@nac3/forge-cli 0.2.0-alpha.1

package/dist/reader/parsers/pdf.d.ts

@@ -0,0 +1,39 @@
+/**
+ * PDF parser for the document reader (V1.16 of HITO 1).
+ *
+ * Uses pdfjs-dist (Mozilla's pdf.js). PDF is dense + complex
+ * to hand-roll, and ADIP specs really need it -- accepting
+ * the ~3MB dependency was Pablo-approved.
+ *
+ * Pipeline:
+ *   1. pdfjs.getDocument(buffer).promise -> PDFDocumentProxy
+ *   2. for each page in document order:
+ *        page.getTextContent() -> text items grouped by y
+ *        coordinate -> reconstruct lines -> reconstruct
+ *        paragraphs by blank-line breaks.
+ *   3. paragraph blocks; one section per page (with heading
+ *      "Pagina N"). The first non-empty line of page 1 becomes
+ *      the document title.
+ *
+ * Voice-readback choices:
+ *   - One section per page so users can navigate "leeme la
+ *     pagina 3".
+ *   - "Pagina N" heading is concise enough not to clutter TTS.
+ *   - Text items with very small font sizes (e.g. footnotes)
+ *     fold into the surrounding paragraph -- the voice reader
+ *     does not need typographic structure.
+ *
+ * Pathological inputs (encrypted PDFs, malformed XObjects)
+ * surface the pdfjs error message in a wrapper -- the voice
+ * reader stays alive.
+ *
+ * ASCII-only.
+ */
+import type { NormalisedDocument } from '../types.js';
+export interface ParsePdfOptions {
+    buffer: Buffer;
+    filename: string;
+    language?: string;
+}
+export declare function parsePdf(opts: ParsePdfOptions): Promise<NormalisedDocument>;
+//# sourceMappingURL=pdf.d.ts.map

package/dist/reader/parsers/pdf.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pdf.d.ts","sourceRoot":"","sources":["../../../src/reader/parsers/pdf.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,KAAK,EAAE,kBAAkB,EAAkB,MAAM,aAAa,CAAC;AAEtE,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AA+HD,wBAAsB,QAAQ,CAAC,IAAI,EAAE,eAAe,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAmFjF"}

package/dist/reader/parsers/pdf.js

@@ -0,0 +1,220 @@
+/**
+ * PDF parser for the document reader (V1.16 of HITO 1).
+ *
+ * Uses pdfjs-dist (Mozilla's pdf.js). PDF is dense + complex
+ * to hand-roll, and ADIP specs really need it -- accepting
+ * the ~3MB dependency was Pablo-approved.
+ *
+ * Pipeline:
+ *   1. pdfjs.getDocument(buffer).promise -> PDFDocumentProxy
+ *   2. for each page in document order:
+ *        page.getTextContent() -> text items grouped by y
+ *        coordinate -> reconstruct lines -> reconstruct
+ *        paragraphs by blank-line breaks.
+ *   3. paragraph blocks; one section per page (with heading
+ *      "Pagina N"). The first non-empty line of page 1 becomes
+ *      the document title.
+ *
+ * Voice-readback choices:
+ *   - One section per page so users can navigate "leeme la
+ *     pagina 3".
+ *   - "Pagina N" heading is concise enough not to clutter TTS.
+ *   - Text items with very small font sizes (e.g. footnotes)
+ *     fold into the surrounding paragraph -- the voice reader
+ *     does not need typographic structure.
+ *
+ * Pathological inputs (encrypted PDFs, malformed XObjects)
+ * surface the pdfjs error message in a wrapper -- the voice
+ * reader stays alive.
+ *
+ * ASCII-only.
+ */
+/** Reconstruct lines + paragraphs from a page's text items.
+ *
+ * pdf.js gives items in approximate reading order but with no
+ * paragraph structure. We:
+ *   1. Group items by y-coordinate (within 2 PDF units).
+ *   2. Sort each group by x-coordinate so left-to-right order
+ *      is preserved.
+ *   3. Sort groups top-to-bottom (y descending; PDF coords
+ *      grow upward).
+ *   4. Detect paragraph breaks when consecutive lines have
+ *      large y-gaps (>= 1.5 * median line height) OR when a
+ *      line ends with sentence punctuation followed by an
+ *      indented next line.
+ */
+function reconstructParagraphs(items) {
+    if (items.length === 0)
+        return [];
+    /* Group by quantised y (round to 2 PDF units). */
+    const buckets = new Map();
+    for (const it of items) {
+        if (!it.str || !it.transform || it.transform.length < 6)
+            continue;
+        const y = Math.round((it.transform[5] ?? 0) / 2) * 2;
+        if (!buckets.has(y))
+            buckets.set(y, []);
+        buckets.get(y).push(it);
+    }
+    /* Build lines top to bottom (y descending). */
+    const ys = Array.from(buckets.keys()).sort((a, b) => b - a);
+    const lines = [];
+    for (const y of ys) {
+        const bucket = buckets.get(y);
+        bucket.sort((a, b) => (a.transform[4] ?? 0) - (b.transform[4] ?? 0));
+        const text = bucket.map((it) => it.str).join('').replace(/\s+/g, ' ').trim();
+        if (text)
+            lines.push({ y, text });
+    }
+    if (lines.length === 0)
+        return [];
+    /* Median gap between consecutive lines. */
+    const gaps = [];
+    for (let i = 1; i < lines.length; i += 1) {
+        gaps.push((lines[i - 1].y) - (lines[i].y));
+    }
+    gaps.sort((a, b) => a - b);
+    const median = gaps[Math.floor(gaps.length / 2)] ?? 12;
+    const paragraphBreakAt = median * 1.6;
+    /* Build paragraphs. */
+    const paras = [];
+    let buf = lines[0].text;
+    for (let i = 1; i < lines.length; i += 1) {
+        const gap = (lines[i - 1].y) - (lines[i].y);
+        if (gap >= paragraphBreakAt) {
+            if (buf.trim())
+                paras.push(buf.trim());
+            buf = lines[i].text;
+        }
+        else {
+            /* Same paragraph -- hyphenated line break joins without
+               a space. */
+            if (buf.endsWith('-') && /^[a-zA-Z]/.test(lines[i].text)) {
+                buf = buf.slice(0, -1) + lines[i].text;
+            }
+            else {
+                buf += ' ' + lines[i].text;
+            }
+        }
+    }
+    if (buf.trim())
+        paras.push(buf.trim());
+    return paras;
+}
+/* Dynamic import wrapper -- pdfjs-dist is ESM only. */
+async function loadPdfjs() {
+    /* eslint-disable-next-line @typescript-eslint/no-explicit-any */
+    return await import('pdfjs-dist/legacy/build/pdf.mjs');
+}
+/** Resolve the path to pdf.worker.mjs inside the installed
+ *  pdfjs-dist. pdfjs v4 requires workerSrc to be set even
+ *  when disableWorker is true (the "fake worker" still loads
+ *  the module). */
+async function resolveWorkerSrc() {
+    /* createRequire is the most reliable cross-platform way to
+       find an installed module's actual file path under both
+       ESM and CJS. The ESM import shape varies across TS lib
+       versions; cast through unknown to get a stable handle. */
+    const mod = await import('node:module');
+    const createRequire = mod.createRequire
+        ?? mod.default.createRequire;
+    const require = createRequire(import.meta.url);
+    return require.resolve('pdfjs-dist/legacy/build/pdf.worker.mjs');
+}
+async function getDocument(buffer) {
+    const pdfjs = await loadPdfjs();
+    if (pdfjs.GlobalWorkerOptions && !pdfjs.GlobalWorkerOptions.workerSrc) {
+        pdfjs.GlobalWorkerOptions.workerSrc = await resolveWorkerSrc();
+    }
+    const loadingTask = pdfjs.getDocument({
+        data: new Uint8Array(buffer),
+        isEvalSupported: false,
+        useSystemFonts: true,
+    });
+    return await loadingTask.promise;
+}
+/* ----- document assembly ----- */
+export async function parsePdf(opts) {
+    let doc;
+    try {
+        doc = await getDocument(opts.buffer);
+    }
+    catch (err) {
+        const msg = err instanceof Error ? err.message : String(err);
+        throw new Error('pdf: pdfjs could not open the document: ' + msg);
+    }
+    const sections = [];
+    let blockCounter = 0;
+    function nextId(prefix) {
+        blockCounter += 1;
+        return 's-' + prefix + '-' + blockCounter;
+    }
+    let title = '';
+    for (let pageNum = 1; pageNum <= doc.numPages; pageNum += 1) {
+        let page;
+        let content;
+        try {
+            page = await doc.getPage(pageNum);
+            content = await page.getTextContent();
+        }
+        catch (err) {
+            /* Skip the page rather than aborting -- voice reader
+               degrades to "pagina N: contenido ilegible". */
+            sections.push({
+                id: 's-' + pageNum,
+                heading: 'Pagina ' + pageNum,
+                level: 1,
+                blocks: [
+                    { kind: 'heading', level: 1, text: 'Pagina ' + pageNum, id: nextId('h') },
+                    { kind: 'paragraph', level: 0, text: 'Contenido ilegible.', id: nextId('p') },
+                ],
+            });
+            continue;
+        }
+        const paras = reconstructParagraphs(content.items);
+        if (pageNum === 1 && paras.length > 0 && !title) {
+            /* First non-empty line of page 1 becomes the title. */
+            const firstLine = paras[0].split(/[\n.;]/)[0].trim();
+            if (firstLine && firstLine.length < 120)
+                title = firstLine;
+        }
+        const heading = 'Pagina ' + pageNum;
+        const blocks = [
+            { kind: 'heading', level: 1, text: heading, id: nextId('h') },
+        ];
+        for (const p of paras) {
+            blocks.push({
+                kind: 'paragraph',
+                level: 0,
+                text: p,
+                id: nextId('p'),
+            });
+        }
+        sections.push({
+            id: 's-' + pageNum,
+            heading,
+            level: 1,
+            blocks,
+        });
+    }
+    if (sections.length === 0) {
+        sections.push({ id: 's-0', heading: '', level: 0, blocks: [] });
+    }
+    const docId = (opts.filename.split(/[\\/]/).pop() || 'doc')
+        .replace(/\.[^.]+$/, '')
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, '-')
+        .replace(/^-+|-+$/g, '') || 'doc';
+    const out = {
+        id: docId,
+        filename: opts.filename,
+        format: 'pdf',
+        bytes: opts.buffer.length,
+        title: title || ('PDF (' + doc.numPages + ' pages)'),
+        sections,
+    };
+    if (opts.language !== undefined)
+        out.language = opts.language;
+    return out;
+}
+//# sourceMappingURL=pdf.js.map

package/dist/reader/parsers/pdf.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pdf.js","sourceRoot":"","sources":["../../../src/reader/parsers/pdf.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAkCH;;;;;;;;;;;;;GAaG;AACH,SAAS,qBAAqB,CAAC,KAAoB;IACjD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,kDAAkD;IAClD,MAAM,OAAO,GAAG,IAAI,GAAG,EAAyB,CAAC;IACjD,KAAK,MAAM,EAAE,IAAI,KAAK,EAAE,CAAC;QACvB,IAAI,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC,SAAS,IAAI,EAAE,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC;YAAE,SAAS;QAClE,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;YAAE,OAAO,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC3B,CAAC;IACD,+CAA+C;IAC/C,MAAM,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5D,MAAM,KAAK,GAAkC,EAAE,CAAC;IAChD,KAAK,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAE,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrE,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;QAC7E,IAAI,IAAI;YAAE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC;IACpC,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,2CAA2C;IAC3C,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/C,CAAC;IACD,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACvD,MAAM,gBAAgB,GAAG,MAAM,GAAG,GAAG,CAAC;IACtC,uBAAuB;IACvB,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,GAAG,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;IACzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzC,MAAM,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC,CAAC,CAAC;QAC9C,IAAI,GAAG,IAAI,gBAAgB,EAAE,CAAC;YAC5B,IAAI,GAAG,CAAC,IAAI,EAAE;gBAAE,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC;YACvC,GAAG,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;QACvB,CAAC;aAAM,CAAC;YACN;0BACc;YACd,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC1D,GAAG,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;YAC1C,CAAC;iBAAM,CAAC;gBACN,GAAG,IAAI,GAAG,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;YAC9B,CAAC;QACH,CAAC;IACH,CAAC;IACD,IAAI,GAAG,CAAC,IAAI,EAAE;QAAE,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC;IACvC,OAAO,KAAK,CAAC;AACf,CAAC;AAED,uDAAuD;AACvD,KAAK,UAAU,SAAS;IACtB,iEAAiE;IACjE,OAAO,MAAM,MAAM,CAAC,iCAAwC,CAAC,CAAC;AAChE,CAAC;AAED;;;mBAGmB;AACnB,KAAK,UAAU,gBAAgB;IAC7B;;;gEAG4D;IAC5D,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC,CAAC;IACxC,MAAM,aAAa,GAAI,GAAgE,CAAC,aAAa;WAC/F,GAA6E,CAAC,OAAO,CAAC,aAAa,CAAC;IAC1G,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC/C,OAAO,OAAO,CAAC,OAAO,CAAC,wCAAwC,CAAC,CAAC;AACnE,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,MAAc;IACvC,MAAM,KAAK,GAAG,MAAM,SAAS,EAAE,CAAC;IAChC,IAAI,KAAK,CAAC,mBAAmB,IAAI,CAAC,KAAK,CAAC,mBAAmB,CAAC,SAAS,EAAE,CAAC;QACtE,KAAK,CAAC,mBAAmB,CAAC,SAAS,GAAG,MAAM,gBAAgB,EAAE,CAAC;IACjE,CAAC;IACD,MAAM,WAAW,GAAG,KAAK,CAAC,WAAW,CAAC;QACpC,IAAI,EAAE,IAAI,UAAU,CAAC,MAAM,CAAC;QAC5B,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,IAAI;KACrB,CAAC,CAAC;IACH,OAAO,MAAM,WAAW,CAAC,OAAO,CAAC;AACnC,CAAC;AAED,mCAAmC;AAEnC,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,IAAqB;IAClD,IAAI,GAAW,CAAC;IAChB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACvC,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,MAAM,IAAI,KAAK,CAAC,0CAA0C,GAAG,GAAG,CAAC,CAAC;IACpE,CAAC;IAED,MAAM,QAAQ,GAAc,EAAE,CAAC;IAC/B,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,SAAS,MAAM,CAAC,MAAc;QAC5B,YAAY,IAAI,CAAC,CAAC;QAClB,OAAO,IAAI,GAAG,MAAM,GAAG,GAAG,GAAG,YAAY,CAAC;IAC5C,CAAC;IACD,IAAI,KAAK,GAAG,EAAE,CAAC;IAEf,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,GAAG,CAAC,QAAQ,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;QAC5D,IAAI,IAAa,CAAC;QAClB,IAAI,OAAuB,CAAC;QAC5B,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;YAClC,OAAO,GAAG,MAAM,IAAI,CAAC,cAAc,EAAE,CAAC;QACxC,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb;6DACiD;YACjD,QAAQ,CAAC,IAAI,CAAC;gBACZ,EAAE,EAAE,IAAI,GAAG,OAAO;gBAClB,OAAO,EAAE,SAAS,GAAG,OAAO;gBAC5B,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE;oBACN,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,SAAS,GAAG,OAAO,EAAE,EAAE,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE;oBACzE,EAAE,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,EAAE,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE;iBAC9E;aACF,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QACD,MAAM,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QACnD,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC;YAChD,uDAAuD;YACvD,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC;YACvD,IAAI,SAAS,IAAI,SAAS,CAAC,MAAM,GAAG,GAAG;gBAAE,KAAK,GAAG,SAAS,CAAC;QAC7D,CAAC;QACD,MAAM,OAAO,GAAG,SAAS,GAAG,OAAO,CAAC;QACpC,MAAM,MAAM,GAAY;YACtB,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE;SAC9D,CAAC;QACF,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;YACtB,MAAM,CAAC,IAAI,CAAC;gBACV,IAAI,EAAE,WAAW;gBACjB,KAAK,EAAE,CAAC;gBACR,IAAI,EAAE,CAAC;gBACP,EAAE,EAAE,MAAM,CAAC,GAAG,CAAC;aAChB,CAAC,CAAC;QACL,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC;YACZ,EAAE,EAAE,IAAI,GAAG,OAAO;YAClB,OAAO;YACP,KAAK,EAAE,CAAC;YACR,MAAM;SACP,CAAC,CAAC;IACL,CAAC;IAED,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,IAAI,KAAK,CAAC;SACxD,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,IAAI,KAAK,CAAC;IAEpC,MAAM,GAAG,GAAuB;QAC9B,EAAE,EAAE,KAAK;QACT,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,MAAM,EAAE,KAAK;QACb,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM;QACzB,KAAK,EAAE,KAAK,IAAI,CAAC,OAAO,GAAG,GAAG,CAAC,QAAQ,GAAG,SAAS,CAAC;QACpD,QAAQ;KACT,CAAC;IACF,IAAI,IAAI,CAAC,QAAQ,KAAK,SAAS;QAAE,GAAG,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;IAC9D,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/reader/parsers/rtf.d.ts

@@ -0,0 +1,37 @@
+/**
+ * RTF parser for the document reader (V1.20 of HITO 1).
+ *
+ * Tiny hand-rolled parser. No npm dep. RTF (Rich Text Format)
+ * is plain text with control words ({\rtf1 ... \par ... }),
+ * groups in {curly braces}, hex escapes (\'XX), and unicode
+ * escapes (\uNNNN). The parser walks the byte stream as a
+ * tiny state machine, dropping anything inside groups whose
+ * destination is metadata (fonttbl, colortbl, stylesheet,
+ * info, header, footer, pict, object, ...), and keeps the
+ * plain-text contents of the body.
+ *
+ * Paragraph breaks come from \par. Headings are detected
+ * heuristically (a line whose paragraph is fully bold and
+ * short, or a line tagged with \outlinelevelN, opens a new
+ * section). Plain paragraphs read as paragraph blocks.
+ *
+ * Bold/italic spans inside a paragraph are NOT preserved as
+ * structure -- they just contribute their text content. The
+ * reader spec doesn't need inline emphasis for voice readback;
+ * TTS prosody is handled elsewhere.
+ *
+ * Pathological inputs (mismatched braces, half-finished
+ * control words) degrade to "stop emitting and keep what we
+ * have so far" so the reader never crashes on a malformed
+ * file.
+ *
+ * ASCII-only.
+ */
+import type { NormalisedDocument } from '../types.js';
+export interface ParseRtfOptions {
+    buffer: Buffer;
+    filename: string;
+    language?: string;
+}
+export declare function parseRtf(opts: ParseRtfOptions): NormalisedDocument;
+//# sourceMappingURL=rtf.d.ts.map

package/dist/reader/parsers/rtf.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"rtf.d.ts","sourceRoot":"","sources":["../../../src/reader/parsers/rtf.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,OAAO,KAAK,EAAE,kBAAkB,EAAkB,MAAM,aAAa,CAAC;AAEtE,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAkSD,wBAAgB,QAAQ,CAAC,IAAI,EAAE,eAAe,GAAG,kBAAkB,CA4ElE"}

package/dist/reader/parsers/rtf.js

@@ -0,0 +1,347 @@
+/**
+ * RTF parser for the document reader (V1.20 of HITO 1).
+ *
+ * Tiny hand-rolled parser. No npm dep. RTF (Rich Text Format)
+ * is plain text with control words ({\rtf1 ... \par ... }),
+ * groups in {curly braces}, hex escapes (\'XX), and unicode
+ * escapes (\uNNNN). The parser walks the byte stream as a
+ * tiny state machine, dropping anything inside groups whose
+ * destination is metadata (fonttbl, colortbl, stylesheet,
+ * info, header, footer, pict, object, ...), and keeps the
+ * plain-text contents of the body.
+ *
+ * Paragraph breaks come from \par. Headings are detected
+ * heuristically (a line whose paragraph is fully bold and
+ * short, or a line tagged with \outlinelevelN, opens a new
+ * section). Plain paragraphs read as paragraph blocks.
+ *
+ * Bold/italic spans inside a paragraph are NOT preserved as
+ * structure -- they just contribute their text content. The
+ * reader spec doesn't need inline emphasis for voice readback;
+ * TTS prosody is handled elsewhere.
+ *
+ * Pathological inputs (mismatched braces, half-finished
+ * control words) degrade to "stop emitting and keep what we
+ * have so far" so the reader never crashes on a malformed
+ * file.
+ *
+ * ASCII-only.
+ */
+/** Destinations whose content is metadata and should NOT
+ *  appear in spoken text. When we enter a group whose first
+ *  control word is one of these, we drop everything until the
+ *  group closes. */
+const DROP_DESTINATIONS = new Set([
+    'fonttbl', 'colortbl', 'stylesheet', 'info', 'header',
+    'footer', 'pict', 'object', 'shp', 'shppict', 'nonshppict',
+    'datafield', 'panose', 'listtable', 'listoverridetable',
+    'rsidtbl', 'mmathPr', 'themedata', 'colorschememapping',
+    'latentstyles', 'datastore', 'wgrffmtfilter', 'xmlnstbl',
+    'fldinst',
+    /* Document protection / revision tracking -- not useful
+       for voice readback. */
+    'protect',
+]);
+function emptyPara() {
+    return { spans: [], outline: -1, hadPlain: false, hadBold: false };
+}
+function paraText(p) {
+    return p.spans.map((s) => s.text).join('').replace(/\s+/g, ' ').trim();
+}
+function emit(state) {
+    const text = paraText(state.para);
+    if (text) {
+        state.out.push({
+            text,
+            outline: state.para.outline,
+            bold: state.para.hadBold && !state.para.hadPlain,
+        });
+    }
+    state.para = emptyPara();
+}
+function appendText(state, text) {
+    if (state.dropDepth >= 0)
+        return;
+    if (!text)
+        return;
+    /* Decide whether this run is bold for the "fully bold para
+       = heading" heuristic. */
+    state.para.spans.push({ text, bold: state.bold, outline: state.outline });
+    if (state.bold)
+        state.para.hadBold = true;
+    else
+        state.para.hadPlain = true;
+}
+/** Handle one control word (and its optional numeric arg). */
+function handleControlWord(state, word, arg) {
+    /* Destination-introducing keywords. The RTF spec marks
+       destinations with a leading '*\\' but in practice we
+       can recognise them by name. */
+    if (DROP_DESTINATIONS.has(word)) {
+        state.dropDepth = state.depth;
+        return;
+    }
+    /* Skip the destination switch if we're already dropping. */
+    if (state.dropDepth >= 0)
+        return;
+    switch (word) {
+        case 'par':
+        case 'sect':
+        case 'page':
+        case 'line':
+            emit(state);
+            return;
+        case 'tab':
+            appendText(state, '\t');
+            return;
+        case 'b':
+            state.bold = arg === 0 ? false : true;
+            return;
+        case 'plain':
+            state.bold = false;
+            return;
+        case 'outlinelevel':
+            if (arg !== null && arg >= 0 && arg <= 5) {
+                state.outline = arg;
+                if (state.para.outline < 0 || arg < state.para.outline) {
+                    state.para.outline = arg;
+                }
+            }
+            return;
+        default:
+            /* Unknown control word -> ignore. */
+            return;
+    }
+}
+/** Handle \'XX hex escape. Interprets as a single byte in
+ *  CP1252 (the most common default RTF code page). For
+ *  bytes < 0x80, this matches ASCII. */
+function handleHexEscape(state, hex) {
+    const code = parseInt(hex, 16);
+    if (!isFinite(code))
+        return;
+    let ch;
+    if (code < 0x80) {
+        ch = String.fromCharCode(code);
+    }
+    else {
+        /* Map a few common CP1252 codepoints. Most RTF docs in
+           the wild use ASCII + utf-8 escapes; this fallback is
+           sufficient for voice readback. */
+        const CP1252 = {
+            0x91: "'", 0x92: "'", 0x93: '"', 0x94: '"',
+            0x95: '*', 0x96: '-', 0x97: '--', 0x85: '...',
+            0x99: '(tm)', 0xa9: '(c)', 0xae: '(R)',
+            0xa0: ' ',
+        };
+        ch = CP1252[code] || '?';
+    }
+    appendText(state, ch);
+}
+/** Handle \uNNNN unicode escape. RTF emits the code point as
+ *  a signed 16-bit integer, optionally followed by a fallback
+ *  ASCII character we must skip. */
+function handleUnicodeEscape(state, value) {
+    /* Cast signed 16-bit to actual code point. */
+    let code = value;
+    if (code < 0)
+        code = code + 0x10000;
+    if (code >= 0 && code <= 0x10ffff) {
+        try {
+            appendText(state, String.fromCodePoint(code));
+        }
+        catch {
+            appendText(state, '?');
+        }
+    }
+    /* The fallback char count is \ucN; we don't track that
+       dynamically here -- assume default uc=1. */
+    return { skipChars: 1 };
+}
+function parseRtfBody(src) {
+    const state = {
+        src,
+        i: 0,
+        depth: 0,
+        dropDepth: -1,
+        bold: false,
+        outline: -1,
+        para: emptyPara(),
+        out: [],
+    };
+    while (state.i < state.src.length) {
+        const ch = state.src.charAt(state.i);
+        if (ch === '{') {
+            state.depth += 1;
+            state.i += 1;
+            continue;
+        }
+        if (ch === '}') {
+            state.depth -= 1;
+            /* Closing the group that started a drop ends the drop. */
+            if (state.dropDepth >= 0 && state.depth < state.dropDepth) {
+                state.dropDepth = -1;
+            }
+            state.i += 1;
+            continue;
+        }
+        if (ch === '\\') {
+            const next = state.src.charAt(state.i + 1);
+            /* Special single-char escapes. */
+            if (next === '\\' || next === '{' || next === '}') {
+                appendText(state, next);
+                state.i += 2;
+                continue;
+            }
+            if (next === '\n' || next === '\r') {
+                /* Backslash followed by newline = literal newline
+                   paragraph break in some flavours. Treat as space. */
+                state.i += 2;
+                continue;
+            }
+            if (next === '*') {
+                /* Optional-destination prefix. The next control word
+                   (if a destination) MUST be dropped; we just skip the
+                   '*' and let the next token classify itself. */
+                state.i += 2;
+                continue;
+            }
+            if (next === "'") {
+                /* \'XX hex escape. */
+                const hex = state.src.substr(state.i + 2, 2);
+                handleHexEscape(state, hex);
+                state.i += 4;
+                continue;
+            }
+            if (next === 'u' && /[0-9-]/.test(state.src.charAt(state.i + 2) || '')) {
+                /* \uNNNN unicode escape. */
+                const m = state.src.slice(state.i + 1).match(/^u(-?\d+)/);
+                if (m) {
+                    const result = handleUnicodeEscape(state, parseInt(m[1], 10));
+                    state.i += 1 + m[0].length;
+                    /* Skip the configured number of fallback chars. */
+                    let skip = result.skipChars;
+                    while (skip > 0 && state.i < state.src.length) {
+                        const fc = state.src.charAt(state.i);
+                        if (fc === ' ') {
+                            state.i += 1;
+                            skip -= 1;
+                            continue;
+                        }
+                        if (fc === '\\') {
+                            /* A fallback that itself is an escape -- bail. */
+                            break;
+                        }
+                        state.i += 1;
+                        skip -= 1;
+                    }
+                    continue;
+                }
+            }
+            /* Generic control word: \wordN where N is optional
+               signed numeric argument, optionally terminated by a
+               space. */
+            const m = state.src.slice(state.i + 1).match(/^([a-zA-Z]+)(-?\d+)?/);
+            if (!m) {
+                /* Stray backslash -- skip. */
+                state.i += 1;
+                continue;
+            }
+            const word = m[1].toLowerCase();
+            const arg = m[2] !== undefined ? parseInt(m[2], 10) : null;
+            state.i += 1 + m[0].length;
+            /* The control word may be terminated by a single space
+               that does NOT contribute to output. */
+            if (state.src.charAt(state.i) === ' ')
+                state.i += 1;
+            handleControlWord(state, word, arg);
+            continue;
+        }
+        /* Plain character. Drop CR/LF (RTF uses them as readability
+           breaks, not content). */
+        if (ch === '\r' || ch === '\n') {
+            state.i += 1;
+            continue;
+        }
+        appendText(state, ch);
+        state.i += 1;
+    }
+    /* Flush the final paragraph. */
+    emit(state);
+    return state.out;
+}
+/* ----- document assembly ----- */
+export function parseRtf(opts) {
+    const text = opts.buffer.toString('utf-8');
+    const paras = parseRtfBody(text);
+    const sections = [];
+    let current = { id: 's-0', heading: '', level: 0, blocks: [] };
+    let sectionCounter = 0;
+    let blockCounter = 0;
+    let firstHeading = '';
+    function flush() {
+        if (current.blocks.length > 0 || current.heading)
+            sections.push(current);
+    }
+    function nextId(prefix) {
+        blockCounter += 1;
+        return 's-' + prefix + '-' + blockCounter;
+    }
+    function openSection(heading, level) {
+        flush();
+        sectionCounter += 1;
+        current = {
+            id: 's-' + sectionCounter,
+            heading,
+            level,
+            blocks: [{ kind: 'heading', level, text: heading, id: nextId('h') }],
+        };
+    }
+    for (const p of paras) {
+        /* Heading heuristics:
+           1. explicit \outlinelevelN -> level = N+1.
+           2. fully-bold short paragraph with no sentence punctuation
+              and <= 8 words -> heading at level 1. */
+        let level = 0;
+        if (p.outline >= 0) {
+            level = p.outline + 1;
+        }
+        else if (p.bold) {
+            const words = p.text.split(/\s+/).length;
+            if (words <= 8 && !/[.;:?!]$/.test(p.text)) {
+                level = 1;
+            }
+        }
+        const block = level > 0
+            ? { kind: 'heading', level, text: p.text, id: nextId('h') }
+            : { kind: 'paragraph', level: 0, text: p.text, id: nextId('p') };
+        if (level > 0) {
+            if (!firstHeading)
+                firstHeading = p.text;
+            openSection(p.text, level);
+            continue;
+        }
+        current.blocks.push(block);
+    }
+    flush();
+    if (sections.length === 0) {
+        sections.push({ id: 's-0', heading: '', level: 0, blocks: [] });
+    }
+    const docId = (opts.filename.split(/[\\/]/).pop() || 'doc')
+        .replace(/\.[^.]+$/, '')
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, '-')
+        .replace(/^-+|-+$/g, '') || 'doc';
+    const doc = {
+        id: docId,
+        filename: opts.filename,
+        format: 'rtf',
+        bytes: opts.buffer.length,
+        title: firstHeading,
+        sections,
+    };
+    if (opts.language !== undefined)
+        doc.language = opts.language;
+    return doc;
+}
+//# sourceMappingURL=rtf.js.map