one place\n * that knows both coordinate systems.\n */\nexport function originOfSpans(source: string, spans: readonly Span[]): number[] {\n  const codePointIndexOf = new Array<number>(source.length + 1);\n  let cpIndex = 0;\n  for (let i = 0; i < source.length;) {\n    codePointIndexOf[i] = cpIndex;\n    const code = source.codePointAt(i) as number;\n    const width = code > 0xffff ? 2 : 1;\n    if (width === 2) codePointIndexOf[i + 1] = cpIndex;\n    i += width;\n    cpIndex += 1;\n  }\n  codePointIndexOf[source.length] = cpIndex;\n\n  const origin: number[] = [];\n  let previous: Span | undefined;\n  for (const span of spans) {\n    if (previous !== undefined) origin.push(NO_ORIGIN);\n    const base = codePointIndexOf[span.start] as number;\n    const length = toCodePoints(source.slice(span.start, span.end)).length;\n    for (let k = 0; k < length; k += 1) origin.push(base + k);\n    previous = span;\n  }\n  return origin;\n}\n","import {\n  concatenateSpans,\n  filterBoundaryEdits,\n  normalizeSpans,\n  originOfSpans,\n  spanRangesOf,\n  splitOnMarker,\n  type Span,\n} from \"../modes/spans.js\";\nimport { RULES } from \"../rules/registry.js\";\nimport type { LocaleData, Mode, RuleId } from \"../types.js\";\nimport { applyEdits } from \"./edits.js\";\nimport { fromCodePoints, toCodePoints } from \"./codepoints.js\";\nimport type { Change } from \"./origin.js\";\nimport { runRulesRecording } from \"./rule-runner.js\";\n\n/**\n * The same sequence as `runRules` (`./rule-runner.js`), with the two boundary filters of\n * modes.md 3.4 interposed. The span extents are recomputed after every rule, because applying\n * an edit shifts every index after it; the markers themselves always survive, since no edit may\n * contain one.\n */\nfunction runRulesOverSpans(\n  cp: readonly number[],\n  planned: readonly RuleId[],\n  locale: LocaleData,\n  mode: Mode,\n  narrowTarget: number,\n): readonly number[] {\n  let current = cp;\n  for (const id of planned) {\n    const rule = RULES[id];\n    if (rule === undefined) continue;\n    const edits = rule.apply({ cp: current, locale, mode, narrowTarget });\n    current = applyEdits(current, filterBoundaryEdits(current, edits, spanRangesOf(current)), id);\n  }\n  return current;\n}\n\n/**\n * modes.md 3.5. The pipeline runs **once**, over the marker-separated concatenation of every\n * processable span — not per span, which would pair quotation marks in isolation, and not over a\n * naive concatenation, which would manufacture adjacencies the document does not have.\n *\n * The output is the input with a set of disjoint substring replacements applied and nothing else\n * (modes.md 4). A span whose content the rules did not change contributes no replacement, so a\n * document needing no changes comes back byte-identical; the parser located the spans and was\n * then discarded, and the document is never serialised.\n */\nexport function runOverSpans(\n  source: string,\n  spans: readonly Span[],\n  planned: readonly RuleId[],\n  locale: LocaleData,\n  mode: Mode,\n  narrowTarget: number,\n): string {\n  const normalized = normalizeSpans(spans);\n  if (normalized.length === 0) return source;\n\n  const transformed = runRulesOverSpans(\n    concatenateSpans(source, normalized),\n    planned,\n    locale,\n    mode,\n    narrowTarget,\n  );\n  const pieces = splitOnMarker(transformed, normalized.length);\n\n  let out = \"\";\n  let cursor = 0;\n  for (let i = 0; i < normalized.length; i += 1) {\n    const span = normalized[i] as Span;\n    const replacement = fromCodePoints(pieces[i] as number[]);\n    const original = source.slice(span.start, span.end);\n    out += source.slice(cursor, span.start);\n    out += replacement === original ? original : replacement;\n    cursor = span.end;\n  }\n  return out + source.slice(cursor);\n}\n\n/**\n * `runOverSpans`, reporting instead of applying (analyze.md §1). The span table supplies the\n * origin map, so every change comes back in DOCUMENT coordinates — analyze.md §6 names a\n * runtime that reports span-local offsets here as the mistake that passes every text-mode test.\n */\nexport function analyzeOverSpans(\n  source: string,\n  spans: readonly Span[],\n  planned: readonly RuleId[],\n  locale: LocaleData,\n  mode: Mode,\n  narrowTarget: number,\n): Change[] {\n  const normalized = normalizeSpans(spans);\n  if (normalized.length === 0) return [];\n  return runRulesRecording(\n    concatenateSpans(source, normalized),\n    planned,\n    locale,\n    mode,\n    narrowTarget,\n    originOfSpans(source, normalized),\n    toCodePoints(source).length,\n    (current, edits) => filterBoundaryEdits(current, edits, spanRangesOf(current)),\n  );\n}\n"]}