polytypo 1.1.0 → 1.2.0

package/dist/chunk-HGUA2NVX.cjs

@@ -0,0 +1,21 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});
+
+
+var _chunkLS57O4KJcjs = require('./chunk-LS57O4KJ.cjs');
+
+
+
+var _chunkB4O4R6HIcjs = require('./chunk-B4O4R6HI.cjs');
+
+// src/engine/html-pipeline.ts
+function runHtmlPipeline(input, options) {
+  const planned = _chunkB4O4R6HIcjs.planRules.call(void 0, options.rules);
+  const locale = _chunkB4O4R6HIcjs.getLocaleData.call(void 0, options.locale);
+  const spans = _chunkLS57O4KJcjs.htmlSpans.call(void 0, input);
+  return _chunkLS57O4KJcjs.runOverSpans.call(void 0, input, spans, planned, locale, "html");
+}
+
+
+
+exports.runHtmlPipeline = runHtmlPipeline;
+//# sourceMappingURL=chunk-HGUA2NVX.cjs.map

package/dist/{chunk-IS7WKMGA.cjs.map → chunk-HGUA2NVX.cjs.map}

@@ -1 +1 @@
-{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-IS7WKMGA.cjs","../src/engine/html-pipeline.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACF,wDAA6B;AAC7B;AACE;AACA;AACF,wDAA6B;AAC7B;AACA;ACMO,SAAS,eAAA,CAAgB,KAAA,EAAe,OAAA,EAAmC;AAChF,EAAA,MAAM,QAAA,EAAU,yCAAA,OAAU,CAAQ,KAAK,CAAA;AACvC,EAAA,MAAM,OAAA,EAAS,6CAAA,OAAc,CAAQ,MAAM,CAAA;AAC3C,EAAA,MAAM,MAAA,EAAQ,yCAAA,KAAe,CAAA;AAC7B,EAAA,OAAO,4CAAA,KAAa,EAAO,KAAA,EAAO,OAAA,EAAS,MAAA,EAAQ,MAAM,CAAA;AAC3D;ADJA;AACA;AACE;AACF,0CAAC","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-IS7WKMGA.cjs","sourcesContent":[null,"import { htmlSpans } from \"../modes/html.js\";\nimport type { Options } from \"../types.js\";\nimport { getLocaleData } from \"./locale.js\";\nimport { planRules } from \"./rule-runner.js\";\nimport { runOverSpans } from \"./span-runner.js\";\n\n/**\n * `html` mode only. Imports `parse5` (via `../modes/html.js`) and nothing from\n * `../modes/markdown.js` or the Micromark/MDX stack — this is what makes `polytypo/html`'s\n * module graph exclude the Markdown parser (AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 5.1).\n *\n * Validation order — `planRules` before `getLocaleData` before parsing — mirrors the pre-Stage-5\n * aggregate `runPipeline` exactly (an unknown-rule error must win over an unknown-locale error,\n * and both must win over a parse failure, since that is public, tested behaviour).\n */\nexport function runHtmlPipeline(input: string, options: Partial<Options>): string {\n  const planned = planRules(options.rules);\n  const locale = getLocaleData(options.locale);\n  const spans = htmlSpans(input);\n  return runOverSpans(input, spans, planned, locale, \"html\");\n}\n"]}
+{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-HGUA2NVX.cjs","../src/engine/html-pipeline.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACF,wDAA6B;AAC7B;AACE;AACA;AACF,wDAA6B;AAC7B;AACA;ACMO,SAAS,eAAA,CAAgB,KAAA,EAAe,OAAA,EAAmC;AAChF,EAAA,MAAM,QAAA,EAAU,yCAAA,OAAU,CAAQ,KAAK,CAAA;AACvC,EAAA,MAAM,OAAA,EAAS,6CAAA,OAAc,CAAQ,MAAM,CAAA;AAC3C,EAAA,MAAM,MAAA,EAAQ,yCAAA,KAAe,CAAA;AAC7B,EAAA,OAAO,4CAAA,KAAa,EAAO,KAAA,EAAO,OAAA,EAAS,MAAA,EAAQ,MAAM,CAAA;AAC3D;ADJA;AACA;AACE;AACF,0CAAC","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-HGUA2NVX.cjs","sourcesContent":[null,"import { htmlSpans } from \"../modes/html.js\";\nimport type { Options } from \"../types.js\";\nimport { getLocaleData } from \"./locale.js\";\nimport { planRules } from \"./rule-runner.js\";\nimport { runOverSpans } from \"./span-runner.js\";\n\n/**\n * `html` mode only. Imports `parse5` (via `../modes/html.js`) and nothing from\n * `../modes/markdown.js` or the Micromark/MDX stack — this is what makes `polytypo/html`'s\n * module graph exclude the Markdown parser (AUDIT_REMEDIATION_AND_RELEASE_PLAN.md 5.1).\n *\n * Validation order — `planRules` before `getLocaleData` before parsing — mirrors the pre-Stage-5\n * aggregate `runPipeline` exactly (an unknown-rule error must win over an unknown-locale error,\n * and both must win over a parse failure, since that is public, tested behaviour).\n */\nexport function runHtmlPipeline(input: string, options: Partial<Options>): string {\n  const planned = planRules(options.rules);\n  const locale = getLocaleData(options.locale);\n  const spans = htmlSpans(input);\n  return runOverSpans(input, spans, planned, locale, \"html\");\n}\n"]}

package/dist/{chunk-4FZZJB7Y.js → chunk-KAO74OJ6.js}

@@ -1,6 +1,6 @@
 import {
   PolytypoError
-} from "./chunk-3XRMEZDM.js";
+} from "./chunk-OHF7OFWY.js";
 
 // src/engine/assert-fixed-mode.ts
 function assertFixedMode(mode, fixed, entryPoint) {
@@ -14,4 +14,4 @@ function assertFixedMode(mode, fixed, entryPoint) {
 export {
   assertFixedMode
 };
-//# sourceMappingURL=chunk-4FZZJB7Y.js.map
+//# sourceMappingURL=chunk-KAO74OJ6.js.map

package/dist/{chunk-RNFW7474.cjs → chunk-LS57O4KJ.cjs}

@@ -7,7 +7,7 @@
 
 
 
-var _chunkADRRRKLBcjs = require('./chunk-ADRRRKLB.cjs');
+var _chunkB4O4R6HIcjs = require('./chunk-B4O4R6HI.cjs');
 
 // src/modes/html.ts
 var _parse5 = require('parse5');
@@ -27,8 +27,8 @@ function wrapParserErrors(what, run) {
   try {
     return run();
   } catch (error) {
-    if (error instanceof _chunkADRRRKLBcjs.PolytypoError) throw error;
-    throw new (0, _chunkADRRRKLBcjs.PolytypoError)(
+    if (error instanceof _chunkB4O4R6HIcjs.PolytypoError) throw error;
+    throw new (0, _chunkB4O4R6HIcjs.PolytypoError)(
       "POLYTYPO_MALFORMED_INPUT",
       `Input does not parse as ${what}: ${describe(error)}`
     );
@@ -171,7 +171,7 @@ function normalizeSpans(spans) {
       continue;
     }
     if (span.start < last.end) {
-      throw new (0, _chunkADRRRKLBcjs.PolytypoError)(
+      throw new (0, _chunkB4O4R6HIcjs.PolytypoError)(
         "POLYTYPO_RULE_CONTRACT",
         `mode extractor produced overlapping spans (${last.start}, ${last.end}) and (${span.start}, ${span.end})`
       );
@@ -189,9 +189,9 @@ function concatenateSpans(source, spans) {
   let previous;
   for (const span of spans) {
     if (previous !== void 0) {
-      cp.push(gapIsLineBoundary(source, previous.end, span.start) ? _chunkADRRRKLBcjs.LINE_MARKER : _chunkADRRRKLBcjs.MARKER);
+      cp.push(gapIsLineBoundary(source, previous.end, span.start) ? _chunkB4O4R6HIcjs.LINE_MARKER : _chunkB4O4R6HIcjs.MARKER);
     }
-    for (const value of _chunkADRRRKLBcjs.toCodePoints.call(void 0, source.slice(span.start, span.end))) cp.push(value);
+    for (const value of _chunkB4O4R6HIcjs.toCodePoints.call(void 0, source.slice(span.start, span.end))) cp.push(value);
     previous = span;
   }
   return cp;
@@ -200,7 +200,7 @@ function spanRangesOf(cp) {
   const ranges = [];
   let first = 0;
   for (let i = 0; i < cp.length; i += 1) {
-    if (_chunkADRRRKLBcjs.isMarker.call(void 0, cp[i])) {
+    if (_chunkB4O4R6HIcjs.isMarker.call(void 0, cp[i])) {
       ranges.push({ first, last: i - 1 });
       first = i + 1;
     }
@@ -219,7 +219,7 @@ function filterBoundaryEdits(cp, edits, ranges) {
   for (const edit of edits) {
     let containsMarker = false;
     for (let i = edit.start; i < edit.end; i += 1) {
-      if (_chunkADRRRKLBcjs.isMarker.call(void 0, cp[i])) {
+      if (_chunkB4O4R6HIcjs.isMarker.call(void 0, cp[i])) {
         containsMarker = true;
         break;
       }
@@ -238,14 +238,14 @@ function filterBoundaryEdits(cp, edits, ranges) {
 function splitOnMarker(cp, expected) {
   const pieces = [[]];
   for (const value of cp) {
-    if (_chunkADRRRKLBcjs.isMarker.call(void 0, value)) {
+    if (_chunkB4O4R6HIcjs.isMarker.call(void 0, value)) {
       pieces.push([]);
       continue;
     }
     pieces[pieces.length - 1].push(value);
   }
   if (pieces.length !== expected) {
-    throw new (0, _chunkADRRRKLBcjs.PolytypoError)(
+    throw new (0, _chunkB4O4R6HIcjs.PolytypoError)(
       "POLYTYPO_RULE_CONTRACT",
       `boundary markers did not survive the pipeline: expected ${expected} spans, found ${pieces.length}`
     );
@@ -257,10 +257,10 @@ function splitOnMarker(cp, expected) {
 function runRulesOverSpans(cp, planned, locale, mode) {
   let current = cp;
   for (const id of planned) {
-    const rule = _chunkADRRRKLBcjs.RULES[id];
+    const rule = _chunkB4O4R6HIcjs.RULES[id];
     if (rule === void 0) continue;
     const edits = rule.apply({ cp: current, locale, mode });
-    current = _chunkADRRRKLBcjs.applyEdits.call(void 0, current, filterBoundaryEdits(current, edits, spanRangesOf(current)), id);
+    current = _chunkB4O4R6HIcjs.applyEdits.call(void 0, current, filterBoundaryEdits(current, edits, spanRangesOf(current)), id);
   }
   return current;
 }
@@ -278,7 +278,7 @@ function runOverSpans(source, spans, planned, locale, mode) {
   let cursor = 0;
   for (let i = 0; i < normalized.length; i += 1) {
     const span = normalized[i];
-    const replacement = _chunkADRRRKLBcjs.fromCodePoints.call(void 0, pieces[i]);
+    const replacement = _chunkB4O4R6HIcjs.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;
@@ -294,4 +294,4 @@ function runOverSpans(source, spans, planned, locale, mode) {
 
 
 exports.wrapParserErrors = wrapParserErrors; exports.isSkippedElement = isSkippedElement; exports.htmlSpans = htmlSpans; exports.htmlFragmentSpans = htmlFragmentSpans; exports.runOverSpans = runOverSpans;
-//# sourceMappingURL=chunk-RNFW7474.cjs.map
+//# sourceMappingURL=chunk-LS57O4KJ.cjs.map

package/dist/{chunk-RNFW7474.cjs.map → chunk-LS57O4KJ.cjs.map}

@@ -1 +1 @@
-{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-RNFW7474.cjs","../src/modes/html.ts","../src/modes/parse-error.ts","../src/modes/spans.ts","../src/engine/span-runner.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACA;ACXA,gCAAqC;ADarC;AACA;AEPA,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;AFVY;AACA;ACtBN;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;ADvBY;AACA;AGlIN;AACJ,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAQ,EAAA;AACvC;AAEQ;AACE,EAAA;AACH,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAuBgB;AACR,EAAA;AACA,EAAA;AACN,EAAA;AACQ,IAAA;AACF,IAAA;AACE,MAAA;AACJ,MAAA;AACF,IAAA;AACI,IAAA;AACI,MAAA;AACJ,QAAA;AACA,QAAA;AACF,MAAA;AACF,IAAA;AACI,IAAA;AACE,MAAA;AACJ,MAAA;AACF,IAAA;AACI,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAGgB;AACR,EAAA;AACF,EAAA;AACJ,EAAA;AACM,IAAA;AACC,MAAA;AACL,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAOgB;AACR,EAAA;AACF,EAAA;AACK,EAAA;AACH,IAAA;AACF,MAAA;AACA,MAAA;AACF,IAAA;AACF,EAAA;AACO,EAAA;AACA,EAAA;AACT;AAES;AACP,EAAA;AACM,IAAA;AACN,EAAA;AACO,EAAA;AACT;AA4BgB;AAKR,EAAA;AACN,EAAA;AACM,IAAA;AACJ,IAAA;AACM,MAAA;AACF,QAAA;AACA,QAAA;AACF,MAAA;AACF,IAAA;AACI,IAAA;AAIE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,IAAA;AAEA,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAGgB;AACR,EAAA;AACN,EAAA;AACM,IAAA;AACF,MAAA;AACA,MAAA;AACF,IAAA;AACQ,IAAA;AACV,EAAA;AACI,EAAA;AACI,IAAA;AACJ,MAAA;AACA,MAAA;AACF,IAAA;AACF,EAAA;AACO,EAAA;AACT;AHqEY;AACA;AI5OH;AAMH,EAAA;AACJ,EAAA;AACQ,IAAA;AACF,IAAA;AACE,IAAA;AACN,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAYgB;AAOR,EAAA;AACF,EAAA;AAEE,EAAA;AACJ,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACM,EAAA;AAEI,EAAA;AACN,EAAA;AACK,EAAA;AACD,IAAA;AACA,IAAA;AACA,IAAA;AACC,IAAA;AACA,IAAA;AACP,IAAA;AACF,EAAA;AACO,EAAA;AACT;AJsNY;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-RNFW7474.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","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\";\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. */\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 && r > d && (p === span.first || q === span.last)) continue;\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","import {\n  concatenateSpans,\n  filterBoundaryEdits,\n  normalizeSpans,\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 } from \"./codepoints.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): 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 });\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): 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  );\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"]}
+{"version":3,"sources":["/home/runner/work/polytypo-js/polytypo-js/dist/chunk-LS57O4KJ.cjs","../src/modes/html.ts","../src/modes/parse-error.ts","../src/modes/spans.ts","../src/engine/span-runner.ts"],"names":[],"mappings":"AAAA;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACA;ACXA,gCAAqC;ADarC;AACA;AEPA,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;AFVY;AACA;ACtBN;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;ADvBY;AACA;AGlIN;AACJ,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAM,EAAA;AAAQ,EAAA;AACvC;AAEQ;AACE,EAAA;AACH,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAuBgB;AACR,EAAA;AACA,EAAA;AACN,EAAA;AACQ,IAAA;AACF,IAAA;AACE,MAAA;AACJ,MAAA;AACF,IAAA;AACI,IAAA;AACI,MAAA;AACJ,QAAA;AACA,QAAA;AACF,MAAA;AACF,IAAA;AACI,IAAA;AACE,MAAA;AACJ,MAAA;AACF,IAAA;AACI,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAGgB;AACR,EAAA;AACF,EAAA;AACJ,EAAA;AACM,IAAA;AACC,MAAA;AACL,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAOgB;AACR,EAAA;AACF,EAAA;AACK,EAAA;AACH,IAAA;AACF,MAAA;AACA,MAAA;AACF,IAAA;AACF,EAAA;AACO,EAAA;AACA,EAAA;AACT;AAES;AACP,EAAA;AACM,IAAA;AACN,EAAA;AACO,EAAA;AACT;AA4BgB;AAKR,EAAA;AACN,EAAA;AACM,IAAA;AACJ,IAAA;AACM,MAAA;AACF,QAAA;AACA,QAAA;AACF,MAAA;AACF,IAAA;AACI,IAAA;AAIE,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,IAAA;AAEA,IAAA;AACN,EAAA;AACO,EAAA;AACT;AAGgB;AACR,EAAA;AACN,EAAA;AACM,IAAA;AACF,MAAA;AACA,MAAA;AACF,IAAA;AACQ,IAAA;AACV,EAAA;AACI,EAAA;AACI,IAAA;AACJ,MAAA;AACA,MAAA;AACF,IAAA;AACF,EAAA;AACO,EAAA;AACT;AHqEY;AACA;AI5OH;AAMH,EAAA;AACJ,EAAA;AACQ,IAAA;AACF,IAAA;AACE,IAAA;AACN,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAYgB;AAOR,EAAA;AACF,EAAA;AAEE,EAAA;AACJ,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACM,EAAA;AAEI,EAAA;AACN,EAAA;AACK,EAAA;AACD,IAAA;AACA,IAAA;AACA,IAAA;AACC,IAAA;AACA,IAAA;AACP,IAAA;AACF,EAAA;AACO,EAAA;AACT;AJsNY;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/home/runner/work/polytypo-js/polytypo-js/dist/chunk-LS57O4KJ.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","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\";\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. */\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 && r > d && (p === span.first || q === span.last)) continue;\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","import {\n  concatenateSpans,\n  filterBoundaryEdits,\n  normalizeSpans,\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 } from \"./codepoints.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): 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 });\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): 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  );\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"]}

package/dist/{chunk-3XRMEZDM.js → chunk-OHF7OFWY.js}

@@ -3897,6 +3897,16 @@ var CLOSEISH = /* @__PURE__ */ new Set([
   8211,
   8212
 ]);
+var OPENQUOTE = /* @__PURE__ */ new Set([
+  171,
+  8216,
+  8218,
+  8219,
+  8220,
+  8222,
+  8223,
+  8249
+]);
 function at(cp, i) {
   const value = cp[i];
   return value === void 0 ? NONE : value;
@@ -3912,6 +3922,7 @@ function isApostrophe(left, right) {
   if (isAlnum(left) && isAlnum(right)) return true;
   if (isLetter(left) && (right === NONE || SPACELIKE.has(right) || CLOSEISH.has(right)))
     return true;
+  if (isLetter(left) && OPENQUOTE.has(right)) return true;
   if ((left === NONE || SPACELIKE.has(left) || OPENISH.has(left)) && isAlnum(right)) return true;
   return false;
 }
@@ -4469,7 +4480,7 @@ function isOpenish(prep, cp) {
   return contains(prep.opens, cp);
 }
 function isCloseish(prep, cp) {
-  return contains(prep.closes, cp);
+  return cp === MARKER || contains(prep.closes, cp);
 }
 function isMark(prep, cp) {
   return contains(prep.beforePunctuation, cp) || contains(prep.narrowBeforePunctuation, cp);
@@ -5247,7 +5258,9 @@ function isEmptyBracketGuarded(left, right) {
 }
 function isLoneDot(cp, e) {
   if (at7(cp, e) !== FULL_STOP4) return true;
-  return !isDotlike2(at7(cp, e + 1));
+  const next = at7(cp, e + 1);
+  if (isLetter(next) || isDigitAscii(next)) return false;
+  return !isDotlike2(next);
 }
 function isDigitAscii(cp) {
   return cp >= DIGIT_ZERO7 && cp <= DIGIT_NINE6;
@@ -5617,4 +5630,4 @@ export {
   planRules,
   runRules
 };
-//# sourceMappingURL=chunk-3XRMEZDM.js.map
+//# sourceMappingURL=chunk-OHF7OFWY.js.map