@yolk-sdk/extractors 0.1.0-canary.98

package/dist/node/sheetjs-xml.mjs

@@ -0,0 +1,253 @@
+import { Buffer } from "node:buffer";
+//#region src/node/sheetjs-xml.ts
+/**
+* Ports of the SheetJS 0.20.3 XML helpers (`xlsx.mjs`) the extractor needs to read a part the way
+* SheetJS would: its tag pattern, `parsexmltag`, `strip_ns`, `utf8read`, and `unescapexml`.
+* Everything here is linear in the text it scans.
+*/
+const swapUtf16ByteOrder = (bytes) => Buffer.from(bytes.subarray(0, bytes.length - bytes.length % 2)).swap16();
+/**
+* The texts SheetJS can read from a part: its Latin-1 ("binary") view and, for BOM-marked parts,
+* the UTF-16 decodings of `cc2str` (little- and big-endian from byte 2, including its
+* `arr[1]/arr[2]` Buffer check) plus an extra odd-offset big-endian decode.
+*/
+const sheetJsTextViews = (content) => {
+	const bytes = Buffer.from(content.buffer, content.byteOffset, content.byteLength);
+	const latin1 = bytes.toString("latin1");
+	const littleEndian = bytes[0] === 255 && bytes[1] === 254;
+	const bigEndian = bytes[0] === 254 && bytes[1] === 255;
+	const offsetBigEndian = bytes[1] === 254 && bytes[2] === 255;
+	if (!littleEndian && !bigEndian && !offsetBigEndian) return [latin1];
+	return [
+		latin1,
+		bytes.subarray(2).toString("utf16le"),
+		swapUtf16ByteOrder(bytes.subarray(2)).toString("utf16le"),
+		swapUtf16ByteOrder(bytes.subarray(3)).toString("utf16le")
+	];
+};
+/**
+* SheetJS's own tag pattern (`tagregex1`, used for every part it parses): quoted values may hold
+* `<` and `>`. Each attempt stops at the next quote of its kind, so a scan stays linear.
+*/
+const sheetJsTagPattern = /<[/?]?[a-zA-Z0-9:_-]+(?:\s+[^"\s?<>/]+\s*=\s*(?:"[^"]*"|'[^']*'|[^'"<>\s=]+))*\s*[/?]?>/gm;
+/** SheetJS `attregexg`. */
+const sheetJsAttribute = /\s([^"\s?>/]+)\s*=\s*((?:")([^"]*)(?:")|(?:')([^']*)(?:')|([^'">\s]+))/g;
+/**
+* Port of SheetJS `parsexmltag`: exact-case keys (plus lower-cased copies), a namespace prefix
+* dropped, an unprefixed name cut at its first `_`, the last value winning. Values are raw.
+*/
+const parseSheetJsTag = (tag) => {
+	let end = 0;
+	for (; end < tag.length; end += 1) {
+		const code = tag.charCodeAt(end);
+		if (code === 32 || code === 10 || code === 13) break;
+	}
+	const attributes = /* @__PURE__ */ new Map();
+	if (end === tag.length) return {
+		head: tag,
+		attributes
+	};
+	for (const [match] of tag.matchAll(sheetJsAttribute)) {
+		const text = match.slice(1);
+		let equals = text.indexOf("=");
+		let name = text.slice(0, equals).trim();
+		while (text.charCodeAt(equals + 1) === 32) equals += 1;
+		const quoteCode = text.charCodeAt(equals + 1);
+		const quoted = quoteCode === 34 || quoteCode === 39 ? 1 : 0;
+		const value = text.slice(equals + 1 + quoted, text.length - quoted);
+		const colon = name.indexOf(":");
+		if (colon < 0) {
+			if (name.indexOf("_") > 0) name = name.slice(0, name.indexOf("_"));
+		} else {
+			const local = (colon === 5 && name.startsWith("xmlns") ? "xmlns" : "") + name.slice(colon + 1);
+			if (attributes.has(local) && name.slice(colon - 3, colon) === "ext") continue;
+			name = local;
+		}
+		attributes.set(name, value);
+		attributes.set(name.toLowerCase(), value);
+	}
+	return {
+		head: tag.slice(0, end),
+		attributes
+	};
+};
+/** Every tag SheetJS's pattern finds in `text`, in document order. */
+function* sheetJsTags(text) {
+	for (const [tag] of text.matchAll(sheetJsTagPattern)) yield parseSheetJsTag(tag);
+}
+/** SheetJS `strip_ns`: the first `<prefix:` (or `</prefix:`) loses its prefix. */
+const stripSheetJsNamespace = (head) => head.replace(/<(\/?)\w+:/, "<$1");
+/** SheetJS `utf8read` in Node: the Latin-1 ("binary") string read back as UTF-8. */
+const sheetJsUtf8Read = (binary) => Buffer.from(binary, "latin1").toString("utf8");
+const encodings = new Map([
+	["&quot;", "\""],
+	["&apos;", "'"],
+	["&gt;", ">"],
+	["&lt;", "<"],
+	["&amp;", "&"]
+]);
+/**
+* SheetJS `unescapexml` for text without CDATA, quirks included: entity names match ignoring
+* case but only lower-case ones map (`&QUOT;` becomes U+0000), `&#X41;` is read as decimal, and
+* numeric references wrap at U+FFFF. Returns `undefined` for text with a CDATA marker, which
+* SheetJS splits recursively (and never ends for an unterminated one).
+*/
+const sheetJsUnescapeXml = (text) => {
+	if (text.includes("<![CDATA[")) return void 0;
+	return text.replace(/&(?:quot|apos|gt|lt|amp|#x?([\da-fA-F]+));/gi, (entity, code) => encodings.get(entity) ?? String.fromCharCode(Number.parseInt(code ?? "", entity.includes("x") ? 16 : 10))).replace(/_x([\da-fA-F]{4})_/gi, (_, code) => String.fromCharCode(Number.parseInt(code, 16)));
+};
+/** An attribute value as SheetJS reads text attributes: `unescapexml(utf8read(raw))`. */
+const sheetJsAttributeText = (raw) => sheetJsUnescapeXml(sheetJsUtf8Read(raw));
+const isHighSurrogate = (code) => code >= 55296 && code <= 56319;
+const isLowSurrogate = (code) => code >= 56320 && code <= 57343;
+const codeEscape = (code) => `_x${code.toString(16).toUpperCase().padStart(4, "0")}_`;
+const escapedCharacters = new Map([
+	["&", "&amp;"],
+	["<", "&lt;"],
+	[">", "&gt;"],
+	["\"", "&quot;"]
+]);
+const startsCodeEscape = /_x[\da-fA-F]{4}_/iy;
+/**
+* Escape `text` for a double-quoted attribute of a generated UTF-8 part so that SheetJS's
+* `unescapexml(utf8read(…))` returns `text` exactly: markup characters become entities, an `_`
+* that would start an `_xHHHH_` code becomes `_x005F_`, and control characters, U+FFFE, U+FFFF,
+* and lone surrogates become `_xHHHH_` codes.
+*/
+const sheetJsAttributeEscape = (text) => {
+	let output = "";
+	for (let index = 0; index < text.length; index += 1) {
+		const character = text.charAt(index);
+		const code = text.charCodeAt(index);
+		const escaped = escapedCharacters.get(character);
+		if (escaped !== void 0) {
+			output += escaped;
+			continue;
+		}
+		if (character === "_") {
+			startsCodeEscape.lastIndex = index;
+			output += startsCodeEscape.test(text) ? codeEscape(code) : character;
+			continue;
+		}
+		if (isHighSurrogate(code) && isLowSurrogate(text.charCodeAt(index + 1))) {
+			output += text.slice(index, index + 2);
+			index += 1;
+			continue;
+		}
+		output += code < 32 || code === 65534 || code === 65535 || isHighSurrogate(code) || isLowSurrogate(code) ? codeEscape(code) : character;
+	}
+	return output;
+};
+/** SheetJS's own `XML_HEADER`, used for every generated part. */
+const sheetJsXmlHeader = "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"yes\"?>\r\n";
+const spreadsheetMainNamespace = "http://schemas.openxmlformats.org/spreadsheetml/2006/main";
+const officeDocumentRelationshipsNamespace = "http://schemas.openxmlformats.org/officeDocument/2006/relationships";
+const cdataMarker = "<![CDATA[";
+/** The two conversions SheetJS chains over cell and shared-string text. */
+const sheetJsTextSteps = [sheetJsUnescapeXml, sheetJsUtf8Read];
+/** Depth first, so at most `steps + 1` derived strings are alive at once. */
+const meetsCdata = (text, steps) => {
+	if (text.includes(cdataMarker)) return true;
+	if (steps === 0) return false;
+	return sheetJsTextSteps.some((step) => {
+		const next = step(text);
+		return next !== void 0 && next !== text && meetsCdata(next, steps - 1);
+	});
+};
+const isNameCode = (code) => code >= 48 && code <= 57 || code >= 65 && code <= 90 || code >= 97 && code <= 122 || code === 95 || code === 46 || code === 45;
+/** The end of the run of name characters (`[\w.-]`) starting at `start`. */
+const nameEnd = (text, start) => {
+	let end = start;
+	while (end < text.length && isNameCode(text.charCodeAt(end))) end += 1;
+	return end;
+};
+/**
+* `text` without its simple opening tags, `<(?:[\w.-]+:)?[\w.-]+>`: a superset of the tags SheetJS
+* removes before decoding (`<(?:\w+:)?(?:si|sstItem)>` in `parse_sst_xml`, `<(?:\w+:)?r>` in
+* `parse_rs`). One forward scan: a `<` that does not start such a tag is kept and the scan resumes
+* at the next `<`, so every character is read at most twice.
+*/
+const withoutSimpleTags = (text) => {
+	let output = "";
+	let kept = 0;
+	let index = text.indexOf("<");
+	while (index >= 0) {
+		let end = nameEnd(text, index + 1);
+		if (end > index + 1 && text.charCodeAt(end) === 58) {
+			const local = nameEnd(text, end + 1);
+			end = local > end + 1 ? local : -1;
+		}
+		if (end > index + 1 && text.charCodeAt(end) === 62) {
+			output += text.slice(kept, index);
+			kept = end + 1;
+			index = text.indexOf("<", kept);
+		} else index = text.indexOf("<", index + 1);
+	}
+	return output + text.slice(kept);
+};
+/** `<<` or `<!`: never written in worksheets or shared strings by Excel, LibreOffice, or Sheets. */
+const hasMarkupOpener = (text) => text.includes("<<") || text.includes("<!");
+/**
+* Whether SheetJS could meet a CDATA marker in this part. SheetJS's `unescapexml` handles CDATA by
+* recursing on a string two characters shorter and copying the whole tail at every level, so an
+* unterminated marker costs quadratic time and memory.
+*
+* What SheetJS hands to `unescapexml` in a worksheet or shared-strings part comes from the part
+* text through two kinds of transformation:
+*
+* - Decodes: raw (`<v>` of every cell), `utf8read(raw)` (shared and inline strings), and
+*   `utf8read(unescapexml(raw))` (cells of type `str`, decoded again after `utf8read`).
+*   `utf8read` keeps only the low byte of each character, so U+013C from `_x013C_`, `&#x13C;`, or
+*   `&#316;` becomes `<`.
+* - Tag removal before decoding: `parse_sst_xml` removes every `<si>`/`<sstItem>` opening tag
+*   from the whole shared-strings table, and `parse_rs` removes every `<r>` opening tag from rich
+*   text (after `utf8read`). Inline strings (`t="inlineStr"`) call `parse_si` without options,
+*   so their rich text is processed even with `cellHTML: false`. `A<<r>![CDATA[B` thus reaches
+*   `unescapexml` as `A<![CDATA[B`.
+*
+* So the check rejects a text view (`sheetJsTextViews`) when:
+*
+* - the view, or `utf8read` of it, contains `<<` or `<!`. A marker assembled by removing tags
+*   needs a literal `<` (in the view, or from `utf8read`) followed by a removed tag, or by `!`,
+*   and every removed tag starts with `<`;
+* - the view, or the view without any simple opening tag (`withoutSimpleTags`, a superset of
+*   SheetJS's removals), meets the marker raw or after any chain of up to two steps of
+*   `unescapexml` and `utf8read`, in any order (a superset of SheetJS's decode sequences).
+*
+* Every step is a linear pass, at most a few per view. The first rule deliberately fails closed:
+* it also rejects XML comments, `<!DOCTYPE`, and any other `<!…` declaration, which Excel,
+* LibreOffice, and Google Sheets never write in worksheets or shared strings (nor a literal `<<`).
+*/
+const sheetJsCouldReadCdata = (content) => sheetJsTextViews(content).some((view) => hasMarkupOpener(view) || hasMarkupOpener(sheetJsUtf8Read(view)) || meetsCdata(view, 2) || meetsCdata(withoutSimpleTags(view), 2));
+const xmlBoundary = new Set([
+	" ",
+	"	",
+	"\r",
+	"\n",
+	">"
+]);
+/** SheetJS `str_match_xml`: the first `<tag` element's inner text (exact prefix and case). */
+const sheetJsElementText = (text, tag) => {
+	const width = tag.length + 1;
+	let start = text.indexOf(`<${tag}`);
+	while (start >= 0 && start <= text.length - width && !xmlBoundary.has(text.charAt(start + width))) start = text.indexOf(`<${tag}`, start + 1);
+	if (start === -1) return void 0;
+	const contentStart = text.indexOf(">", start + tag.length);
+	if (contentStart === -1) return void 0;
+	const end = text.indexOf(`</${tag}>`, contentStart);
+	return end === -1 ? void 0 : text.slice(contentStart + 1, end);
+};
+/**
+* The `dc:title` of a core-properties part, read as SheetJS `parse_core_props` reads it, without
+* handing the part to SheetJS. A title holding CDATA is ignored.
+*/
+const coreTitle = (content) => {
+	if (content === void 0) return void 0;
+	const raw = sheetJsElementText(Buffer.from(content.buffer, content.byteOffset, content.byteLength).toString("utf8"), "dc:title");
+	const title = raw === void 0 ? void 0 : sheetJsUnescapeXml(raw);
+	return title !== void 0 && title.trim().length > 0 ? title : void 0;
+};
+//#endregion
+export { coreTitle, officeDocumentRelationshipsNamespace, parseSheetJsTag, sheetJsAttributeEscape, sheetJsAttributeText, sheetJsCouldReadCdata, sheetJsTags, sheetJsTextViews, sheetJsUnescapeXml, sheetJsUtf8Read, sheetJsXmlHeader, spreadsheetMainNamespace, stripSheetJsNamespace, withoutSimpleTags };
+
+//# sourceMappingURL=sheetjs-xml.mjs.map

package/dist/node/sheetjs-xml.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"sheetjs-xml.mjs","names":[],"sources":["../../src/node/sheetjs-xml.ts"],"sourcesContent":["import { Buffer } from 'node:buffer'\n\n/**\n * Ports of the SheetJS 0.20.3 XML helpers (`xlsx.mjs`) the extractor needs to read a part the way\n * SheetJS would: its tag pattern, `parsexmltag`, `strip_ns`, `utf8read`, and `unescapexml`.\n * Everything here is linear in the text it scans.\n */\n\nconst swapUtf16ByteOrder = (bytes: Buffer) =>\n  Buffer.from(bytes.subarray(0, bytes.length - (bytes.length % 2))).swap16()\n\n/**\n * The texts SheetJS can read from a part: its Latin-1 (\"binary\") view and, for BOM-marked parts,\n * the UTF-16 decodings of `cc2str` (little- and big-endian from byte 2, including its\n * `arr[1]/arr[2]` Buffer check) plus an extra odd-offset big-endian decode.\n */\nexport const sheetJsTextViews = (content: Uint8Array): ReadonlyArray<string> => {\n  const bytes = Buffer.from(content.buffer, content.byteOffset, content.byteLength)\n  const latin1 = bytes.toString('latin1')\n  const littleEndian = bytes[0] === 0xff && bytes[1] === 0xfe\n  const bigEndian = bytes[0] === 0xfe && bytes[1] === 0xff\n  const offsetBigEndian = bytes[1] === 0xfe && bytes[2] === 0xff\n\n  if (!littleEndian && !bigEndian && !offsetBigEndian) return [latin1]\n\n  return [\n    latin1,\n    bytes.subarray(2).toString('utf16le'),\n    swapUtf16ByteOrder(bytes.subarray(2)).toString('utf16le'),\n    swapUtf16ByteOrder(bytes.subarray(3)).toString('utf16le')\n  ]\n}\n\n/**\n * SheetJS's own tag pattern (`tagregex1`, used for every part it parses): quoted values may hold\n * `<` and `>`. Each attempt stops at the next quote of its kind, so a scan stays linear.\n */\nconst sheetJsTagPattern =\n  /<[/?]?[a-zA-Z0-9:_-]+(?:\\s+[^\"\\s?<>/]+\\s*=\\s*(?:\"[^\"]*\"|'[^']*'|[^'\"<>\\s=]+))*\\s*[/?]?>/gm\n\n/** SheetJS `attregexg`. */\nconst sheetJsAttribute = /\\s([^\"\\s?>/]+)\\s*=\\s*((?:\")([^\"]*)(?:\")|(?:')([^']*)(?:')|([^'\">\\s]+))/g\n\nexport type SheetJsTag = {\n  /** The tag up to its first space, line feed, or carriage return (SheetJS `y[0]`). */\n  readonly head: string\n  /** Raw (still escaped) attribute values by SheetJS key, plus lower-cased copies. */\n  readonly attributes: ReadonlyMap<string, string>\n}\n\n/**\n * Port of SheetJS `parsexmltag`: exact-case keys (plus lower-cased copies), a namespace prefix\n * dropped, an unprefixed name cut at its first `_`, the last value winning. Values are raw.\n */\nexport const parseSheetJsTag = (tag: string): SheetJsTag => {\n  let end = 0\n\n  for (; end < tag.length; end += 1) {\n    const code = tag.charCodeAt(end)\n\n    if (code === 32 || code === 10 || code === 13) break\n  }\n\n  const attributes = new Map<string, string>()\n\n  if (end === tag.length) return { head: tag, attributes }\n\n  for (const [match] of tag.matchAll(sheetJsAttribute)) {\n    const text = match.slice(1)\n    let equals = text.indexOf('=')\n    let name = text.slice(0, equals).trim()\n\n    while (text.charCodeAt(equals + 1) === 32) equals += 1\n\n    const quoteCode = text.charCodeAt(equals + 1)\n    const quoted = quoteCode === 34 || quoteCode === 39 ? 1 : 0\n    const value = text.slice(equals + 1 + quoted, text.length - quoted)\n    const colon = name.indexOf(':')\n\n    if (colon < 0) {\n      if (name.indexOf('_') > 0) name = name.slice(0, name.indexOf('_'))\n    } else {\n      const local = (colon === 5 && name.startsWith('xmlns') ? 'xmlns' : '') + name.slice(colon + 1)\n\n      if (attributes.has(local) && name.slice(colon - 3, colon) === 'ext') continue\n\n      name = local\n    }\n\n    attributes.set(name, value)\n    attributes.set(name.toLowerCase(), value)\n  }\n\n  return { head: tag.slice(0, end), attributes }\n}\n\n/** Every tag SheetJS's pattern finds in `text`, in document order. */\nexport function* sheetJsTags(text: string): Generator<SheetJsTag> {\n  for (const [tag] of text.matchAll(sheetJsTagPattern)) yield parseSheetJsTag(tag)\n}\n\n/** SheetJS `strip_ns`: the first `<prefix:` (or `</prefix:`) loses its prefix. */\nexport const stripSheetJsNamespace = (head: string) => head.replace(/<(\\/?)\\w+:/, '<$1')\n\n/** SheetJS `utf8read` in Node: the Latin-1 (\"binary\") string read back as UTF-8. */\nexport const sheetJsUtf8Read = (binary: string) => Buffer.from(binary, 'latin1').toString('utf8')\n\nconst encodings: ReadonlyMap<string, string> = new Map([\n  ['&quot;', '\"'],\n  ['&apos;', \"'\"],\n  ['&gt;', '>'],\n  ['&lt;', '<'],\n  ['&amp;', '&']\n])\n\n/**\n * SheetJS `unescapexml` for text without CDATA, quirks included: entity names match ignoring\n * case but only lower-case ones map (`&QUOT;` becomes U+0000), `&#X41;` is read as decimal, and\n * numeric references wrap at U+FFFF. Returns `undefined` for text with a CDATA marker, which\n * SheetJS splits recursively (and never ends for an unterminated one).\n */\nexport const sheetJsUnescapeXml = (text: string): string | undefined => {\n  if (text.includes('<![CDATA[')) return undefined\n\n  return text\n    .replace(\n      /&(?:quot|apos|gt|lt|amp|#x?([\\da-fA-F]+));/gi,\n      (entity, code: string | undefined) =>\n        encodings.get(entity) ??\n        String.fromCharCode(Number.parseInt(code ?? '', entity.includes('x') ? 16 : 10))\n    )\n    .replace(/_x([\\da-fA-F]{4})_/gi, (_, code: string) =>\n      String.fromCharCode(Number.parseInt(code, 16))\n    )\n}\n\n/** An attribute value as SheetJS reads text attributes: `unescapexml(utf8read(raw))`. */\nexport const sheetJsAttributeText = (raw: string) => sheetJsUnescapeXml(sheetJsUtf8Read(raw))\n\nconst isHighSurrogate = (code: number) => code >= 0xd800 && code <= 0xdbff\n\nconst isLowSurrogate = (code: number) => code >= 0xdc00 && code <= 0xdfff\n\nconst codeEscape = (code: number) => `_x${code.toString(16).toUpperCase().padStart(4, '0')}_`\n\nconst escapedCharacters: ReadonlyMap<string, string> = new Map([\n  ['&', '&amp;'],\n  ['<', '&lt;'],\n  ['>', '&gt;'],\n  ['\"', '&quot;']\n])\n\n// SheetJS matches `_xHHHH_` codes ignoring case (`coderegex`), so `_X0041_` is a code too.\nconst startsCodeEscape = /_x[\\da-fA-F]{4}_/iy\n\n/**\n * Escape `text` for a double-quoted attribute of a generated UTF-8 part so that SheetJS's\n * `unescapexml(utf8read(…))` returns `text` exactly: markup characters become entities, an `_`\n * that would start an `_xHHHH_` code becomes `_x005F_`, and control characters, U+FFFE, U+FFFF,\n * and lone surrogates become `_xHHHH_` codes.\n */\nexport const sheetJsAttributeEscape = (text: string) => {\n  let output = ''\n\n  for (let index = 0; index < text.length; index += 1) {\n    const character = text.charAt(index)\n    const code = text.charCodeAt(index)\n    const escaped = escapedCharacters.get(character)\n\n    if (escaped !== undefined) {\n      output += escaped\n      continue\n    }\n\n    if (character === '_') {\n      startsCodeEscape.lastIndex = index\n      output += startsCodeEscape.test(text) ? codeEscape(code) : character\n      continue\n    }\n\n    if (isHighSurrogate(code) && isLowSurrogate(text.charCodeAt(index + 1))) {\n      output += text.slice(index, index + 2)\n      index += 1\n      continue\n    }\n\n    output +=\n      code < 0x20 ||\n      code === 0xfffe ||\n      code === 0xffff ||\n      isHighSurrogate(code) ||\n      isLowSurrogate(code)\n        ? codeEscape(code)\n        : character\n  }\n\n  return output\n}\n\n/** SheetJS's own `XML_HEADER`, used for every generated part. */\nexport const sheetJsXmlHeader = '<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"yes\"?>\\r\\n'\n\nexport const spreadsheetMainNamespace = 'http://schemas.openxmlformats.org/spreadsheetml/2006/main'\n\nexport const officeDocumentRelationshipsNamespace =\n  'http://schemas.openxmlformats.org/officeDocument/2006/relationships'\n\nconst cdataMarker = '<![CDATA['\n\ntype TextStep = (text: string) => string | undefined\n\n/** The two conversions SheetJS chains over cell and shared-string text. */\nconst sheetJsTextSteps: ReadonlyArray<TextStep> = [sheetJsUnescapeXml, sheetJsUtf8Read]\n\n/** Depth first, so at most `steps + 1` derived strings are alive at once. */\nconst meetsCdata = (text: string, steps: number): boolean => {\n  if (text.includes(cdataMarker)) return true\n\n  if (steps === 0) return false\n\n  return sheetJsTextSteps.some(step => {\n    const next = step(text)\n\n    return next !== undefined && next !== text && meetsCdata(next, steps - 1)\n  })\n}\n\nconst isNameCode = (code: number) =>\n  (code >= 48 && code <= 57) || // 0-9\n  (code >= 65 && code <= 90) || // A-Z\n  (code >= 97 && code <= 122) || // a-z\n  code === 95 || // _\n  code === 46 || // .\n  code === 45 // -\n\n/** The end of the run of name characters (`[\\w.-]`) starting at `start`. */\nconst nameEnd = (text: string, start: number) => {\n  let end = start\n\n  while (end < text.length && isNameCode(text.charCodeAt(end))) end += 1\n\n  return end\n}\n\n/**\n * `text` without its simple opening tags, `<(?:[\\w.-]+:)?[\\w.-]+>`: a superset of the tags SheetJS\n * removes before decoding (`<(?:\\w+:)?(?:si|sstItem)>` in `parse_sst_xml`, `<(?:\\w+:)?r>` in\n * `parse_rs`). One forward scan: a `<` that does not start such a tag is kept and the scan resumes\n * at the next `<`, so every character is read at most twice.\n */\nexport const withoutSimpleTags = (text: string) => {\n  let output = ''\n  let kept = 0\n  let index = text.indexOf('<')\n\n  while (index >= 0) {\n    let end = nameEnd(text, index + 1)\n\n    if (end > index + 1 && text.charCodeAt(end) === 58) {\n      const local = nameEnd(text, end + 1)\n\n      end = local > end + 1 ? local : -1\n    }\n\n    if (end > index + 1 && text.charCodeAt(end) === 62) {\n      output += text.slice(kept, index)\n      kept = end + 1\n      index = text.indexOf('<', kept)\n    } else {\n      index = text.indexOf('<', index + 1)\n    }\n  }\n\n  return output + text.slice(kept)\n}\n\n/** `<<` or `<!`: never written in worksheets or shared strings by Excel, LibreOffice, or Sheets. */\nconst hasMarkupOpener = (text: string) => text.includes('<<') || text.includes('<!')\n\n/**\n * Whether SheetJS could meet a CDATA marker in this part. SheetJS's `unescapexml` handles CDATA by\n * recursing on a string two characters shorter and copying the whole tail at every level, so an\n * unterminated marker costs quadratic time and memory.\n *\n * What SheetJS hands to `unescapexml` in a worksheet or shared-strings part comes from the part\n * text through two kinds of transformation:\n *\n * - Decodes: raw (`<v>` of every cell), `utf8read(raw)` (shared and inline strings), and\n *   `utf8read(unescapexml(raw))` (cells of type `str`, decoded again after `utf8read`).\n *   `utf8read` keeps only the low byte of each character, so U+013C from `_x013C_`, `&#x13C;`, or\n *   `&#316;` becomes `<`.\n * - Tag removal before decoding: `parse_sst_xml` removes every `<si>`/`<sstItem>` opening tag\n *   from the whole shared-strings table, and `parse_rs` removes every `<r>` opening tag from rich\n *   text (after `utf8read`). Inline strings (`t=\"inlineStr\"`) call `parse_si` without options,\n *   so their rich text is processed even with `cellHTML: false`. `A<<r>![CDATA[B` thus reaches\n *   `unescapexml` as `A<![CDATA[B`.\n *\n * So the check rejects a text view (`sheetJsTextViews`) when:\n *\n * - the view, or `utf8read` of it, contains `<<` or `<!`. A marker assembled by removing tags\n *   needs a literal `<` (in the view, or from `utf8read`) followed by a removed tag, or by `!`,\n *   and every removed tag starts with `<`;\n * - the view, or the view without any simple opening tag (`withoutSimpleTags`, a superset of\n *   SheetJS's removals), meets the marker raw or after any chain of up to two steps of\n *   `unescapexml` and `utf8read`, in any order (a superset of SheetJS's decode sequences).\n *\n * Every step is a linear pass, at most a few per view. The first rule deliberately fails closed:\n * it also rejects XML comments, `<!DOCTYPE`, and any other `<!…` declaration, which Excel,\n * LibreOffice, and Google Sheets never write in worksheets or shared strings (nor a literal `<<`).\n */\nexport const sheetJsCouldReadCdata = (content: Uint8Array) =>\n  sheetJsTextViews(content).some(\n    view =>\n      hasMarkupOpener(view) ||\n      hasMarkupOpener(sheetJsUtf8Read(view)) ||\n      meetsCdata(view, 2) ||\n      meetsCdata(withoutSimpleTags(view), 2)\n  )\n\nconst xmlBoundary = new Set([' ', '\\t', '\\r', '\\n', '>'])\n\n/** SheetJS `str_match_xml`: the first `<tag` element's inner text (exact prefix and case). */\nconst sheetJsElementText = (text: string, tag: string) => {\n  const width = tag.length + 1\n  let start = text.indexOf(`<${tag}`)\n\n  while (start >= 0 && start <= text.length - width && !xmlBoundary.has(text.charAt(start + width)))\n    start = text.indexOf(`<${tag}`, start + 1)\n\n  if (start === -1) return undefined\n\n  const contentStart = text.indexOf('>', start + tag.length)\n\n  if (contentStart === -1) return undefined\n\n  const end = text.indexOf(`</${tag}>`, contentStart)\n\n  return end === -1 ? undefined : text.slice(contentStart + 1, end)\n}\n\n/**\n * The `dc:title` of a core-properties part, read as SheetJS `parse_core_props` reads it, without\n * handing the part to SheetJS. A title holding CDATA is ignored.\n */\nexport const coreTitle = (content: Uint8Array | undefined) => {\n  if (content === undefined) return undefined\n\n  const raw = sheetJsElementText(\n    Buffer.from(content.buffer, content.byteOffset, content.byteLength).toString('utf8'),\n    'dc:title'\n  )\n\n  const title = raw === undefined ? undefined : sheetJsUnescapeXml(raw)\n\n  return title !== undefined && title.trim().length > 0 ? title : undefined\n}\n"],"mappings":";;;;;;;AAQA,MAAM,sBAAsB,UAC1B,OAAO,KAAK,MAAM,SAAS,GAAG,MAAM,SAAU,MAAM,SAAS,CAAE,CAAC,EAAE,OAAO;;;;;;AAO3E,MAAa,oBAAoB,YAA+C;CAC9E,MAAM,QAAQ,OAAO,KAAK,QAAQ,QAAQ,QAAQ,YAAY,QAAQ,UAAU;CAChF,MAAM,SAAS,MAAM,SAAS,QAAQ;CACtC,MAAM,eAAe,MAAM,OAAO,OAAQ,MAAM,OAAO;CACvD,MAAM,YAAY,MAAM,OAAO,OAAQ,MAAM,OAAO;CACpD,MAAM,kBAAkB,MAAM,OAAO,OAAQ,MAAM,OAAO;CAE1D,IAAI,CAAC,gBAAgB,CAAC,aAAa,CAAC,iBAAiB,OAAO,CAAC,MAAM;CAEnE,OAAO;EACL;EACA,MAAM,SAAS,CAAC,EAAE,SAAS,SAAS;EACpC,mBAAmB,MAAM,SAAS,CAAC,CAAC,EAAE,SAAS,SAAS;EACxD,mBAAmB,MAAM,SAAS,CAAC,CAAC,EAAE,SAAS,SAAS;CAC1D;AACF;;;;;AAMA,MAAM,oBACJ;;AAGF,MAAM,mBAAmB;;;;;AAazB,MAAa,mBAAmB,QAA4B;CAC1D,IAAI,MAAM;CAEV,OAAO,MAAM,IAAI,QAAQ,OAAO,GAAG;EACjC,MAAM,OAAO,IAAI,WAAW,GAAG;EAE/B,IAAI,SAAS,MAAM,SAAS,MAAM,SAAS,IAAI;CACjD;CAEA,MAAM,6BAAa,IAAI,IAAoB;CAE3C,IAAI,QAAQ,IAAI,QAAQ,OAAO;EAAE,MAAM;EAAK;CAAW;CAEvD,KAAK,MAAM,CAAC,UAAU,IAAI,SAAS,gBAAgB,GAAG;EACpD,MAAM,OAAO,MAAM,MAAM,CAAC;EAC1B,IAAI,SAAS,KAAK,QAAQ,GAAG;EAC7B,IAAI,OAAO,KAAK,MAAM,GAAG,MAAM,EAAE,KAAK;EAEtC,OAAO,KAAK,WAAW,SAAS,CAAC,MAAM,IAAI,UAAU;EAErD,MAAM,YAAY,KAAK,WAAW,SAAS,CAAC;EAC5C,MAAM,SAAS,cAAc,MAAM,cAAc,KAAK,IAAI;EAC1D,MAAM,QAAQ,KAAK,MAAM,SAAS,IAAI,QAAQ,KAAK,SAAS,MAAM;EAClE,MAAM,QAAQ,KAAK,QAAQ,GAAG;EAE9B,IAAI,QAAQ;OACN,KAAK,QAAQ,GAAG,IAAI,GAAG,OAAO,KAAK,MAAM,GAAG,KAAK,QAAQ,GAAG,CAAC;EAAA,OAC5D;GACL,MAAM,SAAS,UAAU,KAAK,KAAK,WAAW,OAAO,IAAI,UAAU,MAAM,KAAK,MAAM,QAAQ,CAAC;GAE7F,IAAI,WAAW,IAAI,KAAK,KAAK,KAAK,MAAM,QAAQ,GAAG,KAAK,MAAM,OAAO;GAErE,OAAO;EACT;EAEA,WAAW,IAAI,MAAM,KAAK;EAC1B,WAAW,IAAI,KAAK,YAAY,GAAG,KAAK;CAC1C;CAEA,OAAO;EAAE,MAAM,IAAI,MAAM,GAAG,GAAG;EAAG;CAAW;AAC/C;;AAGA,UAAiB,YAAY,MAAqC;CAChE,KAAK,MAAM,CAAC,QAAQ,KAAK,SAAS,iBAAiB,GAAG,MAAM,gBAAgB,GAAG;AACjF;;AAGA,MAAa,yBAAyB,SAAiB,KAAK,QAAQ,cAAc,KAAK;;AAGvF,MAAa,mBAAmB,WAAmB,OAAO,KAAK,QAAQ,QAAQ,EAAE,SAAS,MAAM;AAEhG,MAAM,YAAyC,IAAI,IAAI;CACrD,CAAC,UAAU,IAAG;CACd,CAAC,UAAU,GAAG;CACd,CAAC,QAAQ,GAAG;CACZ,CAAC,QAAQ,GAAG;CACZ,CAAC,SAAS,GAAG;AACf,CAAC;;;;;;;AAQD,MAAa,sBAAsB,SAAqC;CACtE,IAAI,KAAK,SAAS,WAAW,GAAG,OAAO,KAAA;CAEvC,OAAO,KACJ,QACC,iDACC,QAAQ,SACP,UAAU,IAAI,MAAM,KACpB,OAAO,aAAa,OAAO,SAAS,QAAQ,IAAI,OAAO,SAAS,GAAG,IAAI,KAAK,EAAE,CAAC,CACnF,EACC,QAAQ,yBAAyB,GAAG,SACnC,OAAO,aAAa,OAAO,SAAS,MAAM,EAAE,CAAC,CAC/C;AACJ;;AAGA,MAAa,wBAAwB,QAAgB,mBAAmB,gBAAgB,GAAG,CAAC;AAE5F,MAAM,mBAAmB,SAAiB,QAAQ,SAAU,QAAQ;AAEpE,MAAM,kBAAkB,SAAiB,QAAQ,SAAU,QAAQ;AAEnE,MAAM,cAAc,SAAiB,KAAK,KAAK,SAAS,EAAE,EAAE,YAAY,EAAE,SAAS,GAAG,GAAG,EAAE;AAE3F,MAAM,oBAAiD,IAAI,IAAI;CAC7D,CAAC,KAAK,OAAO;CACb,CAAC,KAAK,MAAM;CACZ,CAAC,KAAK,MAAM;CACZ,CAAC,MAAK,QAAQ;AAChB,CAAC;AAGD,MAAM,mBAAmB;;;;;;;AAQzB,MAAa,0BAA0B,SAAiB;CACtD,IAAI,SAAS;CAEb,KAAK,IAAI,QAAQ,GAAG,QAAQ,KAAK,QAAQ,SAAS,GAAG;EACnD,MAAM,YAAY,KAAK,OAAO,KAAK;EACnC,MAAM,OAAO,KAAK,WAAW,KAAK;EAClC,MAAM,UAAU,kBAAkB,IAAI,SAAS;EAE/C,IAAI,YAAY,KAAA,GAAW;GACzB,UAAU;GACV;EACF;EAEA,IAAI,cAAc,KAAK;GACrB,iBAAiB,YAAY;GAC7B,UAAU,iBAAiB,KAAK,IAAI,IAAI,WAAW,IAAI,IAAI;GAC3D;EACF;EAEA,IAAI,gBAAgB,IAAI,KAAK,eAAe,KAAK,WAAW,QAAQ,CAAC,CAAC,GAAG;GACvE,UAAU,KAAK,MAAM,OAAO,QAAQ,CAAC;GACrC,SAAS;GACT;EACF;EAEA,UACE,OAAO,MACP,SAAS,SACT,SAAS,SACT,gBAAgB,IAAI,KACpB,eAAe,IAAI,IACf,WAAW,IAAI,IACf;CACR;CAEA,OAAO;AACT;;AAGA,MAAa,mBAAmB;AAEhC,MAAa,2BAA2B;AAExC,MAAa,uCACX;AAEF,MAAM,cAAc;;AAKpB,MAAM,mBAA4C,CAAC,oBAAoB,eAAe;;AAGtF,MAAM,cAAc,MAAc,UAA2B;CAC3D,IAAI,KAAK,SAAS,WAAW,GAAG,OAAO;CAEvC,IAAI,UAAU,GAAG,OAAO;CAExB,OAAO,iBAAiB,MAAK,SAAQ;EACnC,MAAM,OAAO,KAAK,IAAI;EAEtB,OAAO,SAAS,KAAA,KAAa,SAAS,QAAQ,WAAW,MAAM,QAAQ,CAAC;CAC1E,CAAC;AACH;AAEA,MAAM,cAAc,SACjB,QAAQ,MAAM,QAAQ,MACtB,QAAQ,MAAM,QAAQ,MACtB,QAAQ,MAAM,QAAQ,OACvB,SAAS,MACT,SAAS,MACT,SAAS;;AAGX,MAAM,WAAW,MAAc,UAAkB;CAC/C,IAAI,MAAM;CAEV,OAAO,MAAM,KAAK,UAAU,WAAW,KAAK,WAAW,GAAG,CAAC,GAAG,OAAO;CAErE,OAAO;AACT;;;;;;;AAQA,MAAa,qBAAqB,SAAiB;CACjD,IAAI,SAAS;CACb,IAAI,OAAO;CACX,IAAI,QAAQ,KAAK,QAAQ,GAAG;CAE5B,OAAO,SAAS,GAAG;EACjB,IAAI,MAAM,QAAQ,MAAM,QAAQ,CAAC;EAEjC,IAAI,MAAM,QAAQ,KAAK,KAAK,WAAW,GAAG,MAAM,IAAI;GAClD,MAAM,QAAQ,QAAQ,MAAM,MAAM,CAAC;GAEnC,MAAM,QAAQ,MAAM,IAAI,QAAQ;EAClC;EAEA,IAAI,MAAM,QAAQ,KAAK,KAAK,WAAW,GAAG,MAAM,IAAI;GAClD,UAAU,KAAK,MAAM,MAAM,KAAK;GAChC,OAAO,MAAM;GACb,QAAQ,KAAK,QAAQ,KAAK,IAAI;EAChC,OACE,QAAQ,KAAK,QAAQ,KAAK,QAAQ,CAAC;CAEvC;CAEA,OAAO,SAAS,KAAK,MAAM,IAAI;AACjC;;AAGA,MAAM,mBAAmB,SAAiB,KAAK,SAAS,IAAI,KAAK,KAAK,SAAS,IAAI;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCnF,MAAa,yBAAyB,YACpC,iBAAiB,OAAO,EAAE,MACxB,SACE,gBAAgB,IAAI,KACpB,gBAAgB,gBAAgB,IAAI,CAAC,KACrC,WAAW,MAAM,CAAC,KAClB,WAAW,kBAAkB,IAAI,GAAG,CAAC,CACzC;AAEF,MAAM,cAAc,IAAI,IAAI;CAAC;CAAK;CAAM;CAAM;CAAM;AAAG,CAAC;;AAGxD,MAAM,sBAAsB,MAAc,QAAgB;CACxD,MAAM,QAAQ,IAAI,SAAS;CAC3B,IAAI,QAAQ,KAAK,QAAQ,IAAI,KAAK;CAElC,OAAO,SAAS,KAAK,SAAS,KAAK,SAAS,SAAS,CAAC,YAAY,IAAI,KAAK,OAAO,QAAQ,KAAK,CAAC,GAC9F,QAAQ,KAAK,QAAQ,IAAI,OAAO,QAAQ,CAAC;CAE3C,IAAI,UAAU,IAAI,OAAO,KAAA;CAEzB,MAAM,eAAe,KAAK,QAAQ,KAAK,QAAQ,IAAI,MAAM;CAEzD,IAAI,iBAAiB,IAAI,OAAO,KAAA;CAEhC,MAAM,MAAM,KAAK,QAAQ,KAAK,IAAI,IAAI,YAAY;CAElD,OAAO,QAAQ,KAAK,KAAA,IAAY,KAAK,MAAM,eAAe,GAAG,GAAG;AAClE;;;;;AAMA,MAAa,aAAa,YAAoC;CAC5D,IAAI,YAAY,KAAA,GAAW,OAAO,KAAA;CAElC,MAAM,MAAM,mBACV,OAAO,KAAK,QAAQ,QAAQ,QAAQ,YAAY,QAAQ,UAAU,EAAE,SAAS,MAAM,GACnF,UACF;CAEA,MAAM,QAAQ,QAAQ,KAAA,IAAY,KAAA,IAAY,mBAAmB,GAAG;CAEpE,OAAO,UAAU,KAAA,KAAa,MAAM,KAAK,EAAE,SAAS,IAAI,QAAQ,KAAA;AAClE"}

package/dist/node/sheetjs.d.mts

@@ -0,0 +1,62 @@
+import { SheetJsUnavailableError } from "../errors.mjs";
+import { XlsxWorkbook } from "./xlsx-text.mjs";
+import { Effect } from "effect";
+
+//#region src/node/sheetjs.d.ts
+/** Loads the SheetJS module. The default is a lazy `import('xlsx')`. */
+type SheetJsLoader = () => Promise<unknown>;
+declare const defaultSheetJsLoader: SheetJsLoader;
+type SheetJs = {
+  readonly version: string; /** Parse workbook bytes with `sheetJsReadOptions`; the result is checked before use. */
+  readonly read: (bytes: Uint8Array) => unknown;
+};
+/**
+ * SheetJS `read` options: cell values and display text (`cell.w`) only.
+ *
+ * - `cellFormula: false`: no formula text. SheetJS otherwise copies a shifted master formula onto
+ *   every shared-formula dependent and scans every earlier array formula for each cell, before
+ *   any extractor budget runs. Cached values still render; formula-only cells render empty.
+ * - `cellHTML: false`: no rich-text HTML (`cell.h`) we never read. Inline strings ignore it
+ *   (SheetJS calls `parse_si` without options), so their rich text is still rendered; the CDATA
+ *   check (`sheetJsCouldReadCdata`) does not rely on this option.
+ * - `cellText: true`: keep the formatted display text (`cell.w`) the CSV uses. SheetJS formats
+ *   every styled cell inside `read`, with work proportional to the format code, so it only ever
+ *   reads the generated `xl/styles.xml` (codes of at most 255 characters, `xlsx-styles.ts`).
+ * - `cellNF`, `cellStyles`, `cellDates: false`: no format strings or style objects; dates stay
+ *   serial numbers whose `cell.w` carries the formatted date (`cellStyles` would also force
+ *   `sheetStubs`).
+ * - `sheetStubs: false`: no objects for empty cells.
+ * - `bookDeps`, `bookFiles`, `bookProps`, `bookSheets`, `bookVBA: false`: no calculation chain,
+ *   raw archive, or VBA blob, and a full parse (not the properties-only or names-only modes).
+ * - `dense: false`: sheets keyed by address, as `xlsx-text.ts` reads them.
+ * - `WTF: false`: per-sheet parse errors skip the sheet instead of throwing.
+ *
+ * A fresh copy is passed on every call because SheetJS writes defaults into the options object.
+ */
+declare const sheetJsReadOptions: {
+  readonly type: "array";
+  readonly cellFormula: false;
+  readonly cellHTML: false;
+  readonly cellText: true;
+  readonly cellNF: false;
+  readonly cellStyles: false;
+  readonly cellDates: false;
+  readonly sheetStubs: false;
+  readonly bookDeps: false;
+  readonly bookFiles: false;
+  readonly bookProps: false;
+  readonly bookSheets: false;
+  readonly bookVBA: false;
+  readonly dense: false;
+  readonly WTF: false;
+};
+/**
+ * Load SheetJS lazily and refuse anything that is not SheetJS 0.20.3 or newer. The version must be
+ * strict SemVer; anything else fails closed as `invalid`.
+ */
+declare const loadSheetJs: (loader: SheetJsLoader) => Effect.Effect<SheetJs, SheetJsUnavailableError, never>;
+/** Check the parsed workbook shape before reading it. */
+declare const asXlsxWorkbook: (parsed: unknown) => XlsxWorkbook | undefined;
+//#endregion
+export { SheetJs, SheetJsLoader, asXlsxWorkbook, defaultSheetJsLoader, loadSheetJs, sheetJsReadOptions };
+//# sourceMappingURL=sheetjs.d.mts.map

package/dist/node/sheetjs.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sheetjs.d.mts","names":[],"sources":["../../src/node/sheetjs.ts"],"mappings":";;;;;;KAKY,aAAA,SAAsB,OAAO;AAAA,cAE5B,oBAAA,EAAsB,aAAoC;AAAA,KAE3D,OAAA;EAAA,SACD,OAAA,UAL8B;EAAA,SAO9B,IAAA,GAAO,KAAA,EAAO,UAAU;AAAA;;;;AALoC;AAEvE;;;;;;;;;AAGmC;AA0BnC;;;;;;;;;cAAa,kBAAA;EAAA;;;;;;;;;;;;;;;;;;;;cA0EA,WAAA,GAAe,MAAA,EAAQ,aAAA,KAAa,MAAA,CAAA,MAAA,CAAA,OAAA,EAAA,uBAAA;;cA8CpC,cAAA,GAAkB,MAAA,cAAkB,YAAY"}

package/dist/node/sheetjs.mjs

@@ -0,0 +1,122 @@
+import { SheetJsUnavailableError, minimumSheetJsVersion } from "../errors.mjs";
+import { Effect, Predicate } from "effect";
+//#region src/node/sheetjs.ts
+const defaultSheetJsLoader = () => import("xlsx");
+/**
+* SheetJS `read` options: cell values and display text (`cell.w`) only.
+*
+* - `cellFormula: false`: no formula text. SheetJS otherwise copies a shifted master formula onto
+*   every shared-formula dependent and scans every earlier array formula for each cell, before
+*   any extractor budget runs. Cached values still render; formula-only cells render empty.
+* - `cellHTML: false`: no rich-text HTML (`cell.h`) we never read. Inline strings ignore it
+*   (SheetJS calls `parse_si` without options), so their rich text is still rendered; the CDATA
+*   check (`sheetJsCouldReadCdata`) does not rely on this option.
+* - `cellText: true`: keep the formatted display text (`cell.w`) the CSV uses. SheetJS formats
+*   every styled cell inside `read`, with work proportional to the format code, so it only ever
+*   reads the generated `xl/styles.xml` (codes of at most 255 characters, `xlsx-styles.ts`).
+* - `cellNF`, `cellStyles`, `cellDates: false`: no format strings or style objects; dates stay
+*   serial numbers whose `cell.w` carries the formatted date (`cellStyles` would also force
+*   `sheetStubs`).
+* - `sheetStubs: false`: no objects for empty cells.
+* - `bookDeps`, `bookFiles`, `bookProps`, `bookSheets`, `bookVBA: false`: no calculation chain,
+*   raw archive, or VBA blob, and a full parse (not the properties-only or names-only modes).
+* - `dense: false`: sheets keyed by address, as `xlsx-text.ts` reads them.
+* - `WTF: false`: per-sheet parse errors skip the sheet instead of throwing.
+*
+* A fresh copy is passed on every call because SheetJS writes defaults into the options object.
+*/
+const sheetJsReadOptions = {
+	type: "array",
+	cellFormula: false,
+	cellHTML: false,
+	cellText: true,
+	cellNF: false,
+	cellStyles: false,
+	cellDates: false,
+	sheetStubs: false,
+	bookDeps: false,
+	bookFiles: false,
+	bookProps: false,
+	bookSheets: false,
+	bookVBA: false,
+	dense: false,
+	WTF: false
+};
+/** SemVer 2.0.0: `major.minor.patch`, optional `-prerelease`, optional `+build`. */
+const semver = /^(0|[1-9]\d*)\.(0|[1-9]\d*)\.(0|[1-9]\d*)(?:-((?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\.(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\+[0-9a-zA-Z-]+(?:\.[0-9a-zA-Z-]+)*)?$/;
+const parseVersion = (version) => {
+	const match = semver.exec(version);
+	if (match === null) return void 0;
+	const release = [
+		Number(match[1]),
+		Number(match[2]),
+		Number(match[3])
+	];
+	return release.every(Number.isSafeInteger) ? {
+		release,
+		prerelease: match[4] !== void 0
+	} : void 0;
+};
+/**
+* Compare by SemVer precedence against the minimum release: a prerelease of 0.20.3 is below it,
+* prereleases of later releases are above it, and build metadata is ignored.
+*/
+const meetsMinimumVersion = (installed, minimum) => {
+	for (const [index, part] of installed.release.entries()) {
+		const required = minimum.release[index] ?? 0;
+		if (part !== required) return part > required;
+	}
+	return !installed.prerelease;
+};
+const isMissingModule = (cause) => Predicate.hasProperty(cause, "code") && (cause.code === "ERR_MODULE_NOT_FOUND" || cause.code === "MODULE_NOT_FOUND");
+/** ESM namespace, or a CommonJS interop namespace whose `default` is the module. */
+const sheetJsExports = (namespace) => {
+	if (Predicate.hasProperty(namespace, "read")) return namespace;
+	if (Predicate.hasProperty(namespace, "default") && Predicate.hasProperty(namespace.default, "read")) return namespace.default;
+};
+/**
+* Load SheetJS lazily and refuse anything that is not SheetJS 0.20.3 or newer. The version must be
+* strict SemVer; anything else fails closed as `invalid`.
+*/
+const loadSheetJs = (loader) => Effect.gen(function* () {
+	const exports = sheetJsExports(yield* Effect.tryPromise({
+		try: loader,
+		catch: (cause) => new SheetJsUnavailableError({
+			reason: isMissingModule(cause) ? "missing" : "invalid",
+			cause
+		})
+	}));
+	if (exports === void 0 || !Predicate.hasProperty(exports, "read") || !Predicate.isFunction(exports.read) || !Predicate.hasProperty(exports, "version") || !Predicate.isString(exports.version)) return yield* Effect.fail(new SheetJsUnavailableError({ reason: "invalid" }));
+	const { read, version } = exports;
+	const installed = parseVersion(version);
+	const minimum = parseVersion(minimumSheetJsVersion);
+	if (installed === void 0 || minimum === void 0) return yield* Effect.fail(new SheetJsUnavailableError({
+		reason: "invalid",
+		installedVersion: version
+	}));
+	if (!meetsMinimumVersion(installed, minimum)) return yield* Effect.fail(new SheetJsUnavailableError({
+		reason: "outdated",
+		installedVersion: version
+	}));
+	return {
+		version,
+		read: (bytes) => read(bytes, { ...sheetJsReadOptions })
+	};
+});
+/** Check the parsed workbook shape before reading it. */
+const asXlsxWorkbook = (parsed) => {
+	if (!Predicate.hasProperty(parsed, "SheetNames") || !Predicate.hasProperty(parsed, "Sheets") || !Array.isArray(parsed.SheetNames) || !Predicate.isObject(parsed.Sheets)) return void 0;
+	const sheetNames = [];
+	for (const name of parsed.SheetNames) {
+		if (!Predicate.isString(name)) return void 0;
+		sheetNames.push(name);
+	}
+	return {
+		SheetNames: sheetNames,
+		Sheets: parsed.Sheets
+	};
+};
+//#endregion
+export { asXlsxWorkbook, defaultSheetJsLoader, loadSheetJs, sheetJsReadOptions };
+
+//# sourceMappingURL=sheetjs.mjs.map

package/dist/node/sheetjs.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"sheetjs.mjs","names":[],"sources":["../../src/node/sheetjs.ts"],"sourcesContent":["import { Effect, Predicate } from 'effect'\nimport { minimumSheetJsVersion, SheetJsUnavailableError } from '../errors.ts'\nimport type { XlsxWorkbook } from './xlsx-text.ts'\n\n/** Loads the SheetJS module. The default is a lazy `import('xlsx')`. */\nexport type SheetJsLoader = () => Promise<unknown>\n\nexport const defaultSheetJsLoader: SheetJsLoader = () => import('xlsx')\n\nexport type SheetJs = {\n  readonly version: string\n  /** Parse workbook bytes with `sheetJsReadOptions`; the result is checked before use. */\n  readonly read: (bytes: Uint8Array) => unknown\n}\n\n/**\n * SheetJS `read` options: cell values and display text (`cell.w`) only.\n *\n * - `cellFormula: false`: no formula text. SheetJS otherwise copies a shifted master formula onto\n *   every shared-formula dependent and scans every earlier array formula for each cell, before\n *   any extractor budget runs. Cached values still render; formula-only cells render empty.\n * - `cellHTML: false`: no rich-text HTML (`cell.h`) we never read. Inline strings ignore it\n *   (SheetJS calls `parse_si` without options), so their rich text is still rendered; the CDATA\n *   check (`sheetJsCouldReadCdata`) does not rely on this option.\n * - `cellText: true`: keep the formatted display text (`cell.w`) the CSV uses. SheetJS formats\n *   every styled cell inside `read`, with work proportional to the format code, so it only ever\n *   reads the generated `xl/styles.xml` (codes of at most 255 characters, `xlsx-styles.ts`).\n * - `cellNF`, `cellStyles`, `cellDates: false`: no format strings or style objects; dates stay\n *   serial numbers whose `cell.w` carries the formatted date (`cellStyles` would also force\n *   `sheetStubs`).\n * - `sheetStubs: false`: no objects for empty cells.\n * - `bookDeps`, `bookFiles`, `bookProps`, `bookSheets`, `bookVBA: false`: no calculation chain,\n *   raw archive, or VBA blob, and a full parse (not the properties-only or names-only modes).\n * - `dense: false`: sheets keyed by address, as `xlsx-text.ts` reads them.\n * - `WTF: false`: per-sheet parse errors skip the sheet instead of throwing.\n *\n * A fresh copy is passed on every call because SheetJS writes defaults into the options object.\n */\nexport const sheetJsReadOptions = {\n  type: 'array',\n  cellFormula: false,\n  cellHTML: false,\n  cellText: true,\n  cellNF: false,\n  cellStyles: false,\n  cellDates: false,\n  sheetStubs: false,\n  bookDeps: false,\n  bookFiles: false,\n  bookProps: false,\n  bookSheets: false,\n  bookVBA: false,\n  dense: false,\n  WTF: false\n} as const\n\n/** SemVer 2.0.0: `major.minor.patch`, optional `-prerelease`, optional `+build`. */\nconst semver =\n  /^(0|[1-9]\\d*)\\.(0|[1-9]\\d*)\\.(0|[1-9]\\d*)(?:-((?:0|[1-9]\\d*|\\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\\.(?:0|[1-9]\\d*|\\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\\+[0-9a-zA-Z-]+(?:\\.[0-9a-zA-Z-]+)*)?$/\n\ntype ParsedVersion = {\n  readonly release: readonly [number, number, number]\n  readonly prerelease: boolean\n}\n\nconst parseVersion = (version: string): ParsedVersion | undefined => {\n  const match = semver.exec(version)\n\n  if (match === null) return undefined\n\n  const release = [Number(match[1]), Number(match[2]), Number(match[3])] as const\n\n  return release.every(Number.isSafeInteger)\n    ? { release, prerelease: match[4] !== undefined }\n    : undefined\n}\n\n/**\n * Compare by SemVer precedence against the minimum release: a prerelease of 0.20.3 is below it,\n * prereleases of later releases are above it, and build metadata is ignored.\n */\nconst meetsMinimumVersion = (installed: ParsedVersion, minimum: ParsedVersion) => {\n  for (const [index, part] of installed.release.entries()) {\n    const required = minimum.release[index] ?? 0\n\n    if (part !== required) return part > required\n  }\n\n  return !installed.prerelease\n}\n\nconst isMissingModule = (cause: unknown) =>\n  Predicate.hasProperty(cause, 'code') &&\n  (cause.code === 'ERR_MODULE_NOT_FOUND' || cause.code === 'MODULE_NOT_FOUND')\n\n/** ESM namespace, or a CommonJS interop namespace whose `default` is the module. */\nconst sheetJsExports = (namespace: unknown): object | undefined => {\n  if (Predicate.hasProperty(namespace, 'read')) return namespace\n\n  if (\n    Predicate.hasProperty(namespace, 'default') &&\n    Predicate.hasProperty(namespace.default, 'read')\n  )\n    return namespace.default\n\n  return undefined\n}\n\n/**\n * Load SheetJS lazily and refuse anything that is not SheetJS 0.20.3 or newer. The version must be\n * strict SemVer; anything else fails closed as `invalid`.\n */\nexport const loadSheetJs = (loader: SheetJsLoader) =>\n  Effect.gen(function* () {\n    const namespace = yield* Effect.tryPromise({\n      try: loader,\n      catch: cause =>\n        new SheetJsUnavailableError({\n          reason: isMissingModule(cause) ? 'missing' : 'invalid',\n          cause\n        })\n    })\n\n    const exports = sheetJsExports(namespace)\n\n    if (\n      exports === undefined ||\n      !Predicate.hasProperty(exports, 'read') ||\n      !Predicate.isFunction(exports.read) ||\n      !Predicate.hasProperty(exports, 'version') ||\n      !Predicate.isString(exports.version)\n    )\n      return yield* Effect.fail(new SheetJsUnavailableError({ reason: 'invalid' }))\n\n    const { read, version } = exports\n    const installed = parseVersion(version)\n    const minimum = parseVersion(minimumSheetJsVersion)\n\n    // A version that is not strict SemVer is not a SheetJS release we can vouch for.\n    if (installed === undefined || minimum === undefined)\n      return yield* Effect.fail(\n        new SheetJsUnavailableError({ reason: 'invalid', installedVersion: version })\n      )\n\n    if (!meetsMinimumVersion(installed, minimum))\n      return yield* Effect.fail(\n        new SheetJsUnavailableError({ reason: 'outdated', installedVersion: version })\n      )\n\n    const sheetJs: SheetJs = {\n      version,\n      read: bytes => read(bytes, { ...sheetJsReadOptions })\n    }\n\n    return sheetJs\n  })\n\n/** Check the parsed workbook shape before reading it. */\nexport const asXlsxWorkbook = (parsed: unknown): XlsxWorkbook | undefined => {\n  if (\n    !Predicate.hasProperty(parsed, 'SheetNames') ||\n    !Predicate.hasProperty(parsed, 'Sheets') ||\n    !Array.isArray(parsed.SheetNames) ||\n    !Predicate.isObject(parsed.Sheets)\n  )\n    return undefined\n\n  const sheetNames: Array<string> = []\n\n  for (const name of parsed.SheetNames) {\n    if (!Predicate.isString(name)) return undefined\n\n    sheetNames.push(name)\n  }\n\n  return { SheetNames: sheetNames, Sheets: parsed.Sheets }\n}\n"],"mappings":";;;AAOA,MAAa,6BAA4C,OAAO;;;;;;;;;;;;;;;;;;;;;;;;AA+BhE,MAAa,qBAAqB;CAChC,MAAM;CACN,aAAa;CACb,UAAU;CACV,UAAU;CACV,QAAQ;CACR,YAAY;CACZ,WAAW;CACX,YAAY;CACZ,UAAU;CACV,WAAW;CACX,WAAW;CACX,YAAY;CACZ,SAAS;CACT,OAAO;CACP,KAAK;AACP;;AAGA,MAAM,SACJ;AAOF,MAAM,gBAAgB,YAA+C;CACnE,MAAM,QAAQ,OAAO,KAAK,OAAO;CAEjC,IAAI,UAAU,MAAM,OAAO,KAAA;CAE3B,MAAM,UAAU;EAAC,OAAO,MAAM,EAAE;EAAG,OAAO,MAAM,EAAE;EAAG,OAAO,MAAM,EAAE;CAAC;CAErE,OAAO,QAAQ,MAAM,OAAO,aAAa,IACrC;EAAE;EAAS,YAAY,MAAM,OAAO,KAAA;CAAU,IAC9C,KAAA;AACN;;;;;AAMA,MAAM,uBAAuB,WAA0B,YAA2B;CAChF,KAAK,MAAM,CAAC,OAAO,SAAS,UAAU,QAAQ,QAAQ,GAAG;EACvD,MAAM,WAAW,QAAQ,QAAQ,UAAU;EAE3C,IAAI,SAAS,UAAU,OAAO,OAAO;CACvC;CAEA,OAAO,CAAC,UAAU;AACpB;AAEA,MAAM,mBAAmB,UACvB,UAAU,YAAY,OAAO,MAAM,MAClC,MAAM,SAAS,0BAA0B,MAAM,SAAS;;AAG3D,MAAM,kBAAkB,cAA2C;CACjE,IAAI,UAAU,YAAY,WAAW,MAAM,GAAG,OAAO;CAErD,IACE,UAAU,YAAY,WAAW,SAAS,KAC1C,UAAU,YAAY,UAAU,SAAS,MAAM,GAE/C,OAAO,UAAU;AAGrB;;;;;AAMA,MAAa,eAAe,WAC1B,OAAO,IAAI,aAAa;CAUtB,MAAM,UAAU,eAAe,OATN,OAAO,WAAW;EACzC,KAAK;EACL,QAAO,UACL,IAAI,wBAAwB;GAC1B,QAAQ,gBAAgB,KAAK,IAAI,YAAY;GAC7C;EACF,CAAC;CACL,CAAC,CAEuC;CAExC,IACE,YAAY,KAAA,KACZ,CAAC,UAAU,YAAY,SAAS,MAAM,KACtC,CAAC,UAAU,WAAW,QAAQ,IAAI,KAClC,CAAC,UAAU,YAAY,SAAS,SAAS,KACzC,CAAC,UAAU,SAAS,QAAQ,OAAO,GAEnC,OAAO,OAAO,OAAO,KAAK,IAAI,wBAAwB,EAAE,QAAQ,UAAU,CAAC,CAAC;CAE9E,MAAM,EAAE,MAAM,YAAY;CAC1B,MAAM,YAAY,aAAa,OAAO;CACtC,MAAM,UAAU,aAAa,qBAAqB;CAGlD,IAAI,cAAc,KAAA,KAAa,YAAY,KAAA,GACzC,OAAO,OAAO,OAAO,KACnB,IAAI,wBAAwB;EAAE,QAAQ;EAAW,kBAAkB;CAAQ,CAAC,CAC9E;CAEF,IAAI,CAAC,oBAAoB,WAAW,OAAO,GACzC,OAAO,OAAO,OAAO,KACnB,IAAI,wBAAwB;EAAE,QAAQ;EAAY,kBAAkB;CAAQ,CAAC,CAC/E;CAOF,OAAO;EAJL;EACA,OAAM,UAAS,KAAK,OAAO,EAAE,GAAG,mBAAmB,CAAC;CAGzC;AACf,CAAC;;AAGH,MAAa,kBAAkB,WAA8C;CAC3E,IACE,CAAC,UAAU,YAAY,QAAQ,YAAY,KAC3C,CAAC,UAAU,YAAY,QAAQ,QAAQ,KACvC,CAAC,MAAM,QAAQ,OAAO,UAAU,KAChC,CAAC,UAAU,SAAS,OAAO,MAAM,GAEjC,OAAO,KAAA;CAET,MAAM,aAA4B,CAAC;CAEnC,KAAK,MAAM,QAAQ,OAAO,YAAY;EACpC,IAAI,CAAC,UAAU,SAAS,IAAI,GAAG,OAAO,KAAA;EAEtC,WAAW,KAAK,IAAI;CACtB;CAEA,OAAO;EAAE,YAAY;EAAY,QAAQ,OAAO;CAAO;AACzD"}

package/dist/node/worker-admission.d.mts

@@ -0,0 +1,58 @@
+import { Effect } from "effect";
+
+//#region src/node/worker-admission.d.ts
+/**
+ * Worker admission. Every Node `FileExtractor` layer in a JavaScript realm (the main thread, or
+ * each worker thread or `vm` context that builds the layer) shares one pool of
+ * `processWorkerLimit` worker slots, however often the layer is built (a per-request
+ * `Effect.provide` builds a new one each time). The pool lives on `globalThis` under a versioned
+ * `Symbol.for` key and holds only plain data and plain callbacks, so duplicated copies of this
+ * package (and of Effect) in one realm share it. Each layer also has its own pool of
+ * `maxConcurrentWorkers` slots (at most `processWorkerLimit`), which can only lower that layer's
+ * share. An extraction takes its layer's slot, then a realm slot, and waits for both until its
+ * admission deadline; a slot is released only after its worker has terminated.
+ *
+ * Admission is first come, first served: a freed slot is handed to the longest-waiting extraction
+ * whose deadline has not passed, and a new extraction takes a free slot only when nobody waits.
+ */
+/** Worker slots shared by every layer in the realm. */
+declare const processWorkerLimit = 4;
+/**
+ * Takes a freed slot for its waiter and returns `true`, or returns `false` when the waiter's
+ * deadline has passed (it then fails with `busy`). Either way it leaves the queue. It never runs
+ * the waiter's fiber: the resume is deferred to a microtask, so a waiter that finishes at once
+ * cannot release (and hand off) again inside this call, and a long queue drains in constant stack.
+ */
+type SlotHandOff = () => boolean;
+/**
+ * A counting pool of worker slots with a FIFO queue (a `Set` iterates in insertion order). While
+ * anyone waits, every slot is taken: a released slot passes to a waiter without being counted
+ * free.
+ */
+type SlotPool = {
+  readonly capacity: number;
+  active: number;
+  readonly waiters: Set<SlotHandOff>;
+};
+declare const makeSlotPool: (capacity: number) => SlotPool;
+/** The realm-wide pool, created by the first copy of this module that asks for it. */
+declare const processSlotPool: () => SlotPool;
+/** Running and waiting extractions in the realm-wide pool. */
+declare const processAdmissionSnapshot: () => {
+  active: number;
+  waiting: number;
+};
+/**
+ * Run `self` holding one slot of `pool`, waiting for it until `deadline` (epoch ms) and failing
+ * with `busy()` after that. The deadline is checked again whenever a slot would be taken, by a new
+ * extraction or by a hand-off, so a late wake-up cannot turn an expired wait into admission.
+ *
+ * Taking a slot and registering its release happen without an interruption point between them,
+ * so a slot is never leaked; only the wait is interruptible. A slot handed to a waiter whose
+ * fiber is interrupted before it resumes is released again. The wait uses a real timer, not the
+ * Effect `Clock`, so a test clock cannot stall it.
+ */
+declare const withSlot: <E2>(pool: SlotPool, deadline: number, busy: () => E2) => <A, E, R>(self: Effect.Effect<A, E, R>) => Effect.Effect<A, E | E2, R>;
+//#endregion
+export { SlotHandOff, SlotPool, makeSlotPool, processAdmissionSnapshot, processSlotPool, processWorkerLimit, withSlot };
+//# sourceMappingURL=worker-admission.d.mts.map

package/dist/node/worker-admission.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"worker-admission.d.mts","names":[],"sources":["../../src/node/worker-admission.ts"],"mappings":";;;;;AAmBA;;;;AAA+B;AAQ/B;;;;AAAuB;AAOvB;;;cAfa,kBAAA;;;;;;;KAQD,WAAA;AAaZ;;;;AAIE;AAJF,KANY,QAAA;EAAA,SACD,QAAA;EACT,MAAA;EAAA,SACS,OAAA,EAAS,GAAG,CAAC,WAAA;AAAA;AAAA,cAGX,YAAA,GAAgB,QAAA,aAAmB,QAI9C;;cAcW,eAAA,QAAsB,QAUlC;;cAGY,wBAAA;EAIZ,MAAA;EAAA,OAAA;AAAA;;;;;;;;;;;cAsBY,QAAA,OACN,IAAA,EAAM,QAAA,EAAU,QAAA,UAAkB,IAAA,QAAY,EAAA,eACzC,IAAA,EAAM,MAAA,CAAO,MAAA,CAAO,CAAA,EAAG,CAAA,EAAG,CAAA,MAAK,MAAA,CAAO,MAAA,CAAO,CAAA,EAAG,CAAA,GAAI,EAAA,EAAI,CAAA"}