@effortlessmotion/pdf-to-md 0.2.0 → 0.2.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (56) hide show
  1. package/dist/cjs/analyze.js.map +1 -1
  2. package/dist/cjs/audit.js.map +1 -1
  3. package/dist/cjs/cli/audit.js.map +1 -1
  4. package/dist/cjs/cli/extract.js.map +1 -1
  5. package/dist/cjs/cli/index.js.map +1 -1
  6. package/dist/cjs/cli/render.js.map +1 -1
  7. package/dist/cjs/cli/types.js.map +1 -1
  8. package/dist/cjs/cli.js.map +1 -1
  9. package/dist/cjs/collect.js.map +1 -1
  10. package/dist/cjs/dict.js.map +1 -1
  11. package/dist/cjs/emit.js.map +1 -1
  12. package/dist/cjs/engines/tesseract.js.map +1 -1
  13. package/dist/cjs/errors.js.map +1 -1
  14. package/dist/cjs/extract.js.map +1 -1
  15. package/dist/cjs/geometry.js.map +1 -1
  16. package/dist/cjs/index.js.map +1 -1
  17. package/dist/cjs/lines.js.map +1 -1
  18. package/dist/cjs/markdown.js.map +1 -1
  19. package/dist/cjs/page-marker.js.map +1 -1
  20. package/dist/cjs/pdf-open.js.map +1 -1
  21. package/dist/cjs/raster.js.map +1 -1
  22. package/dist/cjs/render-analysis.js.map +1 -1
  23. package/dist/cjs/report.js.map +1 -1
  24. package/dist/cjs/resolve.js.map +1 -1
  25. package/dist/cjs/text-analysis.js.map +1 -1
  26. package/dist/cjs/types.js.map +1 -1
  27. package/dist/cjs/validate.js.map +1 -1
  28. package/dist/esm/analyze.js.map +1 -1
  29. package/dist/esm/audit.js.map +1 -1
  30. package/dist/esm/cli/audit.js.map +1 -1
  31. package/dist/esm/cli/extract.js.map +1 -1
  32. package/dist/esm/cli/index.js.map +1 -1
  33. package/dist/esm/cli/render.js.map +1 -1
  34. package/dist/esm/cli/types.js.map +1 -1
  35. package/dist/esm/cli.js.map +1 -1
  36. package/dist/esm/collect.js.map +1 -1
  37. package/dist/esm/dict.js.map +1 -1
  38. package/dist/esm/emit.js.map +1 -1
  39. package/dist/esm/engines/tesseract.js.map +1 -1
  40. package/dist/esm/errors.js.map +1 -1
  41. package/dist/esm/extract.js.map +1 -1
  42. package/dist/esm/geometry.js.map +1 -1
  43. package/dist/esm/index.js.map +1 -1
  44. package/dist/esm/lines.js.map +1 -1
  45. package/dist/esm/markdown.js.map +1 -1
  46. package/dist/esm/page-marker.js.map +1 -1
  47. package/dist/esm/pdf-open.js.map +1 -1
  48. package/dist/esm/raster.js.map +1 -1
  49. package/dist/esm/render-analysis.js.map +1 -1
  50. package/dist/esm/report.js.map +1 -1
  51. package/dist/esm/resolve.js.map +1 -1
  52. package/dist/esm/text-analysis.js.map +1 -1
  53. package/dist/esm/types.js.map +1 -1
  54. package/dist/esm/validate.js.map +1 -1
  55. package/package.json +2 -3
  56. package/skills/pdf-to-md/SKILL.md +0 -75
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/cli/extract.ts"],"sourcesContent":["// extract command — parse this command's flags, call src/extract.ts, print\n// and write the result. No business logic here — the library returns data and throws; only the CLI prints and exits.\nimport { readFileSync, writeFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { parseArgs } from 'node:util';\nimport { analyze } from '../analyze.ts';\nimport type { OcrWordInput } from '../extract.ts';\nimport { analyzePdfAuto, extractOcr, pdfToMarkdown } from '../extract.ts';\nimport type { Tuning } from '../geometry.ts';\nimport { toText } from '../render-analysis.ts';\nimport type { Analysis, Report } from '../types.ts';\nimport type { Command } from './types.ts';\n\nconst USAGE =\n 'usage: pdf-to-md [extract] <file.pdf> [out.md] [--json] [--page-markers] [--ocr] [--no-ocr] ' +\n '[--pages N[-M]] [--stdout] [--words-json=FILE] [--debug-words=FILE] ' +\n '[--format md|txt|raw (default md; raw = Analysis JSON incl. the report; txt = reading-order plain text)] ' +\n '[--dpi=N (OCR render resolution, default 288; raise for tiny print or chart/label pages, ' +\n '>~400 measured worse on ordinary prose pages)] ' +\n '[--float-max-words=N] [--float-max-width=F] [--float-margin=F] [--para-gap=F] [--heading-scale=F] ' +\n '[--col-max-width=F] [--col-split=F] [--line-y-tol=F] [--line-height-ratio=F]';\n\nconst FORMATS = ['md', 'txt', 'raw'] as const;\ntype Format = (typeof FORMATS)[number];\n\n// A bad tuning-flag value (non-numeric, or <= 0 unless the flag allows zero) is a usage error, not a silent no-op.\nconst TUNING_FLAGS: {\n name: string;\n key: keyof Tuning;\n allowZero?: boolean;\n isInt?: boolean;\n}[] = [\n {\n name: 'float-max-words',\n key: 'floatMaxWords',\n allowZero: true,\n isInt: true,\n },\n { name: 'float-max-width', key: 'floatMaxWidth' },\n { name: 'float-margin', key: 'floatMargin' },\n { name: 'para-gap', key: 'paraGap' },\n { name: 'heading-scale', key: 'headingScale' },\n { name: 'col-max-width', key: 'colMaxWidth' },\n { name: 'col-split', key: 'colSplit' },\n { name: 'line-y-tol', key: 'lineYTol' },\n { name: 'line-height-ratio', key: 'lineHeightRatio' },\n];\n\nconst extract: Command = async (ctx) => {\n // --debug-words takes a required FILE value; a bare --debug-words is a\n // usage error, checked before parseArgs so it can't silently swallow the next token as its value.\n if (ctx.rest.includes('--debug-words')) {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n\n const parsed = (() => {\n try {\n return parseArgs({\n args: ctx.rest,\n options: {\n json: { type: 'boolean', default: false },\n 'page-markers': { type: 'boolean', default: false },\n ocr: { type: 'boolean', default: false },\n 'no-ocr': { type: 'boolean', default: false },\n stdout: { type: 'boolean', default: false },\n 'debug-words': { type: 'string' },\n help: { type: 'boolean', default: false, short: 'h' },\n pages: { type: 'string' },\n dpi: { type: 'string' },\n format: { type: 'string', default: 'md' },\n 'words-json': { type: 'string' },\n ...Object.fromEntries(TUNING_FLAGS.map((f) => [f.name, { type: 'string' as const }])),\n },\n allowPositionals: true,\n });\n } catch (err) {\n // usageError exits; the throw is dead code that tells the compiler so.\n throw ctx.usageError(`${(err as Error).message}\\n${USAGE}`);\n }\n })();\n const values = parsed.values as Record<string, string | boolean | undefined>;\n const positionals = parsed.positionals;\n\n if (values.help) {\n console.log(USAGE);\n return;\n }\n\n const [src, outArg] = positionals;\n const wordsPath = values['words-json'] as string | undefined;\n if (!src && !wordsPath) ctx.usageError(USAGE);\n\n const formatRaw = values.format as string;\n if (!(FORMATS as readonly string[]).includes(formatRaw)) {\n ctx.usageError(`${USAGE}\\n--format must be one of ${FORMATS.join('|')} (got ${JSON.stringify(formatRaw)})`);\n }\n const format = formatRaw as Format;\n\n // --pages: a malformed spec is a usage error; out-of-range page NUMBERS\n // clamp instead (collect.ts / the engine apply the same clamping policy).\n let pages: { first: number; last: number } | undefined;\n if (values.pages !== undefined) {\n const m = /^(\\d+)(?:-(\\d+))?$/.exec(values.pages as string);\n const first = m ? Number(m[1]) : Number.NaN;\n const last = m ? (m[2] !== undefined ? Number(m[2]) : first) : Number.NaN;\n if (!m || first < 1 || last < first) {\n ctx.usageError(`${USAGE}\\n--pages requires N or N-M, N >= 1, M >= N (got ${JSON.stringify(values.pages)})`);\n }\n pages = { first, last };\n }\n\n const tuning: Partial<Tuning> = {};\n for (const spec of TUNING_FLAGS) {\n const raw = values[spec.name] as string | undefined;\n if (raw === undefined) continue;\n const value = Number(raw);\n const ok = raw !== '' && Number.isFinite(value) && (spec.allowZero ? value >= 0 : value > 0);\n if (!ok) {\n ctx.usageError(`${USAGE}\\n--${spec.name} requires a ${spec.allowZero ? 'non-negative' : 'positive'} number (got ${JSON.stringify(raw)})`);\n }\n (tuning[spec.key] as number) = spec.isInt ? Math.trunc(value) : value;\n }\n\n // --dpi: page-render resolution, validated like the tuning flags. Default\n // 288 is the measured operating point; >~400 measured WORSE on ordinary pages — not a \"bigger is safer\" knob.\n let dpi: number | undefined;\n if (values.dpi !== undefined) {\n const raw = values.dpi as string;\n const value = Number(raw);\n if (raw === '' || !Number.isFinite(value) || value <= 0) {\n ctx.usageError(`${USAGE}\\n--dpi requires a positive number (got ${JSON.stringify(raw)})`);\n }\n dpi = value;\n }\n\n // In --words-json mode there may be no PDF positional; the words file\n // stands in for naming, extension stripped so the default dest can never overwrite the input. Default extension follows --format.\n const source = src ?? (wordsPath as string);\n const defaultExt = format === 'raw' ? 'json' : format === 'txt' ? 'txt' : 'md';\n const dest = outArg || `${source.replace(/\\.(pdf|jsonl)$/i, '')}.${defaultExt}`;\n\n const onWarning = (m: string) => console.error(m);\n const showProgress = process.stderr.isTTY;\n // onProgress carries per-page timing/count only; human-readable\n // diagnostics (e.g. the sparse-retry notice) arrive via onWarning instead, so they print regardless of TTY-ness.\n const onProgress = (e: { type: 'page'; page: number; words: number; ms: number }) => {\n if (showProgress) console.error(`p${e.page}: ${e.words} words, ${e.ms}ms`);\n };\n const debugWordsPath = values['debug-words'] as string | undefined;\n if (debugWordsPath !== undefined && debugWordsPath.trim() === '') {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n // The severity-1 audit instrument: every recognized word as clean JSON\n // lines to its own file. stderr stays a human channel (mixed prose), never a dump target — this file is guaranteed byte-for-byte parseable.\n const onWords = debugWordsPath\n ? (words: OcrWordInput[]) => {\n const lines = words.map((w) => {\n const entry: OcrWordInput = { page: w.page, text: w.text, x: w.x, y: w.y, w: w.w, h: w.h };\n if (w.confidence !== undefined) entry.confidence = w.confidence;\n return JSON.stringify(entry);\n });\n writeFileSync(debugWordsPath, lines.length ? `${lines.join('\\n')}\\n` : '');\n }\n : undefined;\n\n const common = { pages, onWarning, onWords };\n const ocrCommon = {\n ...common,\n tuning,\n dpi,\n onProgress,\n onWords,\n };\n\n let words: OcrWordInput[] | undefined;\n let skipped = 0;\n if (wordsPath !== undefined) {\n ({ words, skipped } = parseWordLines(readFileSync(wordsPath, 'utf8')));\n if (skipped > 0) console.error(`--words-json: skipped ${skipped} non-JSON line(s)`);\n }\n\n let content: string;\n let report: Report;\n if (format === 'md') {\n let markdown: string;\n if (words !== undefined) {\n ({ markdown, report } = await extractOcr({ words }, ocrCommon));\n } else {\n ({ markdown, report } = await pdfToMarkdown(src, {\n ...ocrCommon,\n ocr: values.ocr as boolean,\n noOcr: values['no-ocr'] as boolean,\n pageMarkers: values['page-markers'] as boolean,\n }));\n }\n content = markdown;\n } else {\n // raw/txt render the same routed Analysis as the Markdown API, so a\n // format switch cannot discard a sparse image page or change audit input.\n let analysis: Analysis;\n if (words !== undefined) {\n analysis = await analyze({ words }, { ...ocrCommon, path: 'ocr' });\n } else {\n analysis = (\n await analyzePdfAuto(src, {\n ...ocrCommon,\n ocr: values.ocr as boolean,\n noOcr: values['no-ocr'] as boolean,\n pageMarkers: values['page-markers'] as boolean,\n })\n ).analysis;\n }\n report = analysis.report;\n content = format === 'raw' ? JSON.stringify(analysis, null, 1) : toText(analysis);\n }\n\n if (values.stdout) process.stdout.write(content);\n else writeFileSync(dest, content);\n\n // The library's Report stays clean of source/out; --json consumers keep\n // report.json files around and need to know which input produced each one, so the CLI (only) decorates the printed JSON with source/out here.\n if (values.json) console.log(JSON.stringify({ source, out: dest, ...report }, null, 1));\n else if (report.path === 'ocr') {\n console.error(`${path.basename(dest).padEnd(32)} OCR path heads=${report.headings}(junk ${report.junkHeadings}) ` + `paras=${report.paragraphs} callouts=${report.callouts} floats=${report.floats ?? 0} dangling=${report.danglingLong} (long) ` + `~${report.charsPerPage} chars/page`);\n } else {\n console.error(\n `${path.basename(dest).padEnd(32)} body=${report.bodyFontHeight}pt heads=${report.headings}(junk ${report.junkHeadings}) ` +\n `lists=${report.listItems} paras=${report.paragraphs} callouts=${report.callouts} pairs=${report.joinedPairs} ` +\n `~${report.charsPerPage} chars/page` +\n (report.sparse ? ' ⚠️ implausibly sparse — pages are likely images; re-run with --ocr' : '')\n );\n }\n};\n\n// Parse one word-dump JSON line per word — key order is not guaranteed, parse\n// never grep. Non-JSON lines are skipped with a count, not fatal (the auditor applies the same policy).\nfunction parseWordLines(raw: string): { words: OcrWordInput[]; skipped: number } {\n const words: OcrWordInput[] = [];\n let skipped = 0;\n for (const line of raw.split('\\n')) {\n if (!line.trim()) continue;\n try {\n words.push(JSON.parse(line));\n } catch {\n skipped++;\n }\n }\n return { words, skipped };\n}\n\nexport default extract;\n"],"names":["readFileSync","writeFileSync","path","parseArgs","analyze","analyzePdfAuto","extractOcr","pdfToMarkdown","toText","USAGE","FORMATS","TUNING_FLAGS","name","key","allowZero","isInt","extract","ctx","rest","includes","usageError","parsed","args","options","json","type","default","ocr","stdout","help","short","pages","dpi","format","Object","fromEntries","map","f","allowPositionals","err","message","values","positionals","console","log","src","outArg","wordsPath","formatRaw","join","JSON","stringify","undefined","m","exec","first","Number","NaN","last","tuning","spec","raw","value","ok","isFinite","Math","trunc","source","defaultExt","dest","replace","onWarning","error","showProgress","process","stderr","isTTY","onProgress","e","page","words","ms","debugWordsPath","trim","onWords","lines","w","entry","text","x","y","h","confidence","length","common","ocrCommon","skipped","parseWordLines","content","report","markdown","noOcr","pageMarkers","analysis","write","out","basename","padEnd","headings","junkHeadings","paragraphs","callouts","floats","danglingLong","charsPerPage","bodyFontHeight","listItems","joinedPairs","sparse","line","split","push","parse"],"mappings":"AAAA,2EAA2E;AAC3E,qHAAqH;AACrH,SAASA,YAAY,EAAEC,aAAa,QAAQ,UAAU;AACtD,OAAOC,UAAU,YAAY;AAC7B,SAASC,SAAS,QAAQ,YAAY;AACtC,SAASC,OAAO,QAAQ,gBAAgB;AAExC,SAASC,cAAc,EAAEC,UAAU,EAAEC,aAAa,QAAQ,gBAAgB;AAE1E,SAASC,MAAM,QAAQ,wBAAwB;AAI/C,MAAMC,QACJ,iGACA,yEACA,8GACA,8FACA,oDACA,uGACA;AAEF,MAAMC,UAAU;IAAC;IAAM;IAAO;CAAM;AAGpC,mHAAmH;AACnH,MAAMC,eAKA;IACJ;QACEC,MAAM;QACNC,KAAK;QACLC,WAAW;QACXC,OAAO;IACT;IACA;QAAEH,MAAM;QAAmBC,KAAK;IAAgB;IAChD;QAAED,MAAM;QAAgBC,KAAK;IAAc;IAC3C;QAAED,MAAM;QAAYC,KAAK;IAAU;IACnC;QAAED,MAAM;QAAiBC,KAAK;IAAe;IAC7C;QAAED,MAAM;QAAiBC,KAAK;IAAc;IAC5C;QAAED,MAAM;QAAaC,KAAK;IAAW;IACrC;QAAED,MAAM;QAAcC,KAAK;IAAW;IACtC;QAAED,MAAM;QAAqBC,KAAK;IAAkB;CACrD;AAED,MAAMG,UAAmB,OAAOC;IAC9B,uEAAuE;IACvE,kGAAkG;IAClG,IAAIA,IAAIC,IAAI,CAACC,QAAQ,CAAC,kBAAkB;QACtCF,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IAEA,MAAMY,SAAS,AAAC,CAAA;QACd,IAAI;YACF,OAAOlB,UAAU;gBACfmB,MAAML,IAAIC,IAAI;gBACdK,SAAS;oBACPC,MAAM;wBAAEC,MAAM;wBAAWC,SAAS;oBAAM;oBACxC,gBAAgB;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAClDC,KAAK;wBAAEF,MAAM;wBAAWC,SAAS;oBAAM;oBACvC,UAAU;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAC5CE,QAAQ;wBAAEH,MAAM;wBAAWC,SAAS;oBAAM;oBAC1C,eAAe;wBAAED,MAAM;oBAAS;oBAChCI,MAAM;wBAAEJ,MAAM;wBAAWC,SAAS;wBAAOI,OAAO;oBAAI;oBACpDC,OAAO;wBAAEN,MAAM;oBAAS;oBACxBO,KAAK;wBAAEP,MAAM;oBAAS;oBACtBQ,QAAQ;wBAAER,MAAM;wBAAUC,SAAS;oBAAK;oBACxC,cAAc;wBAAED,MAAM;oBAAS;oBAC/B,GAAGS,OAAOC,WAAW,CAACxB,aAAayB,GAAG,CAAC,CAACC,IAAM;4BAACA,EAAEzB,IAAI;4BAAE;gCAAEa,MAAM;4BAAkB;yBAAE,EAAE;gBACvF;gBACAa,kBAAkB;YACpB;QACF,EAAE,OAAOC,KAAK;YACZ,uEAAuE;YACvE,MAAMtB,IAAIG,UAAU,CAAC,GAAG,AAACmB,IAAcC,OAAO,CAAC,EAAE,EAAE/B,OAAO;QAC5D;IACF,CAAA;IACA,MAAMgC,SAASpB,OAAOoB,MAAM;IAC5B,MAAMC,cAAcrB,OAAOqB,WAAW;IAEtC,IAAID,OAAOZ,IAAI,EAAE;QACfc,QAAQC,GAAG,CAACnC;QACZ;IACF;IAEA,MAAM,CAACoC,KAAKC,OAAO,GAAGJ;IACtB,MAAMK,YAAYN,MAAM,CAAC,aAAa;IACtC,IAAI,CAACI,OAAO,CAACE,WAAW9B,IAAIG,UAAU,CAACX;IAEvC,MAAMuC,YAAYP,OAAOR,MAAM;IAC/B,IAAI,CAAC,AAACvB,QAA8BS,QAAQ,CAAC6B,YAAY;QACvD/B,IAAIG,UAAU,CAAC,GAAGX,MAAM,0BAA0B,EAAEC,QAAQuC,IAAI,CAAC,KAAK,MAAM,EAAEC,KAAKC,SAAS,CAACH,WAAW,CAAC,CAAC;IAC5G;IACA,MAAMf,SAASe;IAEf,wEAAwE;IACxE,0EAA0E;IAC1E,IAAIjB;IACJ,IAAIU,OAAOV,KAAK,KAAKqB,WAAW;QAC9B,MAAMC,IAAI,qBAAqBC,IAAI,CAACb,OAAOV,KAAK;QAChD,MAAMwB,QAAQF,IAAIG,OAAOH,CAAC,CAAC,EAAE,IAAIG,OAAOC,GAAG;QAC3C,MAAMC,OAAOL,IAAKA,CAAC,CAAC,EAAE,KAAKD,YAAYI,OAAOH,CAAC,CAAC,EAAE,IAAIE,QAASC,OAAOC,GAAG;QACzE,IAAI,CAACJ,KAAKE,QAAQ,KAAKG,OAAOH,OAAO;YACnCtC,IAAIG,UAAU,CAAC,GAAGX,MAAM,iDAAiD,EAAEyC,KAAKC,SAAS,CAACV,OAAOV,KAAK,EAAE,CAAC,CAAC;QAC5G;QACAA,QAAQ;YAAEwB;YAAOG;QAAK;IACxB;IAEA,MAAMC,SAA0B,CAAC;IACjC,KAAK,MAAMC,QAAQjD,aAAc;QAC/B,MAAMkD,MAAMpB,MAAM,CAACmB,KAAKhD,IAAI,CAAC;QAC7B,IAAIiD,QAAQT,WAAW;QACvB,MAAMU,QAAQN,OAAOK;QACrB,MAAME,KAAKF,QAAQ,MAAML,OAAOQ,QAAQ,CAACF,UAAWF,CAAAA,KAAK9C,SAAS,GAAGgD,SAAS,IAAIA,QAAQ,CAAA;QAC1F,IAAI,CAACC,IAAI;YACP9C,IAAIG,UAAU,CAAC,GAAGX,MAAM,IAAI,EAAEmD,KAAKhD,IAAI,CAAC,YAAY,EAAEgD,KAAK9C,SAAS,GAAG,iBAAiB,WAAW,aAAa,EAAEoC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1I;QACCF,MAAM,CAACC,KAAK/C,GAAG,CAAC,GAAc+C,KAAK7C,KAAK,GAAGkD,KAAKC,KAAK,CAACJ,SAASA;IAClE;IAEA,0EAA0E;IAC1E,8GAA8G;IAC9G,IAAI9B;IACJ,IAAIS,OAAOT,GAAG,KAAKoB,WAAW;QAC5B,MAAMS,MAAMpB,OAAOT,GAAG;QACtB,MAAM8B,QAAQN,OAAOK;QACrB,IAAIA,QAAQ,MAAM,CAACL,OAAOQ,QAAQ,CAACF,UAAUA,SAAS,GAAG;YACvD7C,IAAIG,UAAU,CAAC,GAAGX,MAAM,wCAAwC,EAAEyC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1F;QACA7B,MAAM8B;IACR;IAEA,sEAAsE;IACtE,kIAAkI;IAClI,MAAMK,SAAStB,gBAAAA,iBAAAA,MAAQE;IACvB,MAAMqB,aAAanC,WAAW,QAAQ,SAASA,WAAW,QAAQ,QAAQ;IAC1E,MAAMoC,OAAOvB,UAAU,GAAGqB,OAAOG,OAAO,CAAC,mBAAmB,IAAI,CAAC,EAAEF,YAAY;IAE/E,MAAMG,YAAY,CAAClB,IAAcV,QAAQ6B,KAAK,CAACnB;IAC/C,MAAMoB,eAAeC,QAAQC,MAAM,CAACC,KAAK;IACzC,gEAAgE;IAChE,iHAAiH;IACjH,MAAMC,aAAa,CAACC;QAClB,IAAIL,cAAc9B,QAAQ6B,KAAK,CAAC,CAAC,CAAC,EAAEM,EAAEC,IAAI,CAAC,EAAE,EAAED,EAAEE,KAAK,CAAC,QAAQ,EAAEF,EAAEG,EAAE,CAAC,EAAE,CAAC;IAC3E;IACA,MAAMC,iBAAiBzC,MAAM,CAAC,cAAc;IAC5C,IAAIyC,mBAAmB9B,aAAa8B,eAAeC,IAAI,OAAO,IAAI;QAChElE,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IACA,uEAAuE;IACvE,4IAA4I;IAC5I,MAAM2E,UAAUF,iBACZ,CAACF;QACC,MAAMK,QAAQL,MAAM5C,GAAG,CAAC,CAACkD;YACvB,MAAMC,QAAsB;gBAAER,MAAMO,EAAEP,IAAI;gBAAES,MAAMF,EAAEE,IAAI;gBAAEC,GAAGH,EAAEG,CAAC;gBAAEC,GAAGJ,EAAEI,CAAC;gBAAEJ,GAAGA,EAAEA,CAAC;gBAAEK,GAAGL,EAAEK,CAAC;YAAC;YACzF,IAAIL,EAAEM,UAAU,KAAKxC,WAAWmC,MAAMK,UAAU,GAAGN,EAAEM,UAAU;YAC/D,OAAO1C,KAAKC,SAAS,CAACoC;QACxB;QACAtF,cAAciF,gBAAgBG,MAAMQ,MAAM,GAAG,GAAGR,MAAMpC,IAAI,CAAC,MAAM,EAAE,CAAC,GAAG;IACzE,IACAG;IAEJ,MAAM0C,SAAS;QAAE/D;QAAOwC;QAAWa;IAAQ;IAC3C,MAAMW,YAAY;QAChB,GAAGD,MAAM;QACTnC;QACA3B;QACA6C;QACAO;IACF;IAEA,IAAIJ;IACJ,IAAIgB,UAAU;IACd,IAAIjD,cAAcK,WAAW;QAC1B,CAAA,EAAE4B,KAAK,EAAEgB,OAAO,EAAE,GAAGC,eAAejG,aAAa+C,WAAW,QAAO;QACpE,IAAIiD,UAAU,GAAGrD,QAAQ6B,KAAK,CAAC,CAAC,sBAAsB,EAAEwB,QAAQ,iBAAiB,CAAC;IACpF;IAEA,IAAIE;IACJ,IAAIC;IACJ,IAAIlE,WAAW,MAAM;QACnB,IAAImE;QACJ,IAAIpB,UAAU5B,WAAW;YACtB,CAAA,EAAEgD,QAAQ,EAAED,MAAM,EAAE,GAAG,MAAM7F,WAAW;gBAAE0E;YAAM,GAAGe,UAAS;QAC/D,OAAO;YACJ,CAAA,EAAEK,QAAQ,EAAED,MAAM,EAAE,GAAG,MAAM5F,cAAcsC,KAAK;gBAC/C,GAAGkD,SAAS;gBACZpE,KAAKc,OAAOd,GAAG;gBACf0E,OAAO5D,MAAM,CAAC,SAAS;gBACvB6D,aAAa7D,MAAM,CAAC,eAAe;YACrC,EAAC;QACH;QACAyD,UAAUE;IACZ,OAAO;QACL,oEAAoE;QACpE,0EAA0E;QAC1E,IAAIG;QACJ,IAAIvB,UAAU5B,WAAW;YACvBmD,WAAW,MAAMnG,QAAQ;gBAAE4E;YAAM,GAAG;gBAAE,GAAGe,SAAS;gBAAE7F,MAAM;YAAM;QAClE,OAAO;YACLqG,WAAW,AACT,CAAA,MAAMlG,eAAewC,KAAK;gBACxB,GAAGkD,SAAS;gBACZpE,KAAKc,OAAOd,GAAG;gBACf0E,OAAO5D,MAAM,CAAC,SAAS;gBACvB6D,aAAa7D,MAAM,CAAC,eAAe;YACrC,EAAC,EACD8D,QAAQ;QACZ;QACAJ,SAASI,SAASJ,MAAM;QACxBD,UAAUjE,WAAW,QAAQiB,KAAKC,SAAS,CAACoD,UAAU,MAAM,KAAK/F,OAAO+F;IAC1E;IAEA,IAAI9D,OAAOb,MAAM,EAAE8C,QAAQ9C,MAAM,CAAC4E,KAAK,CAACN;SACnCjG,cAAcoE,MAAM6B;IAEzB,wEAAwE;IACxE,8IAA8I;IAC9I,IAAIzD,OAAOjB,IAAI,EAAEmB,QAAQC,GAAG,CAACM,KAAKC,SAAS,CAAC;QAAEgB;QAAQsC,KAAKpC;QAAM,GAAG8B,MAAM;IAAC,GAAG,MAAM;SAC/E,IAAIA,OAAOjG,IAAI,KAAK,OAAO;YACuJiG;QAArLxD,QAAQ6B,KAAK,CAAC,GAAGtE,KAAKwG,QAAQ,CAACrC,MAAMsC,MAAM,CAAC,IAAI,gBAAgB,EAAER,OAAOS,QAAQ,CAAC,MAAM,EAAET,OAAOU,YAAY,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,EAAEV,OAAOW,UAAU,CAAC,UAAU,EAAEX,OAAOY,QAAQ,CAAC,QAAQ,GAAEZ,iBAAAA,OAAOa,MAAM,cAAbb,4BAAAA,iBAAiB,EAAE,UAAU,EAAEA,OAAOc,YAAY,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAEd,OAAOe,YAAY,CAAC,WAAW,CAAC;IAC1R,OAAO;QACLvE,QAAQ6B,KAAK,CACX,GAAGtE,KAAKwG,QAAQ,CAACrC,MAAMsC,MAAM,CAAC,IAAI,MAAM,EAAER,OAAOgB,cAAc,CAAC,SAAS,EAAEhB,OAAOS,QAAQ,CAAC,MAAM,EAAET,OAAOU,YAAY,CAAC,EAAE,CAAC,GACxH,CAAC,MAAM,EAAEV,OAAOiB,SAAS,CAAC,OAAO,EAAEjB,OAAOW,UAAU,CAAC,UAAU,EAAEX,OAAOY,QAAQ,CAAC,OAAO,EAAEZ,OAAOkB,WAAW,CAAC,CAAC,CAAC,GAC/G,CAAC,CAAC,EAAElB,OAAOe,YAAY,CAAC,WAAW,CAAC,GACnCf,CAAAA,OAAOmB,MAAM,GAAG,0EAA0E,EAAC;IAElG;AACF;AAEA,8EAA8E;AAC9E,wGAAwG;AACxG,SAASrB,eAAepC,GAAW;IACjC,MAAMmB,QAAwB,EAAE;IAChC,IAAIgB,UAAU;IACd,KAAK,MAAMuB,QAAQ1D,IAAI2D,KAAK,CAAC,MAAO;QAClC,IAAI,CAACD,KAAKpC,IAAI,IAAI;QAClB,IAAI;YACFH,MAAMyC,IAAI,CAACvE,KAAKwE,KAAK,CAACH;QACxB,EAAE,OAAM;YACNvB;QACF;IACF;IACA,OAAO;QAAEhB;QAAOgB;IAAQ;AAC1B;AAEA,eAAehF,QAAQ"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/cli/extract.ts"],"sourcesContent":["// extract command — parse this command's flags, call src/extract.ts, print\n// and write the result. No business logic here — the library returns data and throws; only the CLI prints and exits.\nimport { readFileSync, writeFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { parseArgs } from 'node:util';\nimport { analyze } from '../analyze.ts';\nimport type { OcrWordInput } from '../extract.ts';\nimport { analyzePdfAuto, extractOcr, pdfToMarkdown } from '../extract.ts';\nimport type { Tuning } from '../geometry.ts';\nimport { toText } from '../render-analysis.ts';\nimport type { Analysis, Report } from '../types.ts';\nimport type { Command } from './types.ts';\n\nconst USAGE =\n 'usage: pdf-to-md [extract] <file.pdf> [out.md] [--json] [--page-markers] [--ocr] [--no-ocr] ' +\n '[--pages N[-M]] [--stdout] [--words-json=FILE] [--debug-words=FILE] ' +\n '[--format md|txt|raw (default md; raw = Analysis JSON incl. the report; txt = reading-order plain text)] ' +\n '[--dpi=N (OCR render resolution, default 288; raise for tiny print or chart/label pages, ' +\n '>~400 measured worse on ordinary prose pages)] ' +\n '[--float-max-words=N] [--float-max-width=F] [--float-margin=F] [--para-gap=F] [--heading-scale=F] ' +\n '[--col-max-width=F] [--col-split=F] [--line-y-tol=F] [--line-height-ratio=F]';\n\nconst FORMATS = ['md', 'txt', 'raw'] as const;\ntype Format = (typeof FORMATS)[number];\n\n// A bad tuning-flag value (non-numeric, or <= 0 unless the flag allows zero) is a usage error, not a silent no-op.\nconst TUNING_FLAGS: {\n name: string;\n key: keyof Tuning;\n allowZero?: boolean;\n isInt?: boolean;\n}[] = [\n {\n name: 'float-max-words',\n key: 'floatMaxWords',\n allowZero: true,\n isInt: true,\n },\n { name: 'float-max-width', key: 'floatMaxWidth' },\n { name: 'float-margin', key: 'floatMargin' },\n { name: 'para-gap', key: 'paraGap' },\n { name: 'heading-scale', key: 'headingScale' },\n { name: 'col-max-width', key: 'colMaxWidth' },\n { name: 'col-split', key: 'colSplit' },\n { name: 'line-y-tol', key: 'lineYTol' },\n { name: 'line-height-ratio', key: 'lineHeightRatio' },\n];\n\nconst extract: Command = async (ctx) => {\n // --debug-words takes a required FILE value; a bare --debug-words is a\n // usage error, checked before parseArgs so it can't silently swallow the next token as its value.\n if (ctx.rest.includes('--debug-words')) {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n\n const parsed = (() => {\n try {\n return parseArgs({\n args: ctx.rest,\n options: {\n json: { type: 'boolean', default: false },\n 'page-markers': { type: 'boolean', default: false },\n ocr: { type: 'boolean', default: false },\n 'no-ocr': { type: 'boolean', default: false },\n stdout: { type: 'boolean', default: false },\n 'debug-words': { type: 'string' },\n help: { type: 'boolean', default: false, short: 'h' },\n pages: { type: 'string' },\n dpi: { type: 'string' },\n format: { type: 'string', default: 'md' },\n 'words-json': { type: 'string' },\n ...Object.fromEntries(TUNING_FLAGS.map((f) => [f.name, { type: 'string' as const }])),\n },\n allowPositionals: true,\n });\n } catch (err) {\n // usageError exits; the throw is dead code that tells the compiler so.\n throw ctx.usageError(`${(err as Error).message}\\n${USAGE}`);\n }\n })();\n const values = parsed.values as Record<string, string | boolean | undefined>;\n const positionals = parsed.positionals;\n\n if (values.help) {\n console.log(USAGE);\n return;\n }\n\n const [src, outArg] = positionals;\n const wordsPath = values['words-json'] as string | undefined;\n if (!src && !wordsPath) ctx.usageError(USAGE);\n\n const formatRaw = values.format as string;\n if (!(FORMATS as readonly string[]).includes(formatRaw)) {\n ctx.usageError(`${USAGE}\\n--format must be one of ${FORMATS.join('|')} (got ${JSON.stringify(formatRaw)})`);\n }\n const format = formatRaw as Format;\n\n // --pages: a malformed spec is a usage error; out-of-range page NUMBERS\n // clamp instead (collect.ts / the engine apply the same clamping policy).\n let pages: { first: number; last: number } | undefined;\n if (values.pages !== undefined) {\n const m = /^(\\d+)(?:-(\\d+))?$/.exec(values.pages as string);\n const first = m ? Number(m[1]) : Number.NaN;\n const last = m ? (m[2] !== undefined ? Number(m[2]) : first) : Number.NaN;\n if (!m || first < 1 || last < first) {\n ctx.usageError(`${USAGE}\\n--pages requires N or N-M, N >= 1, M >= N (got ${JSON.stringify(values.pages)})`);\n }\n pages = { first, last };\n }\n\n const tuning: Partial<Tuning> = {};\n for (const spec of TUNING_FLAGS) {\n const raw = values[spec.name] as string | undefined;\n if (raw === undefined) continue;\n const value = Number(raw);\n const ok = raw !== '' && Number.isFinite(value) && (spec.allowZero ? value >= 0 : value > 0);\n if (!ok) {\n ctx.usageError(`${USAGE}\\n--${spec.name} requires a ${spec.allowZero ? 'non-negative' : 'positive'} number (got ${JSON.stringify(raw)})`);\n }\n (tuning[spec.key] as number) = spec.isInt ? Math.trunc(value) : value;\n }\n\n // --dpi: page-render resolution, validated like the tuning flags. Default\n // 288 is the measured operating point; >~400 measured WORSE on ordinary pages — not a \"bigger is safer\" knob.\n let dpi: number | undefined;\n if (values.dpi !== undefined) {\n const raw = values.dpi as string;\n const value = Number(raw);\n if (raw === '' || !Number.isFinite(value) || value <= 0) {\n ctx.usageError(`${USAGE}\\n--dpi requires a positive number (got ${JSON.stringify(raw)})`);\n }\n dpi = value;\n }\n\n // In --words-json mode there may be no PDF positional; the words file\n // stands in for naming, extension stripped so the default dest can never overwrite the input. Default extension follows --format.\n const source = src ?? (wordsPath as string);\n const defaultExt = format === 'raw' ? 'json' : format === 'txt' ? 'txt' : 'md';\n const dest = outArg || `${source.replace(/\\.(pdf|jsonl)$/i, '')}.${defaultExt}`;\n\n const onWarning = (m: string) => console.error(m);\n const showProgress = process.stderr.isTTY;\n // onProgress carries per-page timing/count only; human-readable\n // diagnostics (e.g. the sparse-retry notice) arrive via onWarning instead, so they print regardless of TTY-ness.\n const onProgress = (e: { type: 'page'; page: number; words: number; ms: number }) => {\n if (showProgress) console.error(`p${e.page}: ${e.words} words, ${e.ms}ms`);\n };\n const debugWordsPath = values['debug-words'] as string | undefined;\n if (debugWordsPath !== undefined && debugWordsPath.trim() === '') {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n // The severity-1 audit instrument: every recognized word as clean JSON\n // lines to its own file. stderr stays a human channel (mixed prose), never a dump target — this file is guaranteed byte-for-byte parseable.\n const onWords = debugWordsPath\n ? (words: OcrWordInput[]) => {\n const lines = words.map((w) => {\n const entry: OcrWordInput = { page: w.page, text: w.text, x: w.x, y: w.y, w: w.w, h: w.h };\n if (w.confidence !== undefined) entry.confidence = w.confidence;\n return JSON.stringify(entry);\n });\n writeFileSync(debugWordsPath, lines.length ? `${lines.join('\\n')}\\n` : '');\n }\n : undefined;\n\n const common = { pages, onWarning, onWords };\n const ocrCommon = {\n ...common,\n tuning,\n dpi,\n onProgress,\n onWords,\n };\n\n let words: OcrWordInput[] | undefined;\n let skipped = 0;\n if (wordsPath !== undefined) {\n ({ words, skipped } = parseWordLines(readFileSync(wordsPath, 'utf8')));\n if (skipped > 0) console.error(`--words-json: skipped ${skipped} non-JSON line(s)`);\n }\n\n let content: string;\n let report: Report;\n if (format === 'md') {\n let markdown: string;\n if (words !== undefined) {\n ({ markdown, report } = await extractOcr({ words }, ocrCommon));\n } else {\n ({ markdown, report } = await pdfToMarkdown(src, {\n ...ocrCommon,\n ocr: values.ocr as boolean,\n noOcr: values['no-ocr'] as boolean,\n pageMarkers: values['page-markers'] as boolean,\n }));\n }\n content = markdown;\n } else {\n // raw/txt render the same routed Analysis as the Markdown API, so a\n // format switch cannot discard a sparse image page or change audit input.\n let analysis: Analysis;\n if (words !== undefined) {\n analysis = await analyze({ words }, { ...ocrCommon, path: 'ocr' });\n } else {\n analysis = (\n await analyzePdfAuto(src, {\n ...ocrCommon,\n ocr: values.ocr as boolean,\n noOcr: values['no-ocr'] as boolean,\n pageMarkers: values['page-markers'] as boolean,\n })\n ).analysis;\n }\n report = analysis.report;\n content = format === 'raw' ? JSON.stringify(analysis, null, 1) : toText(analysis);\n }\n\n if (values.stdout) process.stdout.write(content);\n else writeFileSync(dest, content);\n\n // The library's Report stays clean of source/out; --json consumers keep\n // report.json files around and need to know which input produced each one, so the CLI (only) decorates the printed JSON with source/out here.\n if (values.json) console.log(JSON.stringify({ source, out: dest, ...report }, null, 1));\n else if (report.path === 'ocr') {\n console.error(`${path.basename(dest).padEnd(32)} OCR path heads=${report.headings}(junk ${report.junkHeadings}) ` + `paras=${report.paragraphs} callouts=${report.callouts} floats=${report.floats ?? 0} dangling=${report.danglingLong} (long) ` + `~${report.charsPerPage} chars/page`);\n } else {\n console.error(\n `${path.basename(dest).padEnd(32)} body=${report.bodyFontHeight}pt heads=${report.headings}(junk ${report.junkHeadings}) ` +\n `lists=${report.listItems} paras=${report.paragraphs} callouts=${report.callouts} pairs=${report.joinedPairs} ` +\n `~${report.charsPerPage} chars/page` +\n (report.sparse ? ' ⚠️ implausibly sparse — pages are likely images; re-run with --ocr' : '')\n );\n }\n};\n\n// Parse one word-dump JSON line per word — key order is not guaranteed, parse\n// never grep. Non-JSON lines are skipped with a count, not fatal (the auditor applies the same policy).\nfunction parseWordLines(raw: string): { words: OcrWordInput[]; skipped: number } {\n const words: OcrWordInput[] = [];\n let skipped = 0;\n for (const line of raw.split('\\n')) {\n if (!line.trim()) continue;\n try {\n words.push(JSON.parse(line));\n } catch {\n skipped++;\n }\n }\n return { words, skipped };\n}\n\nexport default extract;\n"],"names":["readFileSync","writeFileSync","path","parseArgs","analyze","analyzePdfAuto","extractOcr","pdfToMarkdown","toText","USAGE","FORMATS","TUNING_FLAGS","name","key","allowZero","isInt","extract","ctx","rest","includes","usageError","parsed","args","options","json","type","default","ocr","stdout","help","short","pages","dpi","format","Object","fromEntries","map","f","allowPositionals","err","message","values","positionals","console","log","src","outArg","wordsPath","formatRaw","join","JSON","stringify","undefined","m","exec","first","Number","NaN","last","tuning","spec","raw","value","ok","isFinite","Math","trunc","source","defaultExt","dest","replace","onWarning","error","showProgress","process","stderr","isTTY","onProgress","e","page","words","ms","debugWordsPath","trim","onWords","lines","w","entry","text","x","y","h","confidence","length","common","ocrCommon","skipped","parseWordLines","content","report","markdown","noOcr","pageMarkers","analysis","write","out","basename","padEnd","headings","junkHeadings","paragraphs","callouts","floats","danglingLong","charsPerPage","bodyFontHeight","listItems","joinedPairs","sparse","line","split","push","parse"],"mappings":"AAAA,2EAA2E;AAC3E,qHAAqH;AACrH,SAASA,YAAY,EAAEC,aAAa,QAAQ,UAAU;AACtD,OAAOC,UAAU,YAAY;AAC7B,SAASC,SAAS,QAAQ,YAAY;AACtC,SAASC,OAAO,QAAQ,gBAAgB;AAExC,SAASC,cAAc,EAAEC,UAAU,EAAEC,aAAa,QAAQ,gBAAgB;AAE1E,SAASC,MAAM,QAAQ,wBAAwB;AAI/C,MAAMC,QACJ,iGACA,yEACA,8GACA,8FACA,oDACA,uGACA;AAEF,MAAMC,UAAU;IAAC;IAAM;IAAO;CAAM;AAGpC,mHAAmH;AACnH,MAAMC,eAKA;IACJ;QACEC,MAAM;QACNC,KAAK;QACLC,WAAW;QACXC,OAAO;IACT;IACA;QAAEH,MAAM;QAAmBC,KAAK;IAAgB;IAChD;QAAED,MAAM;QAAgBC,KAAK;IAAc;IAC3C;QAAED,MAAM;QAAYC,KAAK;IAAU;IACnC;QAAED,MAAM;QAAiBC,KAAK;IAAe;IAC7C;QAAED,MAAM;QAAiBC,KAAK;IAAc;IAC5C;QAAED,MAAM;QAAaC,KAAK;IAAW;IACrC;QAAED,MAAM;QAAcC,KAAK;IAAW;IACtC;QAAED,MAAM;QAAqBC,KAAK;IAAkB;CACrD;AAED,MAAMG,UAAmB,OAAOC;IAC9B,uEAAuE;IACvE,kGAAkG;IAClG,IAAIA,IAAIC,IAAI,CAACC,QAAQ,CAAC,kBAAkB;QACtCF,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IAEA,MAAMY,SAAS,AAAC,CAAA;QACd,IAAI;YACF,OAAOlB,UAAU;gBACfmB,MAAML,IAAIC,IAAI;gBACdK,SAAS;oBACPC,MAAM;wBAAEC,MAAM;wBAAWC,SAAS;oBAAM;oBACxC,gBAAgB;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAClDC,KAAK;wBAAEF,MAAM;wBAAWC,SAAS;oBAAM;oBACvC,UAAU;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAC5CE,QAAQ;wBAAEH,MAAM;wBAAWC,SAAS;oBAAM;oBAC1C,eAAe;wBAAED,MAAM;oBAAS;oBAChCI,MAAM;wBAAEJ,MAAM;wBAAWC,SAAS;wBAAOI,OAAO;oBAAI;oBACpDC,OAAO;wBAAEN,MAAM;oBAAS;oBACxBO,KAAK;wBAAEP,MAAM;oBAAS;oBACtBQ,QAAQ;wBAAER,MAAM;wBAAUC,SAAS;oBAAK;oBACxC,cAAc;wBAAED,MAAM;oBAAS;oBAC/B,GAAGS,OAAOC,WAAW,CAACxB,aAAayB,GAAG,CAAC,CAACC,IAAM;4BAACA,EAAEzB,IAAI;4BAAE;gCAAEa,MAAM;4BAAkB;yBAAE,EAAE;gBACvF;gBACAa,kBAAkB;YACpB;QACF,EAAE,OAAOC,KAAK;YACZ,uEAAuE;YACvE,MAAMtB,IAAIG,UAAU,CAAC,GAAG,AAACmB,IAAcC,OAAO,CAAC,EAAE,EAAE/B,OAAO;QAC5D;IACF,CAAA;IACA,MAAMgC,SAASpB,OAAOoB,MAAM;IAC5B,MAAMC,cAAcrB,OAAOqB,WAAW;IAEtC,IAAID,OAAOZ,IAAI,EAAE;QACfc,QAAQC,GAAG,CAACnC;QACZ;IACF;IAEA,MAAM,CAACoC,KAAKC,OAAO,GAAGJ;IACtB,MAAMK,YAAYN,MAAM,CAAC,aAAa;IACtC,IAAI,CAACI,OAAO,CAACE,WAAW9B,IAAIG,UAAU,CAACX;IAEvC,MAAMuC,YAAYP,OAAOR,MAAM;IAC/B,IAAI,CAAC,AAACvB,QAA8BS,QAAQ,CAAC6B,YAAY;QACvD/B,IAAIG,UAAU,CAAC,GAAGX,MAAM,0BAA0B,EAAEC,QAAQuC,IAAI,CAAC,KAAK,MAAM,EAAEC,KAAKC,SAAS,CAACH,WAAW,CAAC,CAAC;IAC5G;IACA,MAAMf,SAASe;IAEf,wEAAwE;IACxE,0EAA0E;IAC1E,IAAIjB;IACJ,IAAIU,OAAOV,KAAK,KAAKqB,WAAW;QAC9B,MAAMC,IAAI,qBAAqBC,IAAI,CAACb,OAAOV,KAAK;QAChD,MAAMwB,QAAQF,IAAIG,OAAOH,CAAC,CAAC,EAAE,IAAIG,OAAOC,GAAG;QAC3C,MAAMC,OAAOL,IAAKA,CAAC,CAAC,EAAE,KAAKD,YAAYI,OAAOH,CAAC,CAAC,EAAE,IAAIE,QAASC,OAAOC,GAAG;QACzE,IAAI,CAACJ,KAAKE,QAAQ,KAAKG,OAAOH,OAAO;YACnCtC,IAAIG,UAAU,CAAC,GAAGX,MAAM,iDAAiD,EAAEyC,KAAKC,SAAS,CAACV,OAAOV,KAAK,EAAE,CAAC,CAAC;QAC5G;QACAA,QAAQ;YAAEwB;YAAOG;QAAK;IACxB;IAEA,MAAMC,SAA0B,CAAC;IACjC,KAAK,MAAMC,QAAQjD,aAAc;QAC/B,MAAMkD,MAAMpB,MAAM,CAACmB,KAAKhD,IAAI,CAAC;QAC7B,IAAIiD,QAAQT,WAAW;QACvB,MAAMU,QAAQN,OAAOK;QACrB,MAAME,KAAKF,QAAQ,MAAML,OAAOQ,QAAQ,CAACF,UAAWF,CAAAA,KAAK9C,SAAS,GAAGgD,SAAS,IAAIA,QAAQ,CAAA;QAC1F,IAAI,CAACC,IAAI;YACP9C,IAAIG,UAAU,CAAC,GAAGX,MAAM,IAAI,EAAEmD,KAAKhD,IAAI,CAAC,YAAY,EAAEgD,KAAK9C,SAAS,GAAG,iBAAiB,WAAW,aAAa,EAAEoC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1I;QACCF,MAAM,CAACC,KAAK/C,GAAG,CAAC,GAAc+C,KAAK7C,KAAK,GAAGkD,KAAKC,KAAK,CAACJ,SAASA;IAClE;IAEA,0EAA0E;IAC1E,8GAA8G;IAC9G,IAAI9B;IACJ,IAAIS,OAAOT,GAAG,KAAKoB,WAAW;QAC5B,MAAMS,MAAMpB,OAAOT,GAAG;QACtB,MAAM8B,QAAQN,OAAOK;QACrB,IAAIA,QAAQ,MAAM,CAACL,OAAOQ,QAAQ,CAACF,UAAUA,SAAS,GAAG;YACvD7C,IAAIG,UAAU,CAAC,GAAGX,MAAM,wCAAwC,EAAEyC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1F;QACA7B,MAAM8B;IACR;IAEA,sEAAsE;IACtE,kIAAkI;IAClI,MAAMK,SAAStB,gBAAAA,iBAAAA,MAAQE;IACvB,MAAMqB,aAAanC,WAAW,QAAQ,SAASA,WAAW,QAAQ,QAAQ;IAC1E,MAAMoC,OAAOvB,UAAU,GAAGqB,OAAOG,OAAO,CAAC,mBAAmB,IAAI,CAAC,EAAEF,YAAY;IAE/E,MAAMG,YAAY,CAAClB,IAAcV,QAAQ6B,KAAK,CAACnB;IAC/C,MAAMoB,eAAeC,QAAQC,MAAM,CAACC,KAAK;IACzC,gEAAgE;IAChE,iHAAiH;IACjH,MAAMC,aAAa,CAACC;QAClB,IAAIL,cAAc9B,QAAQ6B,KAAK,CAAC,CAAC,CAAC,EAAEM,EAAEC,IAAI,CAAC,EAAE,EAAED,EAAEE,KAAK,CAAC,QAAQ,EAAEF,EAAEG,EAAE,CAAC,EAAE,CAAC;IAC3E;IACA,MAAMC,iBAAiBzC,MAAM,CAAC,cAAc;IAC5C,IAAIyC,mBAAmB9B,aAAa8B,eAAeC,IAAI,OAAO,IAAI;QAChElE,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IACA,uEAAuE;IACvE,4IAA4I;IAC5I,MAAM2E,UAAUF,iBACZ,CAACF;QACC,MAAMK,QAAQL,MAAM5C,GAAG,CAAC,CAACkD;YACvB,MAAMC,QAAsB;gBAAER,MAAMO,EAAEP,IAAI;gBAAES,MAAMF,EAAEE,IAAI;gBAAEC,GAAGH,EAAEG,CAAC;gBAAEC,GAAGJ,EAAEI,CAAC;gBAAEJ,GAAGA,EAAEA,CAAC;gBAAEK,GAAGL,EAAEK,CAAC;YAAC;YACzF,IAAIL,EAAEM,UAAU,KAAKxC,WAAWmC,MAAMK,UAAU,GAAGN,EAAEM,UAAU;YAC/D,OAAO1C,KAAKC,SAAS,CAACoC;QACxB;QACAtF,cAAciF,gBAAgBG,MAAMQ,MAAM,GAAG,GAAGR,MAAMpC,IAAI,CAAC,MAAM,EAAE,CAAC,GAAG;IACzE,IACAG;IAEJ,MAAM0C,SAAS;QAAE/D;QAAOwC;QAAWa;IAAQ;IAC3C,MAAMW,YAAY;QAChB,GAAGD,MAAM;QACTnC;QACA3B;QACA6C;QACAO;IACF;IAEA,IAAIJ;IACJ,IAAIgB,UAAU;IACd,IAAIjD,cAAcK,WAAW;QAC1B,CAAA,EAAE4B,KAAK,EAAEgB,OAAO,EAAE,GAAGC,eAAejG,aAAa+C,WAAW,QAAO;QACpE,IAAIiD,UAAU,GAAGrD,QAAQ6B,KAAK,CAAC,CAAC,sBAAsB,EAAEwB,QAAQ,iBAAiB,CAAC;IACpF;IAEA,IAAIE;IACJ,IAAIC;IACJ,IAAIlE,WAAW,MAAM;QACnB,IAAImE;QACJ,IAAIpB,UAAU5B,WAAW;YACtB,CAAA,EAAEgD,QAAQ,EAAED,MAAM,EAAE,GAAG,MAAM7F,WAAW;gBAAE0E;YAAM,GAAGe,UAAS;QAC/D,OAAO;YACJ,CAAA,EAAEK,QAAQ,EAAED,MAAM,EAAE,GAAG,MAAM5F,cAAcsC,KAAK;gBAC/C,GAAGkD,SAAS;gBACZpE,KAAKc,OAAOd,GAAG;gBACf0E,OAAO5D,MAAM,CAAC,SAAS;gBACvB6D,aAAa7D,MAAM,CAAC,eAAe;YACrC,EAAC;QACH;QACAyD,UAAUE;IACZ,OAAO;QACL,oEAAoE;QACpE,0EAA0E;QAC1E,IAAIG;QACJ,IAAIvB,UAAU5B,WAAW;YACvBmD,WAAW,MAAMnG,QAAQ;gBAAE4E;YAAM,GAAG;gBAAE,GAAGe,SAAS;gBAAE7F,MAAM;YAAM;QAClE,OAAO;YACLqG,WAAW,AACT,CAAA,MAAMlG,eAAewC,KAAK;gBACxB,GAAGkD,SAAS;gBACZpE,KAAKc,OAAOd,GAAG;gBACf0E,OAAO5D,MAAM,CAAC,SAAS;gBACvB6D,aAAa7D,MAAM,CAAC,eAAe;YACrC,EAAC,EACD8D,QAAQ;QACZ;QACAJ,SAASI,SAASJ,MAAM;QACxBD,UAAUjE,WAAW,QAAQiB,KAAKC,SAAS,CAACoD,UAAU,MAAM,KAAK/F,OAAO+F;IAC1E;IAEA,IAAI9D,OAAOb,MAAM,EAAE8C,QAAQ9C,MAAM,CAAC4E,KAAK,CAACN;SACnCjG,cAAcoE,MAAM6B;IAEzB,wEAAwE;IACxE,8IAA8I;IAC9I,IAAIzD,OAAOjB,IAAI,EAAEmB,QAAQC,GAAG,CAACM,KAAKC,SAAS,CAAC;QAAEgB;QAAQsC,KAAKpC;QAAM,GAAG8B,MAAM;IAAC,GAAG,MAAM;SAC/E,IAAIA,OAAOjG,IAAI,KAAK,OAAO;YACuJiG;QAArLxD,QAAQ6B,KAAK,CAAC,GAAGtE,KAAKwG,QAAQ,CAACrC,MAAMsC,MAAM,CAAC,IAAI,gBAAgB,EAAER,OAAOS,QAAQ,CAAC,MAAM,EAAET,OAAOU,YAAY,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,EAAEV,OAAOW,UAAU,CAAC,UAAU,EAAEX,OAAOY,QAAQ,CAAC,QAAQ,GAAEZ,iBAAAA,OAAOa,MAAM,cAAbb,4BAAAA,iBAAiB,EAAE,UAAU,EAAEA,OAAOc,YAAY,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAEd,OAAOe,YAAY,CAAC,WAAW,CAAC;IAC1R,OAAO;QACLvE,QAAQ6B,KAAK,CACX,GAAGtE,KAAKwG,QAAQ,CAACrC,MAAMsC,MAAM,CAAC,IAAI,MAAM,EAAER,OAAOgB,cAAc,CAAC,SAAS,EAAEhB,OAAOS,QAAQ,CAAC,MAAM,EAAET,OAAOU,YAAY,CAAC,EAAE,CAAC,GACxH,CAAC,MAAM,EAAEV,OAAOiB,SAAS,CAAC,OAAO,EAAEjB,OAAOW,UAAU,CAAC,UAAU,EAAEX,OAAOY,QAAQ,CAAC,OAAO,EAAEZ,OAAOkB,WAAW,CAAC,CAAC,CAAC,GAC/G,CAAC,CAAC,EAAElB,OAAOe,YAAY,CAAC,WAAW,CAAC,GACnCf,CAAAA,OAAOmB,MAAM,GAAG,0EAA0E,EAAC;IAElG;AACF;AAEA,8EAA8E;AAC9E,wGAAwG;AACxG,SAASrB,eAAepC,GAAW;IACjC,MAAMmB,QAAwB,EAAE;IAChC,IAAIgB,UAAU;IACd,KAAK,MAAMuB,QAAQ1D,IAAI2D,KAAK,CAAC,MAAO;QAClC,IAAI,CAACD,KAAKpC,IAAI,IAAI;QAClB,IAAI;YACFH,MAAMyC,IAAI,CAACvE,KAAKwE,KAAK,CAACH;QACxB,EAAE,OAAM;YACNvB;QACF;IACF;IACA,OAAO;QAAEhB;QAAOgB;IAAQ;AAC1B;AAEA,eAAehF,QAAQ"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/cli/index.ts"],"sourcesContent":["import type { Command } from './types.ts';\n\n// Lazy registry: a command's imports load only when it runs. These names are\n// reserved — a PDF literally named `extract`/`render` (no extension) needs a ./ prefix.\nexport const COMMANDS: Record<string, () => Promise<{ default: Command }>> = {\n extract: () => import('./extract.ts'),\n render: () => import('./render.ts'),\n audit: () => import('./audit.ts'),\n};\n\nexport type { Command, Ctx } from './types.ts';\n"],"names":["COMMANDS","extract","render","audit"],"mappings":"AAEA,6EAA6E;AAC7E,wFAAwF;AACxF,OAAO,MAAMA,WAAgE;IAC3EC,SAAS,IAAM,MAAM,CAAC;IACtBC,QAAQ,IAAM,MAAM,CAAC;IACrBC,OAAO,IAAM,MAAM,CAAC;AACtB,EAAE"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/cli/index.ts"],"sourcesContent":["import type { Command } from './types.ts';\n\n// Lazy registry: a command's imports load only when it runs. These names are\n// reserved — a PDF literally named `extract`/`render` (no extension) needs a ./ prefix.\nexport const COMMANDS: Record<string, () => Promise<{ default: Command }>> = {\n extract: () => import('./extract.ts'),\n render: () => import('./render.ts'),\n audit: () => import('./audit.ts'),\n};\n\nexport type { Command, Ctx } from './types.ts';\n"],"names":["COMMANDS","extract","render","audit"],"mappings":"AAEA,6EAA6E;AAC7E,wFAAwF;AACxF,OAAO,MAAMA,WAAgE;IAC3EC,SAAS,IAAM,MAAM,CAAC;IACtBC,QAAQ,IAAM,MAAM,CAAC;IACrBC,OAAO,IAAM,MAAM,CAAC;AACtB,EAAE"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/cli/render.ts"],"sourcesContent":["// render command — render PDF pages to PNG for human/vision-model\n// verification (extraction output is a claim; the page itself is the truth). Thin wrapper: parse flags, call src/raster.ts, write files.\nimport { mkdirSync, writeFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { parseArgs } from 'node:util';\nimport { PdfToMdError } from '../errors.ts';\nimport { awaitPdfOpen, openPdfForRender } from '../pdf-open.ts';\nimport { renderLoadedPage } from '../raster.ts';\nimport type { Command } from './types.ts';\n\nconst USAGE = 'usage: pdf-to-md render <file.pdf> <page>[-<page>] [--dpi|-d N (default 288)] [--out|-o DIR (default .)]';\n\nconst render: Command = async (ctx) => {\n const parsed = (() => {\n try {\n return parseArgs({\n args: ctx.rest,\n options: {\n dpi: { type: 'string', default: '288', short: 'd' },\n out: { type: 'string', short: 'o' },\n help: { type: 'boolean', default: false, short: 'h' },\n },\n allowPositionals: true,\n });\n } catch (err) {\n // usageError exits; the throw is dead code that tells the compiler so.\n throw ctx.usageError(`${(err as Error).message}\\n${USAGE}`);\n }\n })();\n const { values, positionals } = parsed;\n\n if (values.help) {\n console.log(USAGE);\n return;\n }\n\n const [pdfPath, spec] = positionals;\n if (!pdfPath || !spec) ctx.usageError(USAGE);\n\n const m = /^(\\d+)(?:-(\\d+))?$/.exec(spec);\n if (m === null) throw ctx.usageError(`${USAGE}\\ncould not parse page spec: ${spec}`);\n const firstPage = Number(m[1]);\n const lastPage = m[2] !== undefined ? Number(m[2]) : firstPage;\n if (!(firstPage >= 1) || !(lastPage >= firstPage)) throw ctx.usageError(`${USAGE}\\ninvalid page spec: ${spec}`);\n\n const dpi = Number.parseFloat(values.dpi as string);\n if (!(dpi > 0)) ctx.usageError(`${USAGE}\\n--dpi must be positive`);\n\n const outDir = values.out ?? process.cwd();\n mkdirSync(outDir, { recursive: true });\n const stem = path.basename(pdfPath, path.extname(pdfPath)).replaceAll(' ', '-');\n\n const loadingTask = openPdfForRender(pdfPath);\n try {\n const doc = await awaitPdfOpen(loadingTask, pdfPath);\n\n // Whole-range check before any file is written, so a partly-valid range\n // fails atomically; renderLoadedPage re-validates per page.\n if (lastPage > doc.numPages) {\n throw new PdfToMdError('PAGE_RANGE', `page range ${firstPage}-${lastPage} is outside this document (1-${doc.numPages})`);\n }\n for (let n = firstPage; n <= lastPage; n++) {\n const { png } = await renderLoadedPage(doc, n, dpi);\n const outPath = path.join(outDir, `${stem}-p${n}.png`);\n writeFileSync(outPath, png);\n console.log(outPath);\n }\n } finally {\n await loadingTask.destroy();\n }\n};\n\nexport default render;\n"],"names":["mkdirSync","writeFileSync","path","parseArgs","PdfToMdError","awaitPdfOpen","openPdfForRender","renderLoadedPage","USAGE","render","ctx","values","parsed","args","rest","options","dpi","type","default","short","out","help","allowPositionals","err","usageError","message","positionals","console","log","pdfPath","spec","m","exec","firstPage","Number","lastPage","undefined","parseFloat","outDir","process","cwd","recursive","stem","basename","extname","replaceAll","loadingTask","doc","numPages","n","png","outPath","join","destroy"],"mappings":"AAAA,kEAAkE;AAClE,yIAAyI;AACzI,SAASA,SAAS,EAAEC,aAAa,QAAQ,UAAU;AACnD,OAAOC,UAAU,YAAY;AAC7B,SAASC,SAAS,QAAQ,YAAY;AACtC,SAASC,YAAY,QAAQ,eAAe;AAC5C,SAASC,YAAY,EAAEC,gBAAgB,QAAQ,iBAAiB;AAChE,SAASC,gBAAgB,QAAQ,eAAe;AAGhD,MAAMC,QAAQ;AAEd,MAAMC,SAAkB,OAAOC;QAoCdC;IAnCf,MAAMC,SAAS,AAAC,CAAA;QACd,IAAI;YACF,OAAOT,UAAU;gBACfU,MAAMH,IAAII,IAAI;gBACdC,SAAS;oBACPC,KAAK;wBAAEC,MAAM;wBAAUC,SAAS;wBAAOC,OAAO;oBAAI;oBAClDC,KAAK;wBAAEH,MAAM;wBAAUE,OAAO;oBAAI;oBAClCE,MAAM;wBAAEJ,MAAM;wBAAWC,SAAS;wBAAOC,OAAO;oBAAI;gBACtD;gBACAG,kBAAkB;YACpB;QACF,EAAE,OAAOC,KAAK;YACZ,uEAAuE;YACvE,MAAMb,IAAIc,UAAU,CAAC,GAAG,AAACD,IAAcE,OAAO,CAAC,EAAE,EAAEjB,OAAO;QAC5D;IACF,CAAA;IACA,MAAM,EAAEG,MAAM,EAAEe,WAAW,EAAE,GAAGd;IAEhC,IAAID,OAAOU,IAAI,EAAE;QACfM,QAAQC,GAAG,CAACpB;QACZ;IACF;IAEA,MAAM,CAACqB,SAASC,KAAK,GAAGJ;IACxB,IAAI,CAACG,WAAW,CAACC,MAAMpB,IAAIc,UAAU,CAAChB;IAEtC,MAAMuB,IAAI,qBAAqBC,IAAI,CAACF;IACpC,IAAIC,MAAM,MAAM,MAAMrB,IAAIc,UAAU,CAAC,GAAGhB,MAAM,6BAA6B,EAAEsB,MAAM;IACnF,MAAMG,YAAYC,OAAOH,CAAC,CAAC,EAAE;IAC7B,MAAMI,WAAWJ,CAAC,CAAC,EAAE,KAAKK,YAAYF,OAAOH,CAAC,CAAC,EAAE,IAAIE;IACrD,IAAI,CAAEA,CAAAA,aAAa,CAAA,KAAM,CAAEE,CAAAA,YAAYF,SAAQ,GAAI,MAAMvB,IAAIc,UAAU,CAAC,GAAGhB,MAAM,qBAAqB,EAAEsB,MAAM;IAE9G,MAAMd,MAAMkB,OAAOG,UAAU,CAAC1B,OAAOK,GAAG;IACxC,IAAI,CAAEA,CAAAA,MAAM,CAAA,GAAIN,IAAIc,UAAU,CAAC,GAAGhB,MAAM,wBAAwB,CAAC;IAEjE,MAAM8B,UAAS3B,cAAAA,OAAOS,GAAG,cAAVT,yBAAAA,cAAc4B,QAAQC,GAAG;IACxCxC,UAAUsC,QAAQ;QAAEG,WAAW;IAAK;IACpC,MAAMC,OAAOxC,KAAKyC,QAAQ,CAACd,SAAS3B,KAAK0C,OAAO,CAACf,UAAUgB,UAAU,CAAC,KAAK;IAE3E,MAAMC,cAAcxC,iBAAiBuB;IACrC,IAAI;QACF,MAAMkB,MAAM,MAAM1C,aAAayC,aAAajB;QAE5C,wEAAwE;QACxE,4DAA4D;QAC5D,IAAIM,WAAWY,IAAIC,QAAQ,EAAE;YAC3B,MAAM,IAAI5C,aAAa,cAAc,CAAC,WAAW,EAAE6B,UAAU,CAAC,EAAEE,SAAS,6BAA6B,EAAEY,IAAIC,QAAQ,CAAC,CAAC,CAAC;QACzH;QACA,IAAK,IAAIC,IAAIhB,WAAWgB,KAAKd,UAAUc,IAAK;YAC1C,MAAM,EAAEC,GAAG,EAAE,GAAG,MAAM3C,iBAAiBwC,KAAKE,GAAGjC;YAC/C,MAAMmC,UAAUjD,KAAKkD,IAAI,CAACd,QAAQ,GAAGI,KAAK,EAAE,EAAEO,EAAE,IAAI,CAAC;YACrDhD,cAAckD,SAASD;YACvBvB,QAAQC,GAAG,CAACuB;QACd;IACF,SAAU;QACR,MAAML,YAAYO,OAAO;IAC3B;AACF;AAEA,eAAe5C,OAAO"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/cli/render.ts"],"sourcesContent":["// render command — render PDF pages to PNG for human/vision-model\n// verification (extraction output is a claim; the page itself is the truth). Thin wrapper: parse flags, call src/raster.ts, write files.\nimport { mkdirSync, writeFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { parseArgs } from 'node:util';\nimport { PdfToMdError } from '../errors.ts';\nimport { awaitPdfOpen, openPdfForRender } from '../pdf-open.ts';\nimport { renderLoadedPage } from '../raster.ts';\nimport type { Command } from './types.ts';\n\nconst USAGE = 'usage: pdf-to-md render <file.pdf> <page>[-<page>] [--dpi|-d N (default 288)] [--out|-o DIR (default .)]';\n\nconst render: Command = async (ctx) => {\n const parsed = (() => {\n try {\n return parseArgs({\n args: ctx.rest,\n options: {\n dpi: { type: 'string', default: '288', short: 'd' },\n out: { type: 'string', short: 'o' },\n help: { type: 'boolean', default: false, short: 'h' },\n },\n allowPositionals: true,\n });\n } catch (err) {\n // usageError exits; the throw is dead code that tells the compiler so.\n throw ctx.usageError(`${(err as Error).message}\\n${USAGE}`);\n }\n })();\n const { values, positionals } = parsed;\n\n if (values.help) {\n console.log(USAGE);\n return;\n }\n\n const [pdfPath, spec] = positionals;\n if (!pdfPath || !spec) ctx.usageError(USAGE);\n\n const m = /^(\\d+)(?:-(\\d+))?$/.exec(spec);\n if (m === null) throw ctx.usageError(`${USAGE}\\ncould not parse page spec: ${spec}`);\n const firstPage = Number(m[1]);\n const lastPage = m[2] !== undefined ? Number(m[2]) : firstPage;\n if (!(firstPage >= 1) || !(lastPage >= firstPage)) throw ctx.usageError(`${USAGE}\\ninvalid page spec: ${spec}`);\n\n const dpi = Number.parseFloat(values.dpi as string);\n if (!(dpi > 0)) ctx.usageError(`${USAGE}\\n--dpi must be positive`);\n\n const outDir = values.out ?? process.cwd();\n mkdirSync(outDir, { recursive: true });\n const stem = path.basename(pdfPath, path.extname(pdfPath)).replaceAll(' ', '-');\n\n const loadingTask = openPdfForRender(pdfPath);\n try {\n const doc = await awaitPdfOpen(loadingTask, pdfPath);\n\n // Whole-range check before any file is written, so a partly-valid range\n // fails atomically; renderLoadedPage re-validates per page.\n if (lastPage > doc.numPages) {\n throw new PdfToMdError('PAGE_RANGE', `page range ${firstPage}-${lastPage} is outside this document (1-${doc.numPages})`);\n }\n for (let n = firstPage; n <= lastPage; n++) {\n const { png } = await renderLoadedPage(doc, n, dpi);\n const outPath = path.join(outDir, `${stem}-p${n}.png`);\n writeFileSync(outPath, png);\n console.log(outPath);\n }\n } finally {\n await loadingTask.destroy();\n }\n};\n\nexport default render;\n"],"names":["mkdirSync","writeFileSync","path","parseArgs","PdfToMdError","awaitPdfOpen","openPdfForRender","renderLoadedPage","USAGE","render","ctx","values","parsed","args","rest","options","dpi","type","default","short","out","help","allowPositionals","err","usageError","message","positionals","console","log","pdfPath","spec","m","exec","firstPage","Number","lastPage","undefined","parseFloat","outDir","process","cwd","recursive","stem","basename","extname","replaceAll","loadingTask","doc","numPages","n","png","outPath","join","destroy"],"mappings":"AAAA,kEAAkE;AAClE,yIAAyI;AACzI,SAASA,SAAS,EAAEC,aAAa,QAAQ,UAAU;AACnD,OAAOC,UAAU,YAAY;AAC7B,SAASC,SAAS,QAAQ,YAAY;AACtC,SAASC,YAAY,QAAQ,eAAe;AAC5C,SAASC,YAAY,EAAEC,gBAAgB,QAAQ,iBAAiB;AAChE,SAASC,gBAAgB,QAAQ,eAAe;AAGhD,MAAMC,QAAQ;AAEd,MAAMC,SAAkB,OAAOC;QAoCdC;IAnCf,MAAMC,SAAS,AAAC,CAAA;QACd,IAAI;YACF,OAAOT,UAAU;gBACfU,MAAMH,IAAII,IAAI;gBACdC,SAAS;oBACPC,KAAK;wBAAEC,MAAM;wBAAUC,SAAS;wBAAOC,OAAO;oBAAI;oBAClDC,KAAK;wBAAEH,MAAM;wBAAUE,OAAO;oBAAI;oBAClCE,MAAM;wBAAEJ,MAAM;wBAAWC,SAAS;wBAAOC,OAAO;oBAAI;gBACtD;gBACAG,kBAAkB;YACpB;QACF,EAAE,OAAOC,KAAK;YACZ,uEAAuE;YACvE,MAAMb,IAAIc,UAAU,CAAC,GAAG,AAACD,IAAcE,OAAO,CAAC,EAAE,EAAEjB,OAAO;QAC5D;IACF,CAAA;IACA,MAAM,EAAEG,MAAM,EAAEe,WAAW,EAAE,GAAGd;IAEhC,IAAID,OAAOU,IAAI,EAAE;QACfM,QAAQC,GAAG,CAACpB;QACZ;IACF;IAEA,MAAM,CAACqB,SAASC,KAAK,GAAGJ;IACxB,IAAI,CAACG,WAAW,CAACC,MAAMpB,IAAIc,UAAU,CAAChB;IAEtC,MAAMuB,IAAI,qBAAqBC,IAAI,CAACF;IACpC,IAAIC,MAAM,MAAM,MAAMrB,IAAIc,UAAU,CAAC,GAAGhB,MAAM,6BAA6B,EAAEsB,MAAM;IACnF,MAAMG,YAAYC,OAAOH,CAAC,CAAC,EAAE;IAC7B,MAAMI,WAAWJ,CAAC,CAAC,EAAE,KAAKK,YAAYF,OAAOH,CAAC,CAAC,EAAE,IAAIE;IACrD,IAAI,CAAEA,CAAAA,aAAa,CAAA,KAAM,CAAEE,CAAAA,YAAYF,SAAQ,GAAI,MAAMvB,IAAIc,UAAU,CAAC,GAAGhB,MAAM,qBAAqB,EAAEsB,MAAM;IAE9G,MAAMd,MAAMkB,OAAOG,UAAU,CAAC1B,OAAOK,GAAG;IACxC,IAAI,CAAEA,CAAAA,MAAM,CAAA,GAAIN,IAAIc,UAAU,CAAC,GAAGhB,MAAM,wBAAwB,CAAC;IAEjE,MAAM8B,UAAS3B,cAAAA,OAAOS,GAAG,cAAVT,yBAAAA,cAAc4B,QAAQC,GAAG;IACxCxC,UAAUsC,QAAQ;QAAEG,WAAW;IAAK;IACpC,MAAMC,OAAOxC,KAAKyC,QAAQ,CAACd,SAAS3B,KAAK0C,OAAO,CAACf,UAAUgB,UAAU,CAAC,KAAK;IAE3E,MAAMC,cAAcxC,iBAAiBuB;IACrC,IAAI;QACF,MAAMkB,MAAM,MAAM1C,aAAayC,aAAajB;QAE5C,wEAAwE;QACxE,4DAA4D;QAC5D,IAAIM,WAAWY,IAAIC,QAAQ,EAAE;YAC3B,MAAM,IAAI5C,aAAa,cAAc,CAAC,WAAW,EAAE6B,UAAU,CAAC,EAAEE,SAAS,6BAA6B,EAAEY,IAAIC,QAAQ,CAAC,CAAC,CAAC;QACzH;QACA,IAAK,IAAIC,IAAIhB,WAAWgB,KAAKd,UAAUc,IAAK;YAC1C,MAAM,EAAEC,GAAG,EAAE,GAAG,MAAM3C,iBAAiBwC,KAAKE,GAAGjC;YAC/C,MAAMmC,UAAUjD,KAAKkD,IAAI,CAACd,QAAQ,GAAGI,KAAK,EAAE,EAAEO,EAAE,IAAI,CAAC;YACrDhD,cAAckD,SAASD;YACvBvB,QAAQC,GAAG,CAACuB;QACd;IACF,SAAU;QACR,MAAML,YAAYO,OAAO;IAC3B;AACF;AAEA,eAAe5C,OAAO"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/cli/types.ts"],"sourcesContent":["// Handed to every command; the command word is peeled off in src/cli.ts and\n// each command owns its own parseArgs table over `rest` (our commands' flags are disjoint, so a shared table would let `render --para-gap=2` parse silently as a no-op).\nexport interface Ctx {\n /** binary name, for usage strings */\n name: string;\n /** argv after the command word */\n rest: string[];\n /** the CLI's fixed non-zero exit code, for result signals like audit loss */\n readonly errorCode: number;\n /** print the message and exit with the CLI's error code, on every malformed-invocation path.\n * Readonly property (not method) syntax: TS only applies never-return control-flow analysis to readonly property-typed function members. */\n readonly usageError: (message: string) => never;\n}\n\nexport type Command = (ctx: Ctx) => Promise<void> | void;\n"],"names":[],"mappings":"AAAA,4EAA4E;AAC5E,yKAAyK;AAazK,WAAyD"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/cli/types.ts"],"sourcesContent":["// Handed to every command; the command word is peeled off in src/cli.ts and\n// each command owns its own parseArgs table over `rest` (our commands' flags are disjoint, so a shared table would let `render --para-gap=2` parse silently as a no-op).\nexport interface Ctx {\n /** binary name, for usage strings */\n name: string;\n /** argv after the command word */\n rest: string[];\n /** the CLI's fixed non-zero exit code, for result signals like audit loss */\n readonly errorCode: number;\n /** print the message and exit with the CLI's error code, on every malformed-invocation path.\n * Readonly property (not method) syntax: TS only applies never-return control-flow analysis to readonly property-typed function members. */\n readonly usageError: (message: string) => never;\n}\n\nexport type Command = (ctx: Ctx) => Promise<void> | void;\n"],"names":[],"mappings":"AAAA,4EAA4E;AAC5E,yKAAyK;AAazK,WAAyD"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/cli.ts"],"sourcesContent":["// Parsing and dispatch only. Commands live in src/cli/, one file each,\n// lazy-loaded — nothing tree- or dependency-heavy may be imported at the top of this file (`render` must never load the OCR engine, `--version` must load almost nothing).\nimport { readFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { fileURLToPath } from 'node:url';\nimport { COMMANDS } from './cli/index.ts';\nimport type { Ctx } from './cli/types.ts';\n\n// __dirname (CJS) or its ESM equivalent, then two fixed hops up to the\n// package root. NOT `import.meta.resolve('pdf-to-md/package.json')` — Node's self-reference lookup throws ERR_MODULE_NOT_FOUND from a file this deep; a location-relative path is stable across both build targets.\nconst __dirname = path.dirname(typeof __filename !== 'undefined' ? __filename : fileURLToPath(import.meta.url));\nconst ERROR_CODE = 19;\n\nfunction packageVersion(): string {\n try {\n const pkgPath = path.resolve(__dirname, '..', '..', 'package.json');\n const pkg = JSON.parse(readFileSync(pkgPath, 'utf8')) as { version?: string };\n return pkg.version ?? 'unknown';\n } catch {\n return 'unknown';\n }\n}\n\nfunction usage(name: string): string {\n return (\n `usage: ${name} <file.pdf> [out.md] [flags...] convert a PDF to markdown\\n` +\n ` ${name} extract <file.pdf> [out.md] [flags...] (same, explicit)\\n` +\n ` ${name} render <file.pdf> <page>[-<page>] [--dpi|-d N (default 288)] [--out|-o DIR (default .)]\\n` +\n ` ${name} audit <words.jsonl> <file.md> severity-1 word-loss check\\n` +\n ` ${name} --version | --help\\n` +\n '\\n' +\n 'extract flags:\\n' +\n ' --ocr force the OCR path (image-only/scanned pages)\\n' +\n ' --no-ocr never fall back to OCR; keep sparse text output\\n' +\n ' --pages N[-M] convert only these 1-based source pages\\n' +\n ' --stdout write markdown to stdout instead of a file\\n' +\n ' --json print the full QA report as JSON\\n' +\n ` --page-markers emit a \"### pN\" heading per page\\n` +\n ' --format md|txt|raw output format (default md; raw = Analysis JSON incl. the\\n' +\n ' report; txt = reading-order plain text)\\n' +\n ' --words-json=FILE OCR path from a recognized-word dump (offline replay)\\n' +\n ' --debug-words=FILE dump every recognized word as JSON lines to FILE\\n' +\n ' --dpi=N OCR render resolution (default 288; raise for tiny print or\\n' +\n ' chart/label pages, >~400 measured worse on ordinary prose)\\n' +\n '\\n' +\n `extract tuning (measured defaults; see README's tuning table):\\n` +\n ' --float-max-words=N --float-max-width=F --float-margin=F --para-gap=F\\n' +\n ' --heading-scale=F --col-max-width=F --col-split=F --line-y-tol=F\\n' +\n ' --line-height-ratio=F\\n' +\n '\\n' +\n 'examples:\\n' +\n ` ${name} book.pdf # text layer -> book.md (auto-OCR if scanned)\\n` +\n ` ${name} scan.pdf out.md --ocr --json # force OCR, print the QA report\\n` +\n ` ${name} book.pdf --pages 27-31 --stdout # re-extract a page range to stdout\\n` +\n ` ${name} render book.pdf 27 # render p27 to PNG for eyeballing (288 dpi)\\n` +\n ` ${name} audit words.jsonl book.md # check no recognized word was lost`\n );\n}\n\n// All failures exit ERROR_CODE: usage errors with the usage text, thrown errors with the message verbatim.\nexport default async function cli(argv: string[], name: string): Promise<void> {\n if (argv[0] === '--version' || argv[0] === '-v') {\n console.log(packageVersion());\n return;\n }\n if (argv.length === 0 || argv[0] === '--help' || argv[0] === '-h') {\n // No args is a usage ERROR (bare `pdf-to-md` can't do anything); asking\n // for --help is not.\n if (argv.length === 0) {\n console.error(usage(name));\n process.exit(ERROR_CODE);\n }\n console.log(`${name} v${packageVersion()}`);\n console.log('');\n console.log(usage(name));\n return;\n }\n\n const ctx: Ctx = {\n name,\n rest: argv.slice(1),\n usageError(message) {\n console.error(message);\n process.exit(ERROR_CODE);\n },\n errorCode: ERROR_CODE,\n };\n\n try {\n const load = COMMANDS[argv[0]];\n if (load) {\n await (await load()).default(ctx);\n } else {\n // Compat shim: `pdf-to-md file.pdf [...]` — any first word that isn't\n // a reserved command is extract's own first argument.\n ctx.rest = argv;\n await (await COMMANDS.extract()).default(ctx);\n }\n } catch (err) {\n console.error((err as Error).message);\n process.exit(ERROR_CODE);\n }\n}\n"],"names":["readFileSync","path","fileURLToPath","COMMANDS","__dirname","dirname","__filename","url","ERROR_CODE","packageVersion","pkg","pkgPath","resolve","JSON","parse","version","usage","name","cli","argv","console","log","length","error","process","exit","ctx","rest","slice","usageError","message","errorCode","load","default","extract","err"],"mappings":"AAAA,uEAAuE;AACvE,2KAA2K;AAC3K,SAASA,YAAY,QAAQ,UAAU;AACvC,OAAOC,UAAU,YAAY;AAC7B,SAASC,aAAa,QAAQ,WAAW;AACzC,SAASC,QAAQ,QAAQ,iBAAiB;AAG1C,uEAAuE;AACvE,oNAAoN;AACpN,MAAMC,YAAYH,KAAKI,OAAO,CAAC,OAAOC,eAAe,cAAcA,aAAaJ,cAAc,YAAYK,GAAG;AAC7G,MAAMC,aAAa;AAEnB,SAASC;IACP,IAAI;YAGKC;QAFP,MAAMC,UAAUV,KAAKW,OAAO,CAACR,WAAW,MAAM,MAAM;QACpD,MAAMM,MAAMG,KAAKC,KAAK,CAACd,aAAaW,SAAS;QAC7C,QAAOD,eAAAA,IAAIK,OAAO,cAAXL,0BAAAA,eAAe;IACxB,EAAE,OAAM;QACN,OAAO;IACT;AACF;AAEA,SAASM,MAAMC,IAAY;IACzB,OACE,CAAC,OAAO,EAAEA,KAAK,kEAAkE,CAAC,GAClF,CAAC,OAAO,EAAEA,KAAK,0DAA0D,CAAC,GAC1E,CAAC,OAAO,EAAEA,KAAK,0FAA0F,CAAC,GAC1G,CAAC,OAAO,EAAEA,KAAK,oEAAoE,CAAC,GACpF,CAAC,OAAO,EAAEA,KAAK,qBAAqB,CAAC,GACrC,OACA,qBACA,2EACA,6EACA,qEACA,wEACA,8DACA,CAAC,yDAAyD,CAAC,GAC3D,sFACA,qEACA,mFACA,8EACA,yFACA,wFACA,OACA,CAAC,gEAAgE,CAAC,GAClE,8EACA,yEACA,8BACA,OACA,gBACA,CAAC,EAAE,EAAEA,KAAK,gFAAgF,CAAC,GAC3F,CAAC,EAAE,EAAEA,KAAK,mEAAmE,CAAC,GAC9E,CAAC,EAAE,EAAEA,KAAK,sEAAsE,CAAC,GACjF,CAAC,EAAE,EAAEA,KAAK,+EAA+E,CAAC,GAC1F,CAAC,EAAE,EAAEA,KAAK,oEAAoE,CAAC;AAEnF;AAEA,2GAA2G;AAC3G,eAAe,eAAeC,IAAIC,IAAc,EAAEF,IAAY;IAC5D,IAAIE,IAAI,CAAC,EAAE,KAAK,eAAeA,IAAI,CAAC,EAAE,KAAK,MAAM;QAC/CC,QAAQC,GAAG,CAACZ;QACZ;IACF;IACA,IAAIU,KAAKG,MAAM,KAAK,KAAKH,IAAI,CAAC,EAAE,KAAK,YAAYA,IAAI,CAAC,EAAE,KAAK,MAAM;QACjE,wEAAwE;QACxE,qBAAqB;QACrB,IAAIA,KAAKG,MAAM,KAAK,GAAG;YACrBF,QAAQG,KAAK,CAACP,MAAMC;YACpBO,QAAQC,IAAI,CAACjB;QACf;QACAY,QAAQC,GAAG,CAAC,GAAGJ,KAAK,EAAE,EAAER,kBAAkB;QAC1CW,QAAQC,GAAG,CAAC;QACZD,QAAQC,GAAG,CAACL,MAAMC;QAClB;IACF;IAEA,MAAMS,MAAW;QACfT;QACAU,MAAMR,KAAKS,KAAK,CAAC;QACjBC,YAAWC,OAAO;YAChBV,QAAQG,KAAK,CAACO;YACdN,QAAQC,IAAI,CAACjB;QACf;QACAuB,WAAWvB;IACb;IAEA,IAAI;QACF,MAAMwB,OAAO7B,QAAQ,CAACgB,IAAI,CAAC,EAAE,CAAC;QAC9B,IAAIa,MAAM;YACR,MAAM,AAAC,CAAA,MAAMA,MAAK,EAAGC,OAAO,CAACP;QAC/B,OAAO;YACL,sEAAsE;YACtE,sDAAsD;YACtDA,IAAIC,IAAI,GAAGR;YACX,MAAM,AAAC,CAAA,MAAMhB,SAAS+B,OAAO,EAAC,EAAGD,OAAO,CAACP;QAC3C;IACF,EAAE,OAAOS,KAAK;QACZf,QAAQG,KAAK,CAAC,AAACY,IAAcL,OAAO;QACpCN,QAAQC,IAAI,CAACjB;IACf;AACF"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/cli.ts"],"sourcesContent":["// Parsing and dispatch only. Commands live in src/cli/, one file each,\n// lazy-loaded — nothing tree- or dependency-heavy may be imported at the top of this file (`render` must never load the OCR engine, `--version` must load almost nothing).\nimport { readFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { fileURLToPath } from 'node:url';\nimport { COMMANDS } from './cli/index.ts';\nimport type { Ctx } from './cli/types.ts';\n\n// __dirname (CJS) or its ESM equivalent, then two fixed hops up to the\n// package root. NOT `import.meta.resolve('pdf-to-md/package.json')` — Node's self-reference lookup throws ERR_MODULE_NOT_FOUND from a file this deep; a location-relative path is stable across both build targets.\nconst __dirname = path.dirname(typeof __filename !== 'undefined' ? __filename : fileURLToPath(import.meta.url));\nconst ERROR_CODE = 19;\n\nfunction packageVersion(): string {\n try {\n const pkgPath = path.resolve(__dirname, '..', '..', 'package.json');\n const pkg = JSON.parse(readFileSync(pkgPath, 'utf8')) as { version?: string };\n return pkg.version ?? 'unknown';\n } catch {\n return 'unknown';\n }\n}\n\nfunction usage(name: string): string {\n return (\n `usage: ${name} <file.pdf> [out.md] [flags...] convert a PDF to markdown\\n` +\n ` ${name} extract <file.pdf> [out.md] [flags...] (same, explicit)\\n` +\n ` ${name} render <file.pdf> <page>[-<page>] [--dpi|-d N (default 288)] [--out|-o DIR (default .)]\\n` +\n ` ${name} audit <words.jsonl> <file.md> severity-1 word-loss check\\n` +\n ` ${name} --version | --help\\n` +\n '\\n' +\n 'extract flags:\\n' +\n ' --ocr force the OCR path (image-only/scanned pages)\\n' +\n ' --no-ocr never fall back to OCR; keep sparse text output\\n' +\n ' --pages N[-M] convert only these 1-based source pages\\n' +\n ' --stdout write markdown to stdout instead of a file\\n' +\n ' --json print the full QA report as JSON\\n' +\n ` --page-markers emit a \"### pN\" heading per page\\n` +\n ' --format md|txt|raw output format (default md; raw = Analysis JSON incl. the\\n' +\n ' report; txt = reading-order plain text)\\n' +\n ' --words-json=FILE OCR path from a recognized-word dump (offline replay)\\n' +\n ' --debug-words=FILE dump every recognized word as JSON lines to FILE\\n' +\n ' --dpi=N OCR render resolution (default 288; raise for tiny print or\\n' +\n ' chart/label pages, >~400 measured worse on ordinary prose)\\n' +\n '\\n' +\n `extract tuning (measured defaults; see README's tuning table):\\n` +\n ' --float-max-words=N --float-max-width=F --float-margin=F --para-gap=F\\n' +\n ' --heading-scale=F --col-max-width=F --col-split=F --line-y-tol=F\\n' +\n ' --line-height-ratio=F\\n' +\n '\\n' +\n 'examples:\\n' +\n ` ${name} book.pdf # text layer -> book.md (auto-OCR if scanned)\\n` +\n ` ${name} scan.pdf out.md --ocr --json # force OCR, print the QA report\\n` +\n ` ${name} book.pdf --pages 27-31 --stdout # re-extract a page range to stdout\\n` +\n ` ${name} render book.pdf 27 # render p27 to PNG for eyeballing (288 dpi)\\n` +\n ` ${name} audit words.jsonl book.md # check no recognized word was lost`\n );\n}\n\n// All failures exit ERROR_CODE: usage errors with the usage text, thrown errors with the message verbatim.\nexport default async function cli(argv: string[], name: string): Promise<void> {\n if (argv[0] === '--version' || argv[0] === '-v') {\n console.log(packageVersion());\n return;\n }\n if (argv.length === 0 || argv[0] === '--help' || argv[0] === '-h') {\n // No args is a usage ERROR (bare `pdf-to-md` can't do anything); asking\n // for --help is not.\n if (argv.length === 0) {\n console.error(usage(name));\n process.exit(ERROR_CODE);\n }\n console.log(`${name} v${packageVersion()}`);\n console.log('');\n console.log(usage(name));\n return;\n }\n\n const ctx: Ctx = {\n name,\n rest: argv.slice(1),\n usageError(message) {\n console.error(message);\n process.exit(ERROR_CODE);\n },\n errorCode: ERROR_CODE,\n };\n\n try {\n const load = COMMANDS[argv[0]];\n if (load) {\n await (await load()).default(ctx);\n } else {\n // Compat shim: `pdf-to-md file.pdf [...]` — any first word that isn't\n // a reserved command is extract's own first argument.\n ctx.rest = argv;\n await (await COMMANDS.extract()).default(ctx);\n }\n } catch (err) {\n console.error((err as Error).message);\n process.exit(ERROR_CODE);\n }\n}\n"],"names":["readFileSync","path","fileURLToPath","COMMANDS","__dirname","dirname","__filename","url","ERROR_CODE","packageVersion","pkg","pkgPath","resolve","JSON","parse","version","usage","name","cli","argv","console","log","length","error","process","exit","ctx","rest","slice","usageError","message","errorCode","load","default","extract","err"],"mappings":"AAAA,uEAAuE;AACvE,2KAA2K;AAC3K,SAASA,YAAY,QAAQ,UAAU;AACvC,OAAOC,UAAU,YAAY;AAC7B,SAASC,aAAa,QAAQ,WAAW;AACzC,SAASC,QAAQ,QAAQ,iBAAiB;AAG1C,uEAAuE;AACvE,oNAAoN;AACpN,MAAMC,YAAYH,KAAKI,OAAO,CAAC,OAAOC,eAAe,cAAcA,aAAaJ,cAAc,YAAYK,GAAG;AAC7G,MAAMC,aAAa;AAEnB,SAASC;IACP,IAAI;YAGKC;QAFP,MAAMC,UAAUV,KAAKW,OAAO,CAACR,WAAW,MAAM,MAAM;QACpD,MAAMM,MAAMG,KAAKC,KAAK,CAACd,aAAaW,SAAS;QAC7C,QAAOD,eAAAA,IAAIK,OAAO,cAAXL,0BAAAA,eAAe;IACxB,EAAE,OAAM;QACN,OAAO;IACT;AACF;AAEA,SAASM,MAAMC,IAAY;IACzB,OACE,CAAC,OAAO,EAAEA,KAAK,kEAAkE,CAAC,GAClF,CAAC,OAAO,EAAEA,KAAK,0DAA0D,CAAC,GAC1E,CAAC,OAAO,EAAEA,KAAK,0FAA0F,CAAC,GAC1G,CAAC,OAAO,EAAEA,KAAK,oEAAoE,CAAC,GACpF,CAAC,OAAO,EAAEA,KAAK,qBAAqB,CAAC,GACrC,OACA,qBACA,2EACA,6EACA,qEACA,wEACA,8DACA,CAAC,yDAAyD,CAAC,GAC3D,sFACA,qEACA,mFACA,8EACA,yFACA,wFACA,OACA,CAAC,gEAAgE,CAAC,GAClE,8EACA,yEACA,8BACA,OACA,gBACA,CAAC,EAAE,EAAEA,KAAK,gFAAgF,CAAC,GAC3F,CAAC,EAAE,EAAEA,KAAK,mEAAmE,CAAC,GAC9E,CAAC,EAAE,EAAEA,KAAK,sEAAsE,CAAC,GACjF,CAAC,EAAE,EAAEA,KAAK,+EAA+E,CAAC,GAC1F,CAAC,EAAE,EAAEA,KAAK,oEAAoE,CAAC;AAEnF;AAEA,2GAA2G;AAC3G,eAAe,eAAeC,IAAIC,IAAc,EAAEF,IAAY;IAC5D,IAAIE,IAAI,CAAC,EAAE,KAAK,eAAeA,IAAI,CAAC,EAAE,KAAK,MAAM;QAC/CC,QAAQC,GAAG,CAACZ;QACZ;IACF;IACA,IAAIU,KAAKG,MAAM,KAAK,KAAKH,IAAI,CAAC,EAAE,KAAK,YAAYA,IAAI,CAAC,EAAE,KAAK,MAAM;QACjE,wEAAwE;QACxE,qBAAqB;QACrB,IAAIA,KAAKG,MAAM,KAAK,GAAG;YACrBF,QAAQG,KAAK,CAACP,MAAMC;YACpBO,QAAQC,IAAI,CAACjB;QACf;QACAY,QAAQC,GAAG,CAAC,GAAGJ,KAAK,EAAE,EAAER,kBAAkB;QAC1CW,QAAQC,GAAG,CAAC;QACZD,QAAQC,GAAG,CAACL,MAAMC;QAClB;IACF;IAEA,MAAMS,MAAW;QACfT;QACAU,MAAMR,KAAKS,KAAK,CAAC;QACjBC,YAAWC,OAAO;YAChBV,QAAQG,KAAK,CAACO;YACdN,QAAQC,IAAI,CAACjB;QACf;QACAuB,WAAWvB;IACb;IAEA,IAAI;QACF,MAAMwB,OAAO7B,QAAQ,CAACgB,IAAI,CAAC,EAAE,CAAC;QAC9B,IAAIa,MAAM;YACR,MAAM,AAAC,CAAA,MAAMA,MAAK,EAAGC,OAAO,CAACP;QAC/B,OAAO;YACL,sEAAsE;YACtE,sDAAsD;YACtDA,IAAIC,IAAI,GAAGR;YACX,MAAM,AAAC,CAAA,MAAMhB,SAAS+B,OAAO,EAAC,EAAGD,OAAO,CAACP;QAC3C;IACF,EAAE,OAAOS,KAAK;QACZf,QAAQG,KAAK,CAAC,AAACY,IAAcL,OAAO;QACpCN,QAAQC,IAAI,CAACjB;IACf;AACF"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/collect.ts"],"sourcesContent":["// Pass 1: load each PDF text run into viewport points, plus the document-wide\n// char-weighted font-height histogram. Height rounds to 0.1pt for stable histogram keys.\nimport type { PDFPageProxy } from 'pdfjs-dist/legacy/build/pdf.mjs';\nimport { OPS, Util } from 'pdfjs-dist/legacy/build/pdf.mjs';\nimport { openDocument } from './pdf-open.ts';\nimport type { CollectedPage, CollectResult, Glyph } from './types.ts';\n\n// HYBRID-PAGE DETECTION (README's \"Design contract\"): a page can carry a\n// real text layer AND a separately-painted large image with its OWN baked-in text that `page.getTextContent()` never sees.\n//\n// Image presence uses the paint operators for XObjects, inline images, and masks.\n// The large-image threshold uses `paintImageXObject` dimensions when available.\nconst LARGE_IMAGE_MIN_PX = 200;\ninterface PageImageSignals {\n hasImage: boolean;\n largeImage: boolean;\n}\n\nconst IMAGE_PAINT_OPS = new Set([OPS.paintImageMaskXObject, OPS.paintImageMaskXObjectGroup, OPS.paintImageMaskXObjectRepeat, OPS.paintImageXObject, OPS.paintImageXObjectRepeat, OPS.paintInlineImageXObject, OPS.paintInlineImageXObjectGroup, OPS.paintSolidColorImageMask]);\n\nasync function getPageImageSignals(page: PDFPageProxy, textChars: number): Promise<PageImageSignals> {\n const { fnArray, argsArray } = await page.getOperatorList();\n let hasImage = false;\n let largeImage = false;\n for (let i = 0; i < fnArray.length; i++) {\n if (IMAGE_PAINT_OPS.has(fnArray[i])) hasImage = true;\n if (fnArray[i] !== OPS.paintImageXObject) continue;\n const [, w, h] = argsArray[i];\n if (typeof w === 'number' && typeof h === 'number') {\n if (w >= LARGE_IMAGE_MIN_PX && h >= LARGE_IMAGE_MIN_PX) largeImage = true;\n } else largeImage = true; // dims unavailable: coarser \"op present\" signal\n }\n return { hasImage, largeImage: textChars > 80 && largeImage };\n}\n\n// { first, last }: 1-based, inclusive page range (the CLI's --pages).\n// Out-of-range values CLAMP to the document's bounds; `page` on each returned entry is always the SOURCE page number, so markers/splicing stay correct.\n//\n// CAVEAT: `heightChars` is built ONLY from the pages actually collected, so\n// a `--pages` subset's heading levels can differ from a full run's — accepted since re-extracted pages get spliced back in for their prose, not diffed.\n\n// Cheap page-count-only load (no per-page text extraction) — the OCR path\n// needs it to resolve its own [first, last] clamping and print a \"### pN\" marker for every page in range, including ones with no recognized words.\nexport async function pageCount(src: string): Promise<number> {\n const doc = await openDocument(src);\n return doc.numPages;\n}\n\nexport async function collect(src: string, { first, last }: { first?: number; last?: number } = {}): Promise<CollectResult> {\n const doc = await openDocument(src);\n const from = Math.min(Math.max(1, first ?? 1), Math.max(1, doc.numPages));\n const to = Math.min(Math.max(from, last ?? doc.numPages), doc.numPages);\n const pages: CollectedPage[] = [];\n const heightChars = new Map<number, number>();\n for (let p = from; p <= to; p++) {\n const page = await doc.getPage(p);\n const { items } = await page.getTextContent();\n const vp = page.getViewport({ scale: 1 });\n const viewportScale = Math.hypot(vp.transform[0], vp.transform[1]);\n const glyphs: Glyph[] = [];\n let textChars = 0;\n for (const it of items) {\n if (!('str' in it) || typeof it.str !== 'string') continue;\n textChars += it.str.length;\n if (!it.str.trim()) continue;\n const transform = Util.transform(vp.transform, it.transform);\n const baselineLength = Math.hypot(transform[0], transform[1]);\n const w = Math.abs(it.width) * viewportScale;\n const h = Math.round(Math.hypot(transform[2], transform[3]) * 10) / 10;\n const baselineX = baselineLength > 0 ? (transform[0] / baselineLength) * w : w;\n const baselineY = baselineLength > 0 ? (transform[1] / baselineLength) * w : 0;\n const corners = [\n [transform[4], transform[5]],\n [transform[4] + baselineX, transform[5] + baselineY],\n [transform[4] + baselineX + transform[2], transform[5] + baselineY + transform[3]],\n [transform[4] + transform[2], transform[5] + transform[3]],\n ];\n const boundsX = corners.map(([x]) => x);\n const boundsY = corners.map(([, y]) => y);\n glyphs.push({\n s: it.str,\n x: transform[4],\n y: transform[5],\n w,\n h,\n bounds: {\n x: Math.min(...boundsX),\n y: Math.min(...boundsY),\n w: Math.max(...boundsX) - Math.min(...boundsX),\n h: Math.max(...boundsY) - Math.min(...boundsY),\n },\n });\n heightChars.set(h, (heightChars.get(h) || 0) + it.str.length);\n }\n // \"text layer used\" for hybrid-page purposes (page text > 80 chars) — a\n // page with only a couple of stray scraps isn't \"using\" the text layer even if it also carries a large image.\n const { hasImage, largeImage } = await getPageImageSignals(page, textChars);\n pages.push({\n page: p,\n glyphs,\n width: vp.width,\n height: vp.height,\n textChars,\n hasImage,\n largeImage,\n });\n }\n // `numPages` is the SELECTED count, not doc.numPages: the auto-OCR\n // chars/page floor must judge a `--pages` subset against its own page count, or a genuinely image-only page reads as merely \"sparse\".\n return { numPages: pages.length, pages, heightChars };\n}\n"],"names":["OPS","Util","openDocument","LARGE_IMAGE_MIN_PX","IMAGE_PAINT_OPS","Set","paintImageMaskXObject","paintImageMaskXObjectGroup","paintImageMaskXObjectRepeat","paintImageXObject","paintImageXObjectRepeat","paintInlineImageXObject","paintInlineImageXObjectGroup","paintSolidColorImageMask","getPageImageSignals","page","textChars","fnArray","argsArray","getOperatorList","hasImage","largeImage","i","length","has","w","h","pageCount","src","doc","numPages","collect","first","last","from","Math","min","max","to","pages","heightChars","Map","p","getPage","items","getTextContent","vp","getViewport","scale","viewportScale","hypot","transform","glyphs","it","str","trim","baselineLength","abs","width","round","baselineX","baselineY","corners","boundsX","map","x","boundsY","y","push","s","bounds","set","get","height"],"mappings":"AAAA,8EAA8E;AAC9E,yFAAyF;AAEzF,SAASA,GAAG,EAAEC,IAAI,QAAQ,kCAAkC;AAC5D,SAASC,YAAY,QAAQ,gBAAgB;AAG7C,yEAAyE;AACzE,2HAA2H;AAC3H,EAAE;AACF,kFAAkF;AAClF,gFAAgF;AAChF,MAAMC,qBAAqB;AAM3B,MAAMC,kBAAkB,IAAIC,IAAI;IAACL,IAAIM,qBAAqB;IAAEN,IAAIO,0BAA0B;IAAEP,IAAIQ,2BAA2B;IAAER,IAAIS,iBAAiB;IAAET,IAAIU,uBAAuB;IAAEV,IAAIW,uBAAuB;IAAEX,IAAIY,4BAA4B;IAAEZ,IAAIa,wBAAwB;CAAC;AAE7Q,eAAeC,oBAAoBC,IAAkB,EAAEC,SAAiB;IACtE,MAAM,EAAEC,OAAO,EAAEC,SAAS,EAAE,GAAG,MAAMH,KAAKI,eAAe;IACzD,IAAIC,WAAW;IACf,IAAIC,aAAa;IACjB,IAAK,IAAIC,IAAI,GAAGA,IAAIL,QAAQM,MAAM,EAAED,IAAK;QACvC,IAAIlB,gBAAgBoB,GAAG,CAACP,OAAO,CAACK,EAAE,GAAGF,WAAW;QAChD,IAAIH,OAAO,CAACK,EAAE,KAAKtB,IAAIS,iBAAiB,EAAE;QAC1C,MAAM,GAAGgB,GAAGC,EAAE,GAAGR,SAAS,CAACI,EAAE;QAC7B,IAAI,OAAOG,MAAM,YAAY,OAAOC,MAAM,UAAU;YAClD,IAAID,KAAKtB,sBAAsBuB,KAAKvB,oBAAoBkB,aAAa;QACvE,OAAOA,aAAa,MAAM,gDAAgD;IAC5E;IACA,OAAO;QAAED;QAAUC,YAAYL,YAAY,MAAMK;IAAW;AAC9D;AAEA,sEAAsE;AACtE,wJAAwJ;AACxJ,EAAE;AACF,4EAA4E;AAC5E,wJAAwJ;AAExJ,0EAA0E;AAC1E,mJAAmJ;AACnJ,OAAO,eAAeM,UAAUC,GAAW;IACzC,MAAMC,MAAM,MAAM3B,aAAa0B;IAC/B,OAAOC,IAAIC,QAAQ;AACrB;AAEA,OAAO,eAAeC,QAAQH,GAAW,EAAE,EAAEI,KAAK,EAAEC,IAAI,EAAqC,GAAG,CAAC,CAAC;IAChG,MAAMJ,MAAM,MAAM3B,aAAa0B;IAC/B,MAAMM,OAAOC,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,GAAGL,kBAAAA,mBAAAA,QAAS,IAAIG,KAAKE,GAAG,CAAC,GAAGR,IAAIC,QAAQ;IACvE,MAAMQ,KAAKH,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAACH,MAAMD,iBAAAA,kBAAAA,OAAQJ,IAAIC,QAAQ,GAAGD,IAAIC,QAAQ;IACtE,MAAMS,QAAyB,EAAE;IACjC,MAAMC,cAAc,IAAIC;IACxB,IAAK,IAAIC,IAAIR,MAAMQ,KAAKJ,IAAII,IAAK;QAC/B,MAAM3B,OAAO,MAAMc,IAAIc,OAAO,CAACD;QAC/B,MAAM,EAAEE,KAAK,EAAE,GAAG,MAAM7B,KAAK8B,cAAc;QAC3C,MAAMC,KAAK/B,KAAKgC,WAAW,CAAC;YAAEC,OAAO;QAAE;QACvC,MAAMC,gBAAgBd,KAAKe,KAAK,CAACJ,GAAGK,SAAS,CAAC,EAAE,EAAEL,GAAGK,SAAS,CAAC,EAAE;QACjE,MAAMC,SAAkB,EAAE;QAC1B,IAAIpC,YAAY;QAChB,KAAK,MAAMqC,MAAMT,MAAO;YACtB,IAAI,CAAE,CAAA,SAASS,EAAC,KAAM,OAAOA,GAAGC,GAAG,KAAK,UAAU;YAClDtC,aAAaqC,GAAGC,GAAG,CAAC/B,MAAM;YAC1B,IAAI,CAAC8B,GAAGC,GAAG,CAACC,IAAI,IAAI;YACpB,MAAMJ,YAAYlD,KAAKkD,SAAS,CAACL,GAAGK,SAAS,EAAEE,GAAGF,SAAS;YAC3D,MAAMK,iBAAiBrB,KAAKe,KAAK,CAACC,SAAS,CAAC,EAAE,EAAEA,SAAS,CAAC,EAAE;YAC5D,MAAM1B,IAAIU,KAAKsB,GAAG,CAACJ,GAAGK,KAAK,IAAIT;YAC/B,MAAMvB,IAAIS,KAAKwB,KAAK,CAACxB,KAAKe,KAAK,CAACC,SAAS,CAAC,EAAE,EAAEA,SAAS,CAAC,EAAE,IAAI,MAAM;YACpE,MAAMS,YAAYJ,iBAAiB,IAAI,AAACL,SAAS,CAAC,EAAE,GAAGK,iBAAkB/B,IAAIA;YAC7E,MAAMoC,YAAYL,iBAAiB,IAAI,AAACL,SAAS,CAAC,EAAE,GAAGK,iBAAkB/B,IAAI;YAC7E,MAAMqC,UAAU;gBACd;oBAACX,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE;iBAAC;gBAC5B;oBAACA,SAAS,CAAC,EAAE,GAAGS;oBAAWT,SAAS,CAAC,EAAE,GAAGU;iBAAU;gBACpD;oBAACV,SAAS,CAAC,EAAE,GAAGS,YAAYT,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE,GAAGU,YAAYV,SAAS,CAAC,EAAE;iBAAC;gBAClF;oBAACA,SAAS,CAAC,EAAE,GAAGA,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE,GAAGA,SAAS,CAAC,EAAE;iBAAC;aAC3D;YACD,MAAMY,UAAUD,QAAQE,GAAG,CAAC,CAAC,CAACC,EAAE,GAAKA;YACrC,MAAMC,UAAUJ,QAAQE,GAAG,CAAC,CAAC,GAAGG,EAAE,GAAKA;YACvCf,OAAOgB,IAAI,CAAC;gBACVC,GAAGhB,GAAGC,GAAG;gBACTW,GAAGd,SAAS,CAAC,EAAE;gBACfgB,GAAGhB,SAAS,CAAC,EAAE;gBACf1B;gBACAC;gBACA4C,QAAQ;oBACNL,GAAG9B,KAAKC,GAAG,IAAI2B;oBACfI,GAAGhC,KAAKC,GAAG,IAAI8B;oBACfzC,GAAGU,KAAKE,GAAG,IAAI0B,WAAW5B,KAAKC,GAAG,IAAI2B;oBACtCrC,GAAGS,KAAKE,GAAG,IAAI6B,WAAW/B,KAAKC,GAAG,IAAI8B;gBACxC;YACF;YACA1B,YAAY+B,GAAG,CAAC7C,GAAG,AAACc,CAAAA,YAAYgC,GAAG,CAAC9C,MAAM,CAAA,IAAK2B,GAAGC,GAAG,CAAC/B,MAAM;QAC9D;QACA,wEAAwE;QACxE,8GAA8G;QAC9G,MAAM,EAAEH,QAAQ,EAAEC,UAAU,EAAE,GAAG,MAAMP,oBAAoBC,MAAMC;QACjEuB,MAAM6B,IAAI,CAAC;YACTrD,MAAM2B;YACNU;YACAM,OAAOZ,GAAGY,KAAK;YACfe,QAAQ3B,GAAG2B,MAAM;YACjBzD;YACAI;YACAC;QACF;IACF;IACA,mEAAmE;IACnE,sIAAsI;IACtI,OAAO;QAAES,UAAUS,MAAMhB,MAAM;QAAEgB;QAAOC;IAAY;AACtD"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/collect.ts"],"sourcesContent":["// Pass 1: load each PDF text run into viewport points, plus the document-wide\n// char-weighted font-height histogram. Height rounds to 0.1pt for stable histogram keys.\nimport type { PDFPageProxy } from 'pdfjs-dist/legacy/build/pdf.mjs';\nimport { OPS, Util } from 'pdfjs-dist/legacy/build/pdf.mjs';\nimport { openDocument } from './pdf-open.ts';\nimport type { CollectedPage, CollectResult, Glyph } from './types.ts';\n\n// HYBRID-PAGE DETECTION (README's \"Design contract\"): a page can carry a\n// real text layer AND a separately-painted large image with its OWN baked-in text that `page.getTextContent()` never sees.\n//\n// Image presence uses the paint operators for XObjects, inline images, and masks.\n// The large-image threshold uses `paintImageXObject` dimensions when available.\nconst LARGE_IMAGE_MIN_PX = 200;\ninterface PageImageSignals {\n hasImage: boolean;\n largeImage: boolean;\n}\n\nconst IMAGE_PAINT_OPS = new Set([OPS.paintImageMaskXObject, OPS.paintImageMaskXObjectGroup, OPS.paintImageMaskXObjectRepeat, OPS.paintImageXObject, OPS.paintImageXObjectRepeat, OPS.paintInlineImageXObject, OPS.paintInlineImageXObjectGroup, OPS.paintSolidColorImageMask]);\n\nasync function getPageImageSignals(page: PDFPageProxy, textChars: number): Promise<PageImageSignals> {\n const { fnArray, argsArray } = await page.getOperatorList();\n let hasImage = false;\n let largeImage = false;\n for (let i = 0; i < fnArray.length; i++) {\n if (IMAGE_PAINT_OPS.has(fnArray[i])) hasImage = true;\n if (fnArray[i] !== OPS.paintImageXObject) continue;\n const [, w, h] = argsArray[i];\n if (typeof w === 'number' && typeof h === 'number') {\n if (w >= LARGE_IMAGE_MIN_PX && h >= LARGE_IMAGE_MIN_PX) largeImage = true;\n } else largeImage = true; // dims unavailable: coarser \"op present\" signal\n }\n return { hasImage, largeImage: textChars > 80 && largeImage };\n}\n\n// { first, last }: 1-based, inclusive page range (the CLI's --pages).\n// Out-of-range values CLAMP to the document's bounds; `page` on each returned entry is always the SOURCE page number, so markers/splicing stay correct.\n//\n// CAVEAT: `heightChars` is built ONLY from the pages actually collected, so\n// a `--pages` subset's heading levels can differ from a full run's — accepted since re-extracted pages get spliced back in for their prose, not diffed.\n\n// Cheap page-count-only load (no per-page text extraction) — the OCR path\n// needs it to resolve its own [first, last] clamping and print a \"### pN\" marker for every page in range, including ones with no recognized words.\nexport async function pageCount(src: string): Promise<number> {\n const doc = await openDocument(src);\n return doc.numPages;\n}\n\nexport async function collect(src: string, { first, last }: { first?: number; last?: number } = {}): Promise<CollectResult> {\n const doc = await openDocument(src);\n const from = Math.min(Math.max(1, first ?? 1), Math.max(1, doc.numPages));\n const to = Math.min(Math.max(from, last ?? doc.numPages), doc.numPages);\n const pages: CollectedPage[] = [];\n const heightChars = new Map<number, number>();\n for (let p = from; p <= to; p++) {\n const page = await doc.getPage(p);\n const { items } = await page.getTextContent();\n const vp = page.getViewport({ scale: 1 });\n const viewportScale = Math.hypot(vp.transform[0], vp.transform[1]);\n const glyphs: Glyph[] = [];\n let textChars = 0;\n for (const it of items) {\n if (!('str' in it) || typeof it.str !== 'string') continue;\n textChars += it.str.length;\n if (!it.str.trim()) continue;\n const transform = Util.transform(vp.transform, it.transform);\n const baselineLength = Math.hypot(transform[0], transform[1]);\n const w = Math.abs(it.width) * viewportScale;\n const h = Math.round(Math.hypot(transform[2], transform[3]) * 10) / 10;\n const baselineX = baselineLength > 0 ? (transform[0] / baselineLength) * w : w;\n const baselineY = baselineLength > 0 ? (transform[1] / baselineLength) * w : 0;\n const corners = [\n [transform[4], transform[5]],\n [transform[4] + baselineX, transform[5] + baselineY],\n [transform[4] + baselineX + transform[2], transform[5] + baselineY + transform[3]],\n [transform[4] + transform[2], transform[5] + transform[3]],\n ];\n const boundsX = corners.map(([x]) => x);\n const boundsY = corners.map(([, y]) => y);\n glyphs.push({\n s: it.str,\n x: transform[4],\n y: transform[5],\n w,\n h,\n bounds: {\n x: Math.min(...boundsX),\n y: Math.min(...boundsY),\n w: Math.max(...boundsX) - Math.min(...boundsX),\n h: Math.max(...boundsY) - Math.min(...boundsY),\n },\n });\n heightChars.set(h, (heightChars.get(h) || 0) + it.str.length);\n }\n // \"text layer used\" for hybrid-page purposes (page text > 80 chars) — a\n // page with only a couple of stray scraps isn't \"using\" the text layer even if it also carries a large image.\n const { hasImage, largeImage } = await getPageImageSignals(page, textChars);\n pages.push({\n page: p,\n glyphs,\n width: vp.width,\n height: vp.height,\n textChars,\n hasImage,\n largeImage,\n });\n }\n // `numPages` is the SELECTED count, not doc.numPages: the auto-OCR\n // chars/page floor must judge a `--pages` subset against its own page count, or a genuinely image-only page reads as merely \"sparse\".\n return { numPages: pages.length, pages, heightChars };\n}\n"],"names":["OPS","Util","openDocument","LARGE_IMAGE_MIN_PX","IMAGE_PAINT_OPS","Set","paintImageMaskXObject","paintImageMaskXObjectGroup","paintImageMaskXObjectRepeat","paintImageXObject","paintImageXObjectRepeat","paintInlineImageXObject","paintInlineImageXObjectGroup","paintSolidColorImageMask","getPageImageSignals","page","textChars","fnArray","argsArray","getOperatorList","hasImage","largeImage","i","length","has","w","h","pageCount","src","doc","numPages","collect","first","last","from","Math","min","max","to","pages","heightChars","Map","p","getPage","items","getTextContent","vp","getViewport","scale","viewportScale","hypot","transform","glyphs","it","str","trim","baselineLength","abs","width","round","baselineX","baselineY","corners","boundsX","map","x","boundsY","y","push","s","bounds","set","get","height"],"mappings":"AAAA,8EAA8E;AAC9E,yFAAyF;AAEzF,SAASA,GAAG,EAAEC,IAAI,QAAQ,kCAAkC;AAC5D,SAASC,YAAY,QAAQ,gBAAgB;AAG7C,yEAAyE;AACzE,2HAA2H;AAC3H,EAAE;AACF,kFAAkF;AAClF,gFAAgF;AAChF,MAAMC,qBAAqB;AAM3B,MAAMC,kBAAkB,IAAIC,IAAI;IAACL,IAAIM,qBAAqB;IAAEN,IAAIO,0BAA0B;IAAEP,IAAIQ,2BAA2B;IAAER,IAAIS,iBAAiB;IAAET,IAAIU,uBAAuB;IAAEV,IAAIW,uBAAuB;IAAEX,IAAIY,4BAA4B;IAAEZ,IAAIa,wBAAwB;CAAC;AAE7Q,eAAeC,oBAAoBC,IAAkB,EAAEC,SAAiB;IACtE,MAAM,EAAEC,OAAO,EAAEC,SAAS,EAAE,GAAG,MAAMH,KAAKI,eAAe;IACzD,IAAIC,WAAW;IACf,IAAIC,aAAa;IACjB,IAAK,IAAIC,IAAI,GAAGA,IAAIL,QAAQM,MAAM,EAAED,IAAK;QACvC,IAAIlB,gBAAgBoB,GAAG,CAACP,OAAO,CAACK,EAAE,GAAGF,WAAW;QAChD,IAAIH,OAAO,CAACK,EAAE,KAAKtB,IAAIS,iBAAiB,EAAE;QAC1C,MAAM,GAAGgB,GAAGC,EAAE,GAAGR,SAAS,CAACI,EAAE;QAC7B,IAAI,OAAOG,MAAM,YAAY,OAAOC,MAAM,UAAU;YAClD,IAAID,KAAKtB,sBAAsBuB,KAAKvB,oBAAoBkB,aAAa;QACvE,OAAOA,aAAa,MAAM,gDAAgD;IAC5E;IACA,OAAO;QAAED;QAAUC,YAAYL,YAAY,MAAMK;IAAW;AAC9D;AAEA,sEAAsE;AACtE,wJAAwJ;AACxJ,EAAE;AACF,4EAA4E;AAC5E,wJAAwJ;AAExJ,0EAA0E;AAC1E,mJAAmJ;AACnJ,OAAO,eAAeM,UAAUC,GAAW;IACzC,MAAMC,MAAM,MAAM3B,aAAa0B;IAC/B,OAAOC,IAAIC,QAAQ;AACrB;AAEA,OAAO,eAAeC,QAAQH,GAAW,EAAE,EAAEI,KAAK,EAAEC,IAAI,EAAqC,GAAG,CAAC,CAAC;IAChG,MAAMJ,MAAM,MAAM3B,aAAa0B;IAC/B,MAAMM,OAAOC,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,GAAGL,kBAAAA,mBAAAA,QAAS,IAAIG,KAAKE,GAAG,CAAC,GAAGR,IAAIC,QAAQ;IACvE,MAAMQ,KAAKH,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAACH,MAAMD,iBAAAA,kBAAAA,OAAQJ,IAAIC,QAAQ,GAAGD,IAAIC,QAAQ;IACtE,MAAMS,QAAyB,EAAE;IACjC,MAAMC,cAAc,IAAIC;IACxB,IAAK,IAAIC,IAAIR,MAAMQ,KAAKJ,IAAII,IAAK;QAC/B,MAAM3B,OAAO,MAAMc,IAAIc,OAAO,CAACD;QAC/B,MAAM,EAAEE,KAAK,EAAE,GAAG,MAAM7B,KAAK8B,cAAc;QAC3C,MAAMC,KAAK/B,KAAKgC,WAAW,CAAC;YAAEC,OAAO;QAAE;QACvC,MAAMC,gBAAgBd,KAAKe,KAAK,CAACJ,GAAGK,SAAS,CAAC,EAAE,EAAEL,GAAGK,SAAS,CAAC,EAAE;QACjE,MAAMC,SAAkB,EAAE;QAC1B,IAAIpC,YAAY;QAChB,KAAK,MAAMqC,MAAMT,MAAO;YACtB,IAAI,CAAE,CAAA,SAASS,EAAC,KAAM,OAAOA,GAAGC,GAAG,KAAK,UAAU;YAClDtC,aAAaqC,GAAGC,GAAG,CAAC/B,MAAM;YAC1B,IAAI,CAAC8B,GAAGC,GAAG,CAACC,IAAI,IAAI;YACpB,MAAMJ,YAAYlD,KAAKkD,SAAS,CAACL,GAAGK,SAAS,EAAEE,GAAGF,SAAS;YAC3D,MAAMK,iBAAiBrB,KAAKe,KAAK,CAACC,SAAS,CAAC,EAAE,EAAEA,SAAS,CAAC,EAAE;YAC5D,MAAM1B,IAAIU,KAAKsB,GAAG,CAACJ,GAAGK,KAAK,IAAIT;YAC/B,MAAMvB,IAAIS,KAAKwB,KAAK,CAACxB,KAAKe,KAAK,CAACC,SAAS,CAAC,EAAE,EAAEA,SAAS,CAAC,EAAE,IAAI,MAAM;YACpE,MAAMS,YAAYJ,iBAAiB,IAAI,AAACL,SAAS,CAAC,EAAE,GAAGK,iBAAkB/B,IAAIA;YAC7E,MAAMoC,YAAYL,iBAAiB,IAAI,AAACL,SAAS,CAAC,EAAE,GAAGK,iBAAkB/B,IAAI;YAC7E,MAAMqC,UAAU;gBACd;oBAACX,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE;iBAAC;gBAC5B;oBAACA,SAAS,CAAC,EAAE,GAAGS;oBAAWT,SAAS,CAAC,EAAE,GAAGU;iBAAU;gBACpD;oBAACV,SAAS,CAAC,EAAE,GAAGS,YAAYT,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE,GAAGU,YAAYV,SAAS,CAAC,EAAE;iBAAC;gBAClF;oBAACA,SAAS,CAAC,EAAE,GAAGA,SAAS,CAAC,EAAE;oBAAEA,SAAS,CAAC,EAAE,GAAGA,SAAS,CAAC,EAAE;iBAAC;aAC3D;YACD,MAAMY,UAAUD,QAAQE,GAAG,CAAC,CAAC,CAACC,EAAE,GAAKA;YACrC,MAAMC,UAAUJ,QAAQE,GAAG,CAAC,CAAC,GAAGG,EAAE,GAAKA;YACvCf,OAAOgB,IAAI,CAAC;gBACVC,GAAGhB,GAAGC,GAAG;gBACTW,GAAGd,SAAS,CAAC,EAAE;gBACfgB,GAAGhB,SAAS,CAAC,EAAE;gBACf1B;gBACAC;gBACA4C,QAAQ;oBACNL,GAAG9B,KAAKC,GAAG,IAAI2B;oBACfI,GAAGhC,KAAKC,GAAG,IAAI8B;oBACfzC,GAAGU,KAAKE,GAAG,IAAI0B,WAAW5B,KAAKC,GAAG,IAAI2B;oBACtCrC,GAAGS,KAAKE,GAAG,IAAI6B,WAAW/B,KAAKC,GAAG,IAAI8B;gBACxC;YACF;YACA1B,YAAY+B,GAAG,CAAC7C,GAAG,AAACc,CAAAA,YAAYgC,GAAG,CAAC9C,MAAM,CAAA,IAAK2B,GAAGC,GAAG,CAAC/B,MAAM;QAC9D;QACA,wEAAwE;QACxE,8GAA8G;QAC9G,MAAM,EAAEH,QAAQ,EAAEC,UAAU,EAAE,GAAG,MAAMP,oBAAoBC,MAAMC;QACjEuB,MAAM6B,IAAI,CAAC;YACTrD,MAAM2B;YACNU;YACAM,OAAOZ,GAAGY,KAAK;YACfe,QAAQ3B,GAAG2B,MAAM;YACjBzD;YACAI;YACAC;QACF;IACF;IACA,mEAAmE;IACnE,sIAAsI;IACtI,OAAO;QAAES,UAAUS,MAAMhB,MAAM;QAAEgB;QAAOC;IAAY;AACtD"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/dict.ts"],"sourcesContent":["// The hyphen-rejoin dictionary, shared by src/emit.ts and src/markdown.ts.\n// Loaded once, lazily, on first use — never at import — so importing the library does no file IO on its own.\nimport { existsSync, readFileSync } from 'node:fs';\nimport wordListPath from 'word-list';\n\n// Bundled list (not the OS's /usr/share/dict/words) → identical rejoin decisions on every\n// OS/machine. PDF_TO_MD_DICT_PATH overrides it — the test seam for simulating no dictionary.\nexport const DICT_PATH = process.env.PDF_TO_MD_DICT_PATH || wordListPath;\n\nlet cache: { words: Set<string>; available: boolean } | null = null;\n\nfunction load(): { words: Set<string>; available: boolean } {\n if (!cache) {\n const available = existsSync(DICT_PATH);\n cache = { words: available ? new Set(readFileSync(DICT_PATH, 'utf8').toLowerCase().split('\\n')) : new Set(), available };\n }\n return cache;\n}\n\n/** The hyphen-rejoin dictionary; an empty set if none was found. */\nexport function getDict(): Set<string> {\n return load().words;\n}\n\n/** False when no dictionary file exists at DICT_PATH — every line-final\n * hyphen then stays unresolved. */\nexport function dictAvailable(): boolean {\n return load().available;\n}\n\n/** The notice callers surface via warn() when dictAvailable() is false. */\nexport function dictUnavailableWarning(): string {\n return `no dictionary at ${DICT_PATH} — line-final hyphens kept unresolved`;\n}\n"],"names":["existsSync","readFileSync","wordListPath","DICT_PATH","process","env","PDF_TO_MD_DICT_PATH","cache","load","available","words","Set","toLowerCase","split","getDict","dictAvailable","dictUnavailableWarning"],"mappings":"AAAA,2EAA2E;AAC3E,6GAA6G;AAC7G,SAASA,UAAU,EAAEC,YAAY,QAAQ,UAAU;AACnD,OAAOC,kBAAkB,YAAY;AAErC,0FAA0F;AAC1F,6FAA6F;AAC7F,OAAO,MAAMC,YAAYC,QAAQC,GAAG,CAACC,mBAAmB,IAAIJ,aAAa;AAEzE,IAAIK,QAA2D;AAE/D,SAASC;IACP,IAAI,CAACD,OAAO;QACV,MAAME,YAAYT,WAAWG;QAC7BI,QAAQ;YAAEG,OAAOD,YAAY,IAAIE,IAAIV,aAAaE,WAAW,QAAQS,WAAW,GAAGC,KAAK,CAAC,SAAS,IAAIF;YAAOF;QAAU;IACzH;IACA,OAAOF;AACT;AAEA,kEAAkE,GAClE,OAAO,SAASO;IACd,OAAON,OAAOE,KAAK;AACrB;AAEA;iCACiC,GACjC,OAAO,SAASK;IACd,OAAOP,OAAOC,SAAS;AACzB;AAEA,yEAAyE,GACzE,OAAO,SAASO;IACd,OAAO,CAAC,iBAAiB,EAAEb,UAAU,qCAAqC,CAAC;AAC7E"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/dict.ts"],"sourcesContent":["// The hyphen-rejoin dictionary, shared by src/emit.ts and src/markdown.ts.\n// Loaded once, lazily, on first use — never at import — so importing the library does no file IO on its own.\nimport { existsSync, readFileSync } from 'node:fs';\nimport wordListPath from 'word-list';\n\n// Bundled list (not the OS's /usr/share/dict/words) → identical rejoin decisions on every\n// OS/machine. PDF_TO_MD_DICT_PATH overrides it — the test seam for simulating no dictionary.\nexport const DICT_PATH = process.env.PDF_TO_MD_DICT_PATH || wordListPath;\n\nlet cache: { words: Set<string>; available: boolean } | null = null;\n\nfunction load(): { words: Set<string>; available: boolean } {\n if (!cache) {\n const available = existsSync(DICT_PATH);\n cache = { words: available ? new Set(readFileSync(DICT_PATH, 'utf8').toLowerCase().split('\\n')) : new Set(), available };\n }\n return cache;\n}\n\n/** The hyphen-rejoin dictionary; an empty set if none was found. */\nexport function getDict(): Set<string> {\n return load().words;\n}\n\n/** False when no dictionary file exists at DICT_PATH — every line-final\n * hyphen then stays unresolved. */\nexport function dictAvailable(): boolean {\n return load().available;\n}\n\n/** The notice callers surface via warn() when dictAvailable() is false. */\nexport function dictUnavailableWarning(): string {\n return `no dictionary at ${DICT_PATH} — line-final hyphens kept unresolved`;\n}\n"],"names":["existsSync","readFileSync","wordListPath","DICT_PATH","process","env","PDF_TO_MD_DICT_PATH","cache","load","available","words","Set","toLowerCase","split","getDict","dictAvailable","dictUnavailableWarning"],"mappings":"AAAA,2EAA2E;AAC3E,6GAA6G;AAC7G,SAASA,UAAU,EAAEC,YAAY,QAAQ,UAAU;AACnD,OAAOC,kBAAkB,YAAY;AAErC,0FAA0F;AAC1F,6FAA6F;AAC7F,OAAO,MAAMC,YAAYC,QAAQC,GAAG,CAACC,mBAAmB,IAAIJ,aAAa;AAEzE,IAAIK,QAA2D;AAE/D,SAASC;IACP,IAAI,CAACD,OAAO;QACV,MAAME,YAAYT,WAAWG;QAC7BI,QAAQ;YAAEG,OAAOD,YAAY,IAAIE,IAAIV,aAAaE,WAAW,QAAQS,WAAW,GAAGC,KAAK,CAAC,SAAS,IAAIF;YAAOF;QAAU;IACzH;IACA,OAAOF;AACT;AAEA,kEAAkE,GAClE,OAAO,SAASO;IACd,OAAON,OAAOE,KAAK;AACrB;AAEA;iCACiC,GACjC,OAAO,SAASK;IACd,OAAOP,OAAOC,SAAS;AACzB;AAEA,yEAAyE,GACzE,OAAO,SAASO;IACd,OAAO,CAAC,iBAAiB,EAAEb,UAAU,qCAAqC,CAAC;AAC7E"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/emit.ts"],"sourcesContent":["// emit.ts — OCR-path pass 3: lines -> markdown blocks.\nimport { getDict } from './dict.ts';\nimport type { OcrLine, PageOCR, Tuning, Word } from './geometry.ts';\nimport { bucket } from './geometry.ts';\n\n// Join a wrapped line onto the open paragraph, closing a line-final hyphen\n// only when the dictionary confirms the joined form is a word.\nfunction hyphenJoin(acc: string, next: string): string {\n if (acc.endsWith('-')) {\n const withoutDash = acc.slice(0, -1);\n const pen = withoutDash.at(-1);\n if (pen && /[A-Za-z0-9]/.test(pen)) {\n const head = withoutDash.split(' ').pop() ?? '';\n const firstTok = next.split(' ')[0] ?? '';\n const filtered = [...firstTok].filter((c) => /[A-Za-z-]/.test(c)).join('');\n const tail = filtered.split('-')[0] ?? '';\n const close = getDict().has((head + tail).toLowerCase());\n return (close ? withoutDash : acc) + next;\n }\n }\n return `${acc} ${next}`;\n}\n\n// Blocks with too few word-like tokens to trust as prose (see\n// UNSTRUCTURED_WORDLIKE_FRACTION below) are labeled `> [review: reason] text` instead of guessed at.\nconst REVIEW_REASON = {\n unstructuredLabels: 'unstructured labels',\n} as const;\n\n/** Share of word-like tokens (3+ letter run) below which a block counts as\n * unstructured. Measured (`review-blocks.test.ts`): junk 0.000-0.100 vs prose 0.743-0.857, cut 0.4. */\nexport const UNSTRUCTURED_WORDLIKE_FRACTION = 0.4;\nfunction isUnstructured(text: string): boolean {\n const tokens = text.split(/\\s+/).filter(Boolean);\n if (tokens.length === 0) return false; // never mark an empty block\n const wordLike = tokens.filter((t) => /[A-Za-z]{3}/.test(t)).length;\n return wordLike / tokens.length < UNSTRUCTURED_WORDLIKE_FRACTION;\n}\n\nfunction reviewLine(reason: string, text: string): string {\n return `> [review: ${reason}] ${text}`;\n}\n\n// One classified block, carrying constituent words for analyze()'s\n// block->word linkage; discriminated on `kind` so `reason`/`level` are compile-time guaranteed present on their variants.\nexport type EmitBlock =\n | { kind: 'heading'; text: string; level: number; words: Word[] }\n | { kind: 'paragraph'; text: string; words: Word[] }\n | { kind: 'quote'; text: string; words: Word[] }\n | {\n kind: 'review';\n text: string;\n /** The parseable reason inside `> [review: ...]`, carried on the block\n * so more reasons can be added without changing the render format. */\n reason: string;\n words: Word[];\n };\n\n// Renders one EmitBlock to the exact markdown emitPage used to push\n// directly — shared so extract.ts and analyze.ts produce byte-identical text.\nexport function renderEmitBlock(b: EmitBlock): string {\n switch (b.kind) {\n case 'heading':\n return `${'#'.repeat(b.level)} ${b.text}`;\n case 'quote':\n return `> ${b.text}`;\n case 'review':\n return reviewLine(b.reason, b.text);\n default:\n return b.text;\n }\n}\n\nexport function emitPage(p: PageOCR, levelOf: Map<number, number>, tuning: Tuning): EmitBlock[] {\n const blocks: EmitBlock[] = [];\n let para: OcrLine[] = [];\n const flush = () => {\n if (!para.length) return;\n const texts = para.map((l) => l.text);\n const merged = texts.slice(1).reduce(hyphenJoin, texts[0]);\n const words = para.flatMap((l) => l.words);\n // Same review-marker rule as the heading path below: no word may be\n // dropped or reordered here, only how the block renders changes.\n blocks.push(isUnstructured(merged) ? { kind: 'review', text: merged, reason: REVIEW_REASON.unstructuredLabels, words } : { kind: 'paragraph', text: merged, words });\n para = [];\n };\n\n let i = 0;\n while (i < p.lines.length) {\n const line: OcrLine = p.lines[i];\n const prev = i > 0 ? p.lines[i - 1] : null;\n if (prev && (line.col !== prev.col || prev.midY - line.midY > p.pitch * tuning.paraGap)) {\n // Join across a column break (instead of flushing) only when the column\n // changed, the paragraph dangles (no terminal punct), and the new text starts lowercase; a miss stays visible via pageStats.danglingLong.\n const danglingEnd = para.length > 0 && !/[.!?:;\"'”’)\\]]$/.test(para[para.length - 1].text);\n const continuationStart = /^[a-z]/.test(line.text);\n const joinAcrossColumns = line.col !== prev.col && danglingEnd && continuationStart && !levelOf.has(bucket(line.h));\n if (!joinAcrossColumns) flush();\n }\n\n if (levelOf.has(bucket(line.h))) {\n // Gather the run of consecutive heading-size lines: a wrapped title is\n // one heading, a long run is a pull quote in a bigger face.\n flush();\n const run: OcrLine[] = [line];\n let maxB = bucket(line.h);\n while (i + 1 < p.lines.length && levelOf.has(bucket(p.lines[i + 1].h)) && p.lines[i + 1].col === p.lines[i].col && p.lines[i].midY - p.lines[i + 1].midY < Math.max(p.lines[i].h, p.lines[i + 1].h) * 2.2) {\n i++;\n run.push(p.lines[i]);\n maxB = Math.max(maxB, bucket(p.lines[i].h));\n }\n const runTexts = run.map((l) => l.text);\n const merged = runTexts.slice(1).reduce(hyphenJoin, runTexts[0]);\n const words = run.flatMap((l) => l.words);\n // Below UNSTRUCTURED_WORDLIKE_FRACTION = chart labels/stray numerals,\n // not a heading; checked FIRST so a long numeral run isn't asserted as a pull quote.\n const junk = isUnstructured(merged);\n if (junk) blocks.push({ kind: 'review', text: merged, reason: REVIEW_REASON.unstructuredLabels, words });\n else if (merged.length > 150) blocks.push({ kind: 'quote', text: merged, words });\n else blocks.push({ kind: 'heading', text: merged, level: levelOf.get(maxB) as number, words });\n } else {\n para.push(line);\n }\n i++;\n }\n flush();\n return blocks;\n}\n"],"names":["getDict","bucket","hyphenJoin","acc","next","endsWith","withoutDash","slice","pen","at","test","filtered","head","split","pop","firstTok","filter","c","join","tail","close","has","toLowerCase","REVIEW_REASON","unstructuredLabels","UNSTRUCTURED_WORDLIKE_FRACTION","isUnstructured","text","tokens","Boolean","length","wordLike","t","reviewLine","reason","renderEmitBlock","b","kind","repeat","level","emitPage","p","levelOf","tuning","blocks","para","flush","texts","map","l","merged","reduce","words","flatMap","push","i","lines","line","prev","col","midY","pitch","paraGap","danglingEnd","continuationStart","joinAcrossColumns","h","run","maxB","Math","max","runTexts","junk","get"],"mappings":"AAAA,uDAAuD;AACvD,SAASA,OAAO,QAAQ,YAAY;AAEpC,SAASC,MAAM,QAAQ,gBAAgB;AAEvC,2EAA2E;AAC3E,+DAA+D;AAC/D,SAASC,WAAWC,GAAW,EAAEC,IAAY;IAC3C,IAAID,IAAIE,QAAQ,CAAC,MAAM;QACrB,MAAMC,cAAcH,IAAII,KAAK,CAAC,GAAG,CAAC;QAClC,MAAMC,MAAMF,YAAYG,EAAE,CAAC,CAAC;QAC5B,IAAID,OAAO,cAAcE,IAAI,CAACF,MAAM;gBACrBF,wBACIF,cAEJO;YAHb,MAAMC,QAAON,yBAAAA,YAAYO,KAAK,CAAC,KAAKC,GAAG,gBAA1BR,oCAAAA,yBAAgC;YAC7C,MAAMS,YAAWX,eAAAA,KAAKS,KAAK,CAAC,IAAI,CAAC,EAAE,cAAlBT,0BAAAA,eAAsB;YACvC,MAAMO,WAAW;mBAAII;aAAS,CAACC,MAAM,CAAC,CAACC,IAAM,YAAYP,IAAI,CAACO,IAAIC,IAAI,CAAC;YACvE,MAAMC,QAAOR,mBAAAA,SAASE,KAAK,CAAC,IAAI,CAAC,EAAE,cAAtBF,8BAAAA,mBAA0B;YACvC,MAAMS,QAAQpB,UAAUqB,GAAG,CAAC,AAACT,CAAAA,OAAOO,IAAG,EAAGG,WAAW;YACrD,OAAO,AAACF,CAAAA,QAAQd,cAAcH,GAAE,IAAKC;QACvC;IACF;IACA,OAAO,GAAGD,IAAI,CAAC,EAAEC,MAAM;AACzB;AAEA,8DAA8D;AAC9D,qGAAqG;AACrG,MAAMmB,gBAAgB;IACpBC,oBAAoB;AACtB;AAEA;qGACqG,GACrG,OAAO,MAAMC,iCAAiC,IAAI;AAClD,SAASC,eAAeC,IAAY;IAClC,MAAMC,SAASD,KAAKd,KAAK,CAAC,OAAOG,MAAM,CAACa;IACxC,IAAID,OAAOE,MAAM,KAAK,GAAG,OAAO,OAAO,4BAA4B;IACnE,MAAMC,WAAWH,OAAOZ,MAAM,CAAC,CAACgB,IAAM,cAActB,IAAI,CAACsB,IAAIF,MAAM;IACnE,OAAOC,WAAWH,OAAOE,MAAM,GAAGL;AACpC;AAEA,SAASQ,WAAWC,MAAc,EAAEP,IAAY;IAC9C,OAAO,CAAC,WAAW,EAAEO,OAAO,EAAE,EAAEP,MAAM;AACxC;AAiBA,oEAAoE;AACpE,8EAA8E;AAC9E,OAAO,SAASQ,gBAAgBC,CAAY;IAC1C,OAAQA,EAAEC,IAAI;QACZ,KAAK;YACH,OAAO,GAAG,IAAIC,MAAM,CAACF,EAAEG,KAAK,EAAE,CAAC,EAAEH,EAAET,IAAI,EAAE;QAC3C,KAAK;YACH,OAAO,CAAC,EAAE,EAAES,EAAET,IAAI,EAAE;QACtB,KAAK;YACH,OAAOM,WAAWG,EAAEF,MAAM,EAAEE,EAAET,IAAI;QACpC;YACE,OAAOS,EAAET,IAAI;IACjB;AACF;AAEA,OAAO,SAASa,SAASC,CAAU,EAAEC,OAA4B,EAAEC,MAAc;IAC/E,MAAMC,SAAsB,EAAE;IAC9B,IAAIC,OAAkB,EAAE;IACxB,MAAMC,QAAQ;QACZ,IAAI,CAACD,KAAKf,MAAM,EAAE;QAClB,MAAMiB,QAAQF,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAEtB,IAAI;QACpC,MAAMuB,SAASH,MAAMxC,KAAK,CAAC,GAAG4C,MAAM,CAACjD,YAAY6C,KAAK,CAAC,EAAE;QACzD,MAAMK,QAAQP,KAAKQ,OAAO,CAAC,CAACJ,IAAMA,EAAEG,KAAK;QACzC,oEAAoE;QACpE,iEAAiE;QACjER,OAAOU,IAAI,CAAC5B,eAAewB,UAAU;YAAEb,MAAM;YAAUV,MAAMuB;YAAQhB,QAAQX,cAAcC,kBAAkB;YAAE4B;QAAM,IAAI;YAAEf,MAAM;YAAaV,MAAMuB;YAAQE;QAAM;QAClKP,OAAO,EAAE;IACX;IAEA,IAAIU,IAAI;IACR,MAAOA,IAAId,EAAEe,KAAK,CAAC1B,MAAM,CAAE;QACzB,MAAM2B,OAAgBhB,EAAEe,KAAK,CAACD,EAAE;QAChC,MAAMG,OAAOH,IAAI,IAAId,EAAEe,KAAK,CAACD,IAAI,EAAE,GAAG;QACtC,IAAIG,QAASD,CAAAA,KAAKE,GAAG,KAAKD,KAAKC,GAAG,IAAID,KAAKE,IAAI,GAAGH,KAAKG,IAAI,GAAGnB,EAAEoB,KAAK,GAAGlB,OAAOmB,OAAO,AAAD,GAAI;YACvF,wEAAwE;YACxE,0IAA0I;YAC1I,MAAMC,cAAclB,KAAKf,MAAM,GAAG,KAAK,CAAC,kBAAkBpB,IAAI,CAACmC,IAAI,CAACA,KAAKf,MAAM,GAAG,EAAE,CAACH,IAAI;YACzF,MAAMqC,oBAAoB,SAAStD,IAAI,CAAC+C,KAAK9B,IAAI;YACjD,MAAMsC,oBAAoBR,KAAKE,GAAG,KAAKD,KAAKC,GAAG,IAAII,eAAeC,qBAAqB,CAACtB,QAAQrB,GAAG,CAACpB,OAAOwD,KAAKS,CAAC;YACjH,IAAI,CAACD,mBAAmBnB;QAC1B;QAEA,IAAIJ,QAAQrB,GAAG,CAACpB,OAAOwD,KAAKS,CAAC,IAAI;YAC/B,uEAAuE;YACvE,4DAA4D;YAC5DpB;YACA,MAAMqB,MAAiB;gBAACV;aAAK;YAC7B,IAAIW,OAAOnE,OAAOwD,KAAKS,CAAC;YACxB,MAAOX,IAAI,IAAId,EAAEe,KAAK,CAAC1B,MAAM,IAAIY,QAAQrB,GAAG,CAACpB,OAAOwC,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACW,CAAC,MAAMzB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACI,GAAG,KAAKlB,EAAEe,KAAK,CAACD,EAAE,CAACI,GAAG,IAAIlB,EAAEe,KAAK,CAACD,EAAE,CAACK,IAAI,GAAGnB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACK,IAAI,GAAGS,KAAKC,GAAG,CAAC7B,EAAEe,KAAK,CAACD,EAAE,CAACW,CAAC,EAAEzB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACW,CAAC,IAAI,IAAK;gBACzMX;gBACAY,IAAIb,IAAI,CAACb,EAAEe,KAAK,CAACD,EAAE;gBACnBa,OAAOC,KAAKC,GAAG,CAACF,MAAMnE,OAAOwC,EAAEe,KAAK,CAACD,EAAE,CAACW,CAAC;YAC3C;YACA,MAAMK,WAAWJ,IAAInB,GAAG,CAAC,CAACC,IAAMA,EAAEtB,IAAI;YACtC,MAAMuB,SAASqB,SAAShE,KAAK,CAAC,GAAG4C,MAAM,CAACjD,YAAYqE,QAAQ,CAAC,EAAE;YAC/D,MAAMnB,QAAQe,IAAId,OAAO,CAAC,CAACJ,IAAMA,EAAEG,KAAK;YACxC,sEAAsE;YACtE,qFAAqF;YACrF,MAAMoB,OAAO9C,eAAewB;YAC5B,IAAIsB,MAAM5B,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAAUV,MAAMuB;gBAAQhB,QAAQX,cAAcC,kBAAkB;gBAAE4B;YAAM;iBACjG,IAAIF,OAAOpB,MAAM,GAAG,KAAKc,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAASV,MAAMuB;gBAAQE;YAAM;iBAC1ER,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAAWV,MAAMuB;gBAAQX,OAAOG,QAAQ+B,GAAG,CAACL;gBAAiBhB;YAAM;QAC9F,OAAO;YACLP,KAAKS,IAAI,CAACG;QACZ;QACAF;IACF;IACAT;IACA,OAAOF;AACT"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/emit.ts"],"sourcesContent":["// emit.ts — OCR-path pass 3: lines -> markdown blocks.\nimport { getDict } from './dict.ts';\nimport type { OcrLine, PageOCR, Tuning, Word } from './geometry.ts';\nimport { bucket } from './geometry.ts';\n\n// Join a wrapped line onto the open paragraph, closing a line-final hyphen\n// only when the dictionary confirms the joined form is a word.\nfunction hyphenJoin(acc: string, next: string): string {\n if (acc.endsWith('-')) {\n const withoutDash = acc.slice(0, -1);\n const pen = withoutDash.at(-1);\n if (pen && /[A-Za-z0-9]/.test(pen)) {\n const head = withoutDash.split(' ').pop() ?? '';\n const firstTok = next.split(' ')[0] ?? '';\n const filtered = [...firstTok].filter((c) => /[A-Za-z-]/.test(c)).join('');\n const tail = filtered.split('-')[0] ?? '';\n const close = getDict().has((head + tail).toLowerCase());\n return (close ? withoutDash : acc) + next;\n }\n }\n return `${acc} ${next}`;\n}\n\n// Blocks with too few word-like tokens to trust as prose (see\n// UNSTRUCTURED_WORDLIKE_FRACTION below) are labeled `> [review: reason] text` instead of guessed at.\nconst REVIEW_REASON = {\n unstructuredLabels: 'unstructured labels',\n} as const;\n\n/** Share of word-like tokens (3+ letter run) below which a block counts as\n * unstructured. Measured (`review-blocks.test.ts`): junk 0.000-0.100 vs prose 0.743-0.857, cut 0.4. */\nexport const UNSTRUCTURED_WORDLIKE_FRACTION = 0.4;\nfunction isUnstructured(text: string): boolean {\n const tokens = text.split(/\\s+/).filter(Boolean);\n if (tokens.length === 0) return false; // never mark an empty block\n const wordLike = tokens.filter((t) => /[A-Za-z]{3}/.test(t)).length;\n return wordLike / tokens.length < UNSTRUCTURED_WORDLIKE_FRACTION;\n}\n\nfunction reviewLine(reason: string, text: string): string {\n return `> [review: ${reason}] ${text}`;\n}\n\n// One classified block, carrying constituent words for analyze()'s\n// block->word linkage; discriminated on `kind` so `reason`/`level` are compile-time guaranteed present on their variants.\nexport type EmitBlock =\n | { kind: 'heading'; text: string; level: number; words: Word[] }\n | { kind: 'paragraph'; text: string; words: Word[] }\n | { kind: 'quote'; text: string; words: Word[] }\n | {\n kind: 'review';\n text: string;\n /** The parseable reason inside `> [review: ...]`, carried on the block\n * so more reasons can be added without changing the render format. */\n reason: string;\n words: Word[];\n };\n\n// Renders one EmitBlock to the exact markdown emitPage used to push\n// directly — shared so extract.ts and analyze.ts produce byte-identical text.\nexport function renderEmitBlock(b: EmitBlock): string {\n switch (b.kind) {\n case 'heading':\n return `${'#'.repeat(b.level)} ${b.text}`;\n case 'quote':\n return `> ${b.text}`;\n case 'review':\n return reviewLine(b.reason, b.text);\n default:\n return b.text;\n }\n}\n\nexport function emitPage(p: PageOCR, levelOf: Map<number, number>, tuning: Tuning): EmitBlock[] {\n const blocks: EmitBlock[] = [];\n let para: OcrLine[] = [];\n const flush = () => {\n if (!para.length) return;\n const texts = para.map((l) => l.text);\n const merged = texts.slice(1).reduce(hyphenJoin, texts[0]);\n const words = para.flatMap((l) => l.words);\n // Same review-marker rule as the heading path below: no word may be\n // dropped or reordered here, only how the block renders changes.\n blocks.push(isUnstructured(merged) ? { kind: 'review', text: merged, reason: REVIEW_REASON.unstructuredLabels, words } : { kind: 'paragraph', text: merged, words });\n para = [];\n };\n\n let i = 0;\n while (i < p.lines.length) {\n const line: OcrLine = p.lines[i];\n const prev = i > 0 ? p.lines[i - 1] : null;\n if (prev && (line.col !== prev.col || prev.midY - line.midY > p.pitch * tuning.paraGap)) {\n // Join across a column break (instead of flushing) only when the column\n // changed, the paragraph dangles (no terminal punct), and the new text starts lowercase; a miss stays visible via pageStats.danglingLong.\n const danglingEnd = para.length > 0 && !/[.!?:;\"'”’)\\]]$/.test(para[para.length - 1].text);\n const continuationStart = /^[a-z]/.test(line.text);\n const joinAcrossColumns = line.col !== prev.col && danglingEnd && continuationStart && !levelOf.has(bucket(line.h));\n if (!joinAcrossColumns) flush();\n }\n\n if (levelOf.has(bucket(line.h))) {\n // Gather the run of consecutive heading-size lines: a wrapped title is\n // one heading, a long run is a pull quote in a bigger face.\n flush();\n const run: OcrLine[] = [line];\n let maxB = bucket(line.h);\n while (i + 1 < p.lines.length && levelOf.has(bucket(p.lines[i + 1].h)) && p.lines[i + 1].col === p.lines[i].col && p.lines[i].midY - p.lines[i + 1].midY < Math.max(p.lines[i].h, p.lines[i + 1].h) * 2.2) {\n i++;\n run.push(p.lines[i]);\n maxB = Math.max(maxB, bucket(p.lines[i].h));\n }\n const runTexts = run.map((l) => l.text);\n const merged = runTexts.slice(1).reduce(hyphenJoin, runTexts[0]);\n const words = run.flatMap((l) => l.words);\n // Below UNSTRUCTURED_WORDLIKE_FRACTION = chart labels/stray numerals,\n // not a heading; checked FIRST so a long numeral run isn't asserted as a pull quote.\n const junk = isUnstructured(merged);\n if (junk) blocks.push({ kind: 'review', text: merged, reason: REVIEW_REASON.unstructuredLabels, words });\n else if (merged.length > 150) blocks.push({ kind: 'quote', text: merged, words });\n else blocks.push({ kind: 'heading', text: merged, level: levelOf.get(maxB) as number, words });\n } else {\n para.push(line);\n }\n i++;\n }\n flush();\n return blocks;\n}\n"],"names":["getDict","bucket","hyphenJoin","acc","next","endsWith","withoutDash","slice","pen","at","test","filtered","head","split","pop","firstTok","filter","c","join","tail","close","has","toLowerCase","REVIEW_REASON","unstructuredLabels","UNSTRUCTURED_WORDLIKE_FRACTION","isUnstructured","text","tokens","Boolean","length","wordLike","t","reviewLine","reason","renderEmitBlock","b","kind","repeat","level","emitPage","p","levelOf","tuning","blocks","para","flush","texts","map","l","merged","reduce","words","flatMap","push","i","lines","line","prev","col","midY","pitch","paraGap","danglingEnd","continuationStart","joinAcrossColumns","h","run","maxB","Math","max","runTexts","junk","get"],"mappings":"AAAA,uDAAuD;AACvD,SAASA,OAAO,QAAQ,YAAY;AAEpC,SAASC,MAAM,QAAQ,gBAAgB;AAEvC,2EAA2E;AAC3E,+DAA+D;AAC/D,SAASC,WAAWC,GAAW,EAAEC,IAAY;IAC3C,IAAID,IAAIE,QAAQ,CAAC,MAAM;QACrB,MAAMC,cAAcH,IAAII,KAAK,CAAC,GAAG,CAAC;QAClC,MAAMC,MAAMF,YAAYG,EAAE,CAAC,CAAC;QAC5B,IAAID,OAAO,cAAcE,IAAI,CAACF,MAAM;gBACrBF,wBACIF,cAEJO;YAHb,MAAMC,QAAON,yBAAAA,YAAYO,KAAK,CAAC,KAAKC,GAAG,gBAA1BR,oCAAAA,yBAAgC;YAC7C,MAAMS,YAAWX,eAAAA,KAAKS,KAAK,CAAC,IAAI,CAAC,EAAE,cAAlBT,0BAAAA,eAAsB;YACvC,MAAMO,WAAW;mBAAII;aAAS,CAACC,MAAM,CAAC,CAACC,IAAM,YAAYP,IAAI,CAACO,IAAIC,IAAI,CAAC;YACvE,MAAMC,QAAOR,mBAAAA,SAASE,KAAK,CAAC,IAAI,CAAC,EAAE,cAAtBF,8BAAAA,mBAA0B;YACvC,MAAMS,QAAQpB,UAAUqB,GAAG,CAAC,AAACT,CAAAA,OAAOO,IAAG,EAAGG,WAAW;YACrD,OAAO,AAACF,CAAAA,QAAQd,cAAcH,GAAE,IAAKC;QACvC;IACF;IACA,OAAO,GAAGD,IAAI,CAAC,EAAEC,MAAM;AACzB;AAEA,8DAA8D;AAC9D,qGAAqG;AACrG,MAAMmB,gBAAgB;IACpBC,oBAAoB;AACtB;AAEA;qGACqG,GACrG,OAAO,MAAMC,iCAAiC,IAAI;AAClD,SAASC,eAAeC,IAAY;IAClC,MAAMC,SAASD,KAAKd,KAAK,CAAC,OAAOG,MAAM,CAACa;IACxC,IAAID,OAAOE,MAAM,KAAK,GAAG,OAAO,OAAO,4BAA4B;IACnE,MAAMC,WAAWH,OAAOZ,MAAM,CAAC,CAACgB,IAAM,cAActB,IAAI,CAACsB,IAAIF,MAAM;IACnE,OAAOC,WAAWH,OAAOE,MAAM,GAAGL;AACpC;AAEA,SAASQ,WAAWC,MAAc,EAAEP,IAAY;IAC9C,OAAO,CAAC,WAAW,EAAEO,OAAO,EAAE,EAAEP,MAAM;AACxC;AAiBA,oEAAoE;AACpE,8EAA8E;AAC9E,OAAO,SAASQ,gBAAgBC,CAAY;IAC1C,OAAQA,EAAEC,IAAI;QACZ,KAAK;YACH,OAAO,GAAG,IAAIC,MAAM,CAACF,EAAEG,KAAK,EAAE,CAAC,EAAEH,EAAET,IAAI,EAAE;QAC3C,KAAK;YACH,OAAO,CAAC,EAAE,EAAES,EAAET,IAAI,EAAE;QACtB,KAAK;YACH,OAAOM,WAAWG,EAAEF,MAAM,EAAEE,EAAET,IAAI;QACpC;YACE,OAAOS,EAAET,IAAI;IACjB;AACF;AAEA,OAAO,SAASa,SAASC,CAAU,EAAEC,OAA4B,EAAEC,MAAc;IAC/E,MAAMC,SAAsB,EAAE;IAC9B,IAAIC,OAAkB,EAAE;IACxB,MAAMC,QAAQ;QACZ,IAAI,CAACD,KAAKf,MAAM,EAAE;QAClB,MAAMiB,QAAQF,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAEtB,IAAI;QACpC,MAAMuB,SAASH,MAAMxC,KAAK,CAAC,GAAG4C,MAAM,CAACjD,YAAY6C,KAAK,CAAC,EAAE;QACzD,MAAMK,QAAQP,KAAKQ,OAAO,CAAC,CAACJ,IAAMA,EAAEG,KAAK;QACzC,oEAAoE;QACpE,iEAAiE;QACjER,OAAOU,IAAI,CAAC5B,eAAewB,UAAU;YAAEb,MAAM;YAAUV,MAAMuB;YAAQhB,QAAQX,cAAcC,kBAAkB;YAAE4B;QAAM,IAAI;YAAEf,MAAM;YAAaV,MAAMuB;YAAQE;QAAM;QAClKP,OAAO,EAAE;IACX;IAEA,IAAIU,IAAI;IACR,MAAOA,IAAId,EAAEe,KAAK,CAAC1B,MAAM,CAAE;QACzB,MAAM2B,OAAgBhB,EAAEe,KAAK,CAACD,EAAE;QAChC,MAAMG,OAAOH,IAAI,IAAId,EAAEe,KAAK,CAACD,IAAI,EAAE,GAAG;QACtC,IAAIG,QAASD,CAAAA,KAAKE,GAAG,KAAKD,KAAKC,GAAG,IAAID,KAAKE,IAAI,GAAGH,KAAKG,IAAI,GAAGnB,EAAEoB,KAAK,GAAGlB,OAAOmB,OAAO,AAAD,GAAI;YACvF,wEAAwE;YACxE,0IAA0I;YAC1I,MAAMC,cAAclB,KAAKf,MAAM,GAAG,KAAK,CAAC,kBAAkBpB,IAAI,CAACmC,IAAI,CAACA,KAAKf,MAAM,GAAG,EAAE,CAACH,IAAI;YACzF,MAAMqC,oBAAoB,SAAStD,IAAI,CAAC+C,KAAK9B,IAAI;YACjD,MAAMsC,oBAAoBR,KAAKE,GAAG,KAAKD,KAAKC,GAAG,IAAII,eAAeC,qBAAqB,CAACtB,QAAQrB,GAAG,CAACpB,OAAOwD,KAAKS,CAAC;YACjH,IAAI,CAACD,mBAAmBnB;QAC1B;QAEA,IAAIJ,QAAQrB,GAAG,CAACpB,OAAOwD,KAAKS,CAAC,IAAI;YAC/B,uEAAuE;YACvE,4DAA4D;YAC5DpB;YACA,MAAMqB,MAAiB;gBAACV;aAAK;YAC7B,IAAIW,OAAOnE,OAAOwD,KAAKS,CAAC;YACxB,MAAOX,IAAI,IAAId,EAAEe,KAAK,CAAC1B,MAAM,IAAIY,QAAQrB,GAAG,CAACpB,OAAOwC,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACW,CAAC,MAAMzB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACI,GAAG,KAAKlB,EAAEe,KAAK,CAACD,EAAE,CAACI,GAAG,IAAIlB,EAAEe,KAAK,CAACD,EAAE,CAACK,IAAI,GAAGnB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACK,IAAI,GAAGS,KAAKC,GAAG,CAAC7B,EAAEe,KAAK,CAACD,EAAE,CAACW,CAAC,EAAEzB,EAAEe,KAAK,CAACD,IAAI,EAAE,CAACW,CAAC,IAAI,IAAK;gBACzMX;gBACAY,IAAIb,IAAI,CAACb,EAAEe,KAAK,CAACD,EAAE;gBACnBa,OAAOC,KAAKC,GAAG,CAACF,MAAMnE,OAAOwC,EAAEe,KAAK,CAACD,EAAE,CAACW,CAAC;YAC3C;YACA,MAAMK,WAAWJ,IAAInB,GAAG,CAAC,CAACC,IAAMA,EAAEtB,IAAI;YACtC,MAAMuB,SAASqB,SAAShE,KAAK,CAAC,GAAG4C,MAAM,CAACjD,YAAYqE,QAAQ,CAAC,EAAE;YAC/D,MAAMnB,QAAQe,IAAId,OAAO,CAAC,CAACJ,IAAMA,EAAEG,KAAK;YACxC,sEAAsE;YACtE,qFAAqF;YACrF,MAAMoB,OAAO9C,eAAewB;YAC5B,IAAIsB,MAAM5B,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAAUV,MAAMuB;gBAAQhB,QAAQX,cAAcC,kBAAkB;gBAAE4B;YAAM;iBACjG,IAAIF,OAAOpB,MAAM,GAAG,KAAKc,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAASV,MAAMuB;gBAAQE;YAAM;iBAC1ER,OAAOU,IAAI,CAAC;gBAAEjB,MAAM;gBAAWV,MAAMuB;gBAAQX,OAAOG,QAAQ+B,GAAG,CAACL;gBAAiBhB;YAAM;QAC9F,OAAO;YACLP,KAAKS,IAAI,CAACG;QACZ;QACAF;IACF;IACAT;IACA,OAAOF;AACT"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/engines/tesseract.ts"],"sourcesContent":["// engines/tesseract.ts — THE OCR engine: tesseract.js on every platform.\n//\n// Seam contract: words come back in normalized 0-1 page coordinates, origin\n// bottom-left, y increasing upward — the y-flip happens here, and nowhere else.\n//\n// Pure WASM, in-process, no native dependency: no crash containment needed.\nimport { mkdirSync } from 'node:fs';\nimport os from 'node:os';\nimport path from 'node:path';\nimport type { Page as TesseractPage } from 'tesseract.js';\nimport { createWorker, PSM } from 'tesseract.js';\nimport type { PageWord, Word } from '../geometry.ts';\nimport { foldTypographic } from '../lines.ts';\nimport { awaitPdfOpen, openPdfForRender } from '../pdf-open.ts';\nimport { renderLoadedPage } from '../raster.ts';\nimport { resolvePackagePath } from '../resolve.ts';\n\n// 288 dpi: the render resolution the engine's word-overlap numbers were\n// measured at (test/fixtures/ocr-engine-poc/) — the known-good operating point.\nexport const OCR_RENDER_DPI = 288;\n\n// Chart pages starve PSM.AUTO (unrecognized text can't be marked for review).\n// Prose-fraction retry trigger, measured (sparse-retry.test.ts): chart 0.442 vs prose 0.767-0.938, cut 0.6.\nexport const SPARSE_RETRY_PROSE_FRACTION = 0.6;\n\n/** Share of `words` whose text contains a run of 3+ letters (`src/emit.ts`'s isUnstructured canary, applied per-page instead of per-block). */\nfunction proseFraction(words: { text: string }[]): number {\n if (words.length === 0) return 0;\n return words.filter((w) => /[A-Za-z]{3}/.test(w.text)).length / words.length;\n}\n\n/** Does this page's PSM.AUTO pass look chart-like enough to re-recognize\n * with PSM.SPARSE_TEXT? A page with 0 words never retries (nothing to take a prose fraction of). */\nexport function shouldRetrySparse(words: { text: string }[]): boolean {\n return words.length > 0 && proseFraction(words) < SPARSE_RETRY_PROSE_FRACTION;\n}\n\n// The locally installed traineddata (npm package @tesseract.js-data/eng),\n// resolved dynamically so depcheck won't flag it \"unused\" — it's a genuine runtime dependency, listed in .depcheckrc's `ignores`.\nconst TESSDATA_DIR = path.join(path.dirname(resolvePackagePath('@tesseract.js-data/eng/package.json')), '4.0.0');\n\n/** Per-page recognition progress, delivered once per page as it finishes.\n * Engine chatter goes through `onWarning`/`warnings` instead — this is timing/count data only. */\nexport type OcrProgressEvent = { type: 'page'; page: number; words: number; ms: number };\n\nexport interface RecognizeResult {\n words: PageWord[];\n /** pages (1-based) where the engine returned zero words */\n emptyPages: number[];\n /** wall-clock ms per page (render + recognize), keyed by page number */\n msPerPage: Map<number, number>;\n}\n\n/** tesseract.js's nested blocks/paragraphs/lines/words -> this repo's flat Word contract (normalize + y-flip). Same shape regardless of which PSM produced `page`, so both the AUTO pass and the SPARSE_TEXT retry share it. */\nfunction wordsFromPage(page: TesseractPage, width: number, height: number): Word[] {\n const pageWords: Word[] = [];\n for (const block of page.blocks ?? []) {\n for (const para of block.paragraphs ?? []) {\n for (const line of para.lines ?? []) {\n for (const w of line.words ?? []) {\n const { x0, y0, x1, y1 } = w.bbox;\n pageWords.push({\n // Same typographic folding as the text path (src/lines.ts) — both\n // paths must agree on how a printed apostrophe is spelled.\n text: foldTypographic(w.text),\n x: x0 / width,\n y: (height - y1) / height, // y-flip: top-left px -> bottom-left normalized\n w: (x1 - x0) / width,\n h: (y1 - y0) / height,\n // Plumbed through unmodified — tesseract.js's own 0-100\n // per-word confidence. See src/report.ts's LOW_CONFIDENCE_THRESHOLD.\n confidence: w.confidence,\n });\n }\n }\n }\n }\n return pageWords;\n}\n\n// tesseract.js-core's WASM build writes native diagnostics straight to this\n// process's stdout/stderr, with no override to silence them at the source.\n//\n// Suppression must span the whole run: worker-thread stdio arrives\n// asynchronously, on later event-loop turns; the synchronous carve-out below (`withRealStdio`) is the only safe exit.\nlet suppressDepth = 0;\nlet savedStdoutWrite: typeof process.stdout.write | null = null;\nlet savedStderrWrite: typeof process.stderr.write | null = null;\nasync function withSuppressedStdio<T>(fn: () => Promise<T>): Promise<T> {\n if (suppressDepth === 0) {\n savedStdoutWrite = process.stdout.write.bind(process.stdout);\n savedStderrWrite = process.stderr.write.bind(process.stderr);\n process.stdout.write = (() => true) as typeof process.stdout.write;\n process.stderr.write = (() => true) as typeof process.stderr.write;\n }\n suppressDepth++;\n try {\n return await fn();\n } finally {\n suppressDepth--;\n if (suppressDepth === 0) {\n if (savedStdoutWrite) process.stdout.write = savedStdoutWrite;\n if (savedStderrWrite) process.stderr.write = savedStderrWrite;\n savedStdoutWrite = null;\n savedStderrWrite = null;\n }\n }\n}\n\n/** Pierces an active `withSuppressedStdio` region SYNCHRONOUSLY to let this\n * module's own onProgress/onWarning calls through. No-op outside any suppressed region. */\nfunction withRealStdio<T>(fn: () => T): T {\n if (suppressDepth === 0) return fn();\n const realOut = savedStdoutWrite;\n const realErr = savedStderrWrite;\n if (realOut) process.stdout.write = realOut;\n if (realErr) process.stderr.write = realErr;\n try {\n return fn();\n } finally {\n process.stdout.write = (() => true) as typeof process.stdout.write;\n process.stderr.write = (() => true) as typeof process.stderr.write;\n }\n}\n\n/** Recognize words on pages [first..last] of a PDF (renders at `dpi`,\n * default OCR_RENDER_DPI). Progress via `onProgress`; diagnostics via `onWarning` — this module never writes to stdout/stderr itself. */\nexport async function recognizePdf(pdfPath: string, first: number, last: number, { dpi = OCR_RENDER_DPI, onProgress, onWarning }: { dpi?: number; onProgress?: (e: OcrProgressEvent) => void; onWarning?: (message: string) => void } = {}): Promise<RecognizeResult> {\n // tesseract.js writes its decompressed traineddata cache here by default,\n // dropping a model file into the user's cwd — redirect it under tmpdir() instead.\n const cachePath = path.join(os.tmpdir(), 'pdf-to-md-engine');\n mkdirSync(cachePath, { recursive: true });\n\n const words: PageWord[] = [];\n const emptyPages: number[] = [];\n const msPerPage = new Map<number, number>();\n\n // ONE continuous suppressed region for the whole run; onProgress/onWarning\n // below go through `withRealStdio`, the only sanctioned way out of it.\n await withSuppressedStdio(async () => {\n // OEM 1 (LSTM only, tesseract.js's own default): measured 99.0% recall\n // on the hard benchmark pages (2026-08-16).\n const worker = await createWorker('eng', 1, {\n langPath: TESSDATA_DIR,\n cachePath,\n });\n try {\n // NOT optional: tesseract.js's own default PSM (SINGLE_BLOCK) measured\n // 3-8 percentage points worse than AUTO on two-column pages.\n await worker.setParameters({ tessedit_pageseg_mode: PSM.AUTO });\n\n const loadingTask = openPdfForRender(pdfPath);\n try {\n const pdf = await awaitPdfOpen(loadingTask, pdfPath);\n for (let pn = first; pn <= Math.min(last, pdf.numPages); pn++) {\n const t0 = Date.now();\n const { png, width, height } = await renderLoadedPage(pdf, pn, dpi);\n const { data } = await worker.recognize(png, {}, { blocks: true });\n let pageWords: Word[] = wordsFromPage(data, width, height);\n\n // Adaptive second pass: re-recognizes the SAME rendered PNG, so\n // this costs one extra recognize() call only on chart-like pages.\n if (shouldRetrySparse(pageWords)) {\n const autoFraction = proseFraction(pageWords);\n const autoWords = pageWords;\n // setParameters on the shared worker is reversible, so one\n // in-process worker can alternate PSM per page instead of paying for a second worker.\n await worker.setParameters({ tessedit_pageseg_mode: PSM.SPARSE_TEXT });\n const sparse = await worker.recognize(png, {}, { blocks: true });\n const sparseWords = wordsFromPage(sparse.data, width, height);\n await worker.setParameters({ tessedit_pageseg_mode: PSM.AUTO });\n\n if (sparseWords.length > autoWords.length) {\n pageWords = sparseWords;\n const autoCount = autoWords.length;\n const sparseCount = sparseWords.length;\n withRealStdio(() => onWarning?.(`p${pn}: chart-like page (prose fraction ${autoFraction.toFixed(2)}) — retried with sparse-text segmentation, ${autoCount} -> ${sparseCount} words`));\n }\n }\n\n const ms = Date.now() - t0;\n for (const w of pageWords) words.push({ page: pn, ...w });\n if (pageWords.length === 0) emptyPages.push(pn);\n msPerPage.set(pn, ms);\n const wordCount = pageWords.length;\n withRealStdio(() => onProgress?.({ type: 'page', page: pn, words: wordCount, ms }));\n }\n } finally {\n await loadingTask.destroy();\n }\n } finally {\n await worker.terminate();\n }\n });\n\n return { words, emptyPages, msPerPage };\n}\n"],"names":["mkdirSync","os","path","createWorker","PSM","foldTypographic","awaitPdfOpen","openPdfForRender","renderLoadedPage","resolvePackagePath","OCR_RENDER_DPI","SPARSE_RETRY_PROSE_FRACTION","proseFraction","words","length","filter","w","test","text","shouldRetrySparse","TESSDATA_DIR","join","dirname","wordsFromPage","page","width","height","pageWords","block","blocks","para","paragraphs","line","lines","x0","y0","x1","y1","bbox","push","x","y","h","confidence","suppressDepth","savedStdoutWrite","savedStderrWrite","withSuppressedStdio","fn","process","stdout","write","bind","stderr","withRealStdio","realOut","realErr","recognizePdf","pdfPath","first","last","dpi","onProgress","onWarning","cachePath","tmpdir","recursive","emptyPages","msPerPage","Map","worker","langPath","setParameters","tessedit_pageseg_mode","AUTO","loadingTask","pdf","pn","Math","min","numPages","t0","Date","now","png","data","recognize","autoFraction","autoWords","SPARSE_TEXT","sparse","sparseWords","autoCount","sparseCount","toFixed","ms","set","wordCount","type","destroy","terminate"],"mappings":"AAAA,yEAAyE;AACzE,EAAE;AACF,4EAA4E;AAC5E,gFAAgF;AAChF,EAAE;AACF,4EAA4E;AAC5E,SAASA,SAAS,QAAQ,UAAU;AACpC,OAAOC,QAAQ,UAAU;AACzB,OAAOC,UAAU,YAAY;AAE7B,SAASC,YAAY,EAAEC,GAAG,QAAQ,eAAe;AAEjD,SAASC,eAAe,QAAQ,cAAc;AAC9C,SAASC,YAAY,EAAEC,gBAAgB,QAAQ,iBAAiB;AAChE,SAASC,gBAAgB,QAAQ,eAAe;AAChD,SAASC,kBAAkB,QAAQ,gBAAgB;AAEnD,wEAAwE;AACxE,gFAAgF;AAChF,OAAO,MAAMC,iBAAiB,IAAI;AAElC,8EAA8E;AAC9E,4GAA4G;AAC5G,OAAO,MAAMC,8BAA8B,IAAI;AAE/C,6IAA6I,GAC7I,SAASC,cAAcC,KAAyB;IAC9C,IAAIA,MAAMC,MAAM,KAAK,GAAG,OAAO;IAC/B,OAAOD,MAAME,MAAM,CAAC,CAACC,IAAM,cAAcC,IAAI,CAACD,EAAEE,IAAI,GAAGJ,MAAM,GAAGD,MAAMC,MAAM;AAC9E;AAEA;kGACkG,GAClG,OAAO,SAASK,kBAAkBN,KAAyB;IACzD,OAAOA,MAAMC,MAAM,GAAG,KAAKF,cAAcC,SAASF;AACpD;AAEA,0EAA0E;AAC1E,kIAAkI;AAClI,MAAMS,eAAelB,KAAKmB,IAAI,CAACnB,KAAKoB,OAAO,CAACb,mBAAmB,yCAAyC;AAcxG,8NAA8N,GAC9N,SAASc,cAAcC,IAAmB,EAAEC,KAAa,EAAEC,MAAc;QAEnDF;IADpB,MAAMG,YAAoB,EAAE;IAC5B,KAAK,MAAMC,UAASJ,eAAAA,KAAKK,MAAM,cAAXL,0BAAAA,eAAe,EAAE,CAAE;YAClBI;QAAnB,KAAK,MAAME,SAAQF,oBAAAA,MAAMG,UAAU,cAAhBH,+BAAAA,oBAAoB,EAAE,CAAE;gBACtBE;YAAnB,KAAK,MAAME,SAAQF,cAAAA,KAAKG,KAAK,cAAVH,yBAAAA,cAAc,EAAE,CAAE;oBACnBE;gBAAhB,KAAK,MAAMhB,MAAKgB,cAAAA,KAAKnB,KAAK,cAAVmB,yBAAAA,cAAc,EAAE,CAAE;oBAChC,MAAM,EAAEE,EAAE,EAAEC,EAAE,EAAEC,EAAE,EAAEC,EAAE,EAAE,GAAGrB,EAAEsB,IAAI;oBACjCX,UAAUY,IAAI,CAAC;wBACb,kEAAkE;wBAClE,2DAA2D;wBAC3DrB,MAAMb,gBAAgBW,EAAEE,IAAI;wBAC5BsB,GAAGN,KAAKT;wBACRgB,GAAG,AAACf,CAAAA,SAASW,EAAC,IAAKX;wBACnBV,GAAG,AAACoB,CAAAA,KAAKF,EAAC,IAAKT;wBACfiB,GAAG,AAACL,CAAAA,KAAKF,EAAC,IAAKT;wBACf,wDAAwD;wBACxD,qEAAqE;wBACrEiB,YAAY3B,EAAE2B,UAAU;oBAC1B;gBACF;YACF;QACF;IACF;IACA,OAAOhB;AACT;AAEA,4EAA4E;AAC5E,2EAA2E;AAC3E,EAAE;AACF,mEAAmE;AACnE,sHAAsH;AACtH,IAAIiB,gBAAgB;AACpB,IAAIC,mBAAuD;AAC3D,IAAIC,mBAAuD;AAC3D,eAAeC,oBAAuBC,EAAoB;IACxD,IAAIJ,kBAAkB,GAAG;QACvBC,mBAAmBI,QAAQC,MAAM,CAACC,KAAK,CAACC,IAAI,CAACH,QAAQC,MAAM;QAC3DJ,mBAAmBG,QAAQI,MAAM,CAACF,KAAK,CAACC,IAAI,CAACH,QAAQI,MAAM;QAC3DJ,QAAQC,MAAM,CAACC,KAAK,GAAI,IAAM;QAC9BF,QAAQI,MAAM,CAACF,KAAK,GAAI,IAAM;IAChC;IACAP;IACA,IAAI;QACF,OAAO,MAAMI;IACf,SAAU;QACRJ;QACA,IAAIA,kBAAkB,GAAG;YACvB,IAAIC,kBAAkBI,QAAQC,MAAM,CAACC,KAAK,GAAGN;YAC7C,IAAIC,kBAAkBG,QAAQI,MAAM,CAACF,KAAK,GAAGL;YAC7CD,mBAAmB;YACnBC,mBAAmB;QACrB;IACF;AACF;AAEA;yFACyF,GACzF,SAASQ,cAAiBN,EAAW;IACnC,IAAIJ,kBAAkB,GAAG,OAAOI;IAChC,MAAMO,UAAUV;IAChB,MAAMW,UAAUV;IAChB,IAAIS,SAASN,QAAQC,MAAM,CAACC,KAAK,GAAGI;IACpC,IAAIC,SAASP,QAAQI,MAAM,CAACF,KAAK,GAAGK;IACpC,IAAI;QACF,OAAOR;IACT,SAAU;QACRC,QAAQC,MAAM,CAACC,KAAK,GAAI,IAAM;QAC9BF,QAAQI,MAAM,CAACF,KAAK,GAAI,IAAM;IAChC;AACF;AAEA;uIACuI,GACvI,OAAO,eAAeM,aAAaC,OAAe,EAAEC,KAAa,EAAEC,IAAY,EAAE,EAAEC,MAAMnD,cAAc,EAAEoD,UAAU,EAAEC,SAAS,EAAuG,GAAG,CAAC,CAAC;IACxO,0EAA0E;IAC1E,kFAAkF;IAClF,MAAMC,YAAY9D,KAAKmB,IAAI,CAACpB,GAAGgE,MAAM,IAAI;IACzCjE,UAAUgE,WAAW;QAAEE,WAAW;IAAK;IAEvC,MAAMrD,QAAoB,EAAE;IAC5B,MAAMsD,aAAuB,EAAE;IAC/B,MAAMC,YAAY,IAAIC;IAEtB,2EAA2E;IAC3E,uEAAuE;IACvE,MAAMtB,oBAAoB;QACxB,uEAAuE;QACvE,4CAA4C;QAC5C,MAAMuB,SAAS,MAAMnE,aAAa,OAAO,GAAG;YAC1CoE,UAAUnD;YACV4C;QACF;QACA,IAAI;YACF,uEAAuE;YACvE,6DAA6D;YAC7D,MAAMM,OAAOE,aAAa,CAAC;gBAAEC,uBAAuBrE,IAAIsE,IAAI;YAAC;YAE7D,MAAMC,cAAcpE,iBAAiBmD;YACrC,IAAI;gBACF,MAAMkB,MAAM,MAAMtE,aAAaqE,aAAajB;gBAC5C,IAAK,IAAImB,KAAKlB,OAAOkB,MAAMC,KAAKC,GAAG,CAACnB,MAAMgB,IAAII,QAAQ,GAAGH,KAAM;oBAC7D,MAAMI,KAAKC,KAAKC,GAAG;oBACnB,MAAM,EAAEC,GAAG,EAAE3D,KAAK,EAAEC,MAAM,EAAE,GAAG,MAAMlB,iBAAiBoE,KAAKC,IAAIhB;oBAC/D,MAAM,EAAEwB,IAAI,EAAE,GAAG,MAAMf,OAAOgB,SAAS,CAACF,KAAK,CAAC,GAAG;wBAAEvD,QAAQ;oBAAK;oBAChE,IAAIF,YAAoBJ,cAAc8D,MAAM5D,OAAOC;oBAEnD,gEAAgE;oBAChE,kEAAkE;oBAClE,IAAIP,kBAAkBQ,YAAY;wBAChC,MAAM4D,eAAe3E,cAAce;wBACnC,MAAM6D,YAAY7D;wBAClB,2DAA2D;wBAC3D,sFAAsF;wBACtF,MAAM2C,OAAOE,aAAa,CAAC;4BAAEC,uBAAuBrE,IAAIqF,WAAW;wBAAC;wBACpE,MAAMC,SAAS,MAAMpB,OAAOgB,SAAS,CAACF,KAAK,CAAC,GAAG;4BAAEvD,QAAQ;wBAAK;wBAC9D,MAAM8D,cAAcpE,cAAcmE,OAAOL,IAAI,EAAE5D,OAAOC;wBACtD,MAAM4C,OAAOE,aAAa,CAAC;4BAAEC,uBAAuBrE,IAAIsE,IAAI;wBAAC;wBAE7D,IAAIiB,YAAY7E,MAAM,GAAG0E,UAAU1E,MAAM,EAAE;4BACzCa,YAAYgE;4BACZ,MAAMC,YAAYJ,UAAU1E,MAAM;4BAClC,MAAM+E,cAAcF,YAAY7E,MAAM;4BACtCwC,cAAc,IAAMS,sBAAAA,gCAAAA,UAAY,CAAC,CAAC,EAAEc,GAAG,kCAAkC,EAAEU,aAAaO,OAAO,CAAC,GAAG,2CAA2C,EAAEF,UAAU,IAAI,EAAEC,YAAY,MAAM,CAAC;wBACrL;oBACF;oBAEA,MAAME,KAAKb,KAAKC,GAAG,KAAKF;oBACxB,KAAK,MAAMjE,KAAKW,UAAWd,MAAM0B,IAAI,CAAC;wBAAEf,MAAMqD;wBAAI,GAAG7D,CAAC;oBAAC;oBACvD,IAAIW,UAAUb,MAAM,KAAK,GAAGqD,WAAW5B,IAAI,CAACsC;oBAC5CT,UAAU4B,GAAG,CAACnB,IAAIkB;oBAClB,MAAME,YAAYtE,UAAUb,MAAM;oBAClCwC,cAAc,IAAMQ,uBAAAA,iCAAAA,WAAa;4BAAEoC,MAAM;4BAAQ1E,MAAMqD;4BAAIhE,OAAOoF;4BAAWF;wBAAG;gBAClF;YACF,SAAU;gBACR,MAAMpB,YAAYwB,OAAO;YAC3B;QACF,SAAU;YACR,MAAM7B,OAAO8B,SAAS;QACxB;IACF;IAEA,OAAO;QAAEvF;QAAOsD;QAAYC;IAAU;AACxC"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/engines/tesseract.ts"],"sourcesContent":["// engines/tesseract.ts — THE OCR engine: tesseract.js on every platform.\n//\n// Seam contract: words come back in normalized 0-1 page coordinates, origin\n// bottom-left, y increasing upward — the y-flip happens here, and nowhere else.\n//\n// Pure WASM, in-process, no native dependency: no crash containment needed.\nimport { mkdirSync } from 'node:fs';\nimport os from 'node:os';\nimport path from 'node:path';\nimport type { Page as TesseractPage } from 'tesseract.js';\nimport { createWorker, PSM } from 'tesseract.js';\nimport type { PageWord, Word } from '../geometry.ts';\nimport { foldTypographic } from '../lines.ts';\nimport { awaitPdfOpen, openPdfForRender } from '../pdf-open.ts';\nimport { renderLoadedPage } from '../raster.ts';\nimport { resolvePackagePath } from '../resolve.ts';\n\n// 288 dpi: the render resolution the engine's word-overlap numbers were\n// measured at (test/fixtures/ocr-engine-poc/) — the known-good operating point.\nexport const OCR_RENDER_DPI = 288;\n\n// Chart pages starve PSM.AUTO (unrecognized text can't be marked for review).\n// Prose-fraction retry trigger, measured (sparse-retry.test.ts): chart 0.442 vs prose 0.767-0.938, cut 0.6.\nexport const SPARSE_RETRY_PROSE_FRACTION = 0.6;\n\n/** Share of `words` whose text contains a run of 3+ letters (`src/emit.ts`'s isUnstructured canary, applied per-page instead of per-block). */\nfunction proseFraction(words: { text: string }[]): number {\n if (words.length === 0) return 0;\n return words.filter((w) => /[A-Za-z]{3}/.test(w.text)).length / words.length;\n}\n\n/** Does this page's PSM.AUTO pass look chart-like enough to re-recognize\n * with PSM.SPARSE_TEXT? A page with 0 words never retries (nothing to take a prose fraction of). */\nexport function shouldRetrySparse(words: { text: string }[]): boolean {\n return words.length > 0 && proseFraction(words) < SPARSE_RETRY_PROSE_FRACTION;\n}\n\n// The locally installed traineddata (npm package @tesseract.js-data/eng),\n// resolved dynamically so depcheck won't flag it \"unused\" — it's a genuine runtime dependency, listed in .depcheckrc's `ignores`.\nconst TESSDATA_DIR = path.join(path.dirname(resolvePackagePath('@tesseract.js-data/eng/package.json')), '4.0.0');\n\n/** Per-page recognition progress, delivered once per page as it finishes.\n * Engine chatter goes through `onWarning`/`warnings` instead — this is timing/count data only. */\nexport type OcrProgressEvent = { type: 'page'; page: number; words: number; ms: number };\n\nexport interface RecognizeResult {\n words: PageWord[];\n /** pages (1-based) where the engine returned zero words */\n emptyPages: number[];\n /** wall-clock ms per page (render + recognize), keyed by page number */\n msPerPage: Map<number, number>;\n}\n\n/** tesseract.js's nested blocks/paragraphs/lines/words -> this repo's flat Word contract (normalize + y-flip). Same shape regardless of which PSM produced `page`, so both the AUTO pass and the SPARSE_TEXT retry share it. */\nfunction wordsFromPage(page: TesseractPage, width: number, height: number): Word[] {\n const pageWords: Word[] = [];\n for (const block of page.blocks ?? []) {\n for (const para of block.paragraphs ?? []) {\n for (const line of para.lines ?? []) {\n for (const w of line.words ?? []) {\n const { x0, y0, x1, y1 } = w.bbox;\n pageWords.push({\n // Same typographic folding as the text path (src/lines.ts) — both\n // paths must agree on how a printed apostrophe is spelled.\n text: foldTypographic(w.text),\n x: x0 / width,\n y: (height - y1) / height, // y-flip: top-left px -> bottom-left normalized\n w: (x1 - x0) / width,\n h: (y1 - y0) / height,\n // Plumbed through unmodified — tesseract.js's own 0-100\n // per-word confidence. See src/report.ts's LOW_CONFIDENCE_THRESHOLD.\n confidence: w.confidence,\n });\n }\n }\n }\n }\n return pageWords;\n}\n\n// tesseract.js-core's WASM build writes native diagnostics straight to this\n// process's stdout/stderr, with no override to silence them at the source.\n//\n// Suppression must span the whole run: worker-thread stdio arrives\n// asynchronously, on later event-loop turns; the synchronous carve-out below (`withRealStdio`) is the only safe exit.\nlet suppressDepth = 0;\nlet savedStdoutWrite: typeof process.stdout.write | null = null;\nlet savedStderrWrite: typeof process.stderr.write | null = null;\nasync function withSuppressedStdio<T>(fn: () => Promise<T>): Promise<T> {\n if (suppressDepth === 0) {\n savedStdoutWrite = process.stdout.write.bind(process.stdout);\n savedStderrWrite = process.stderr.write.bind(process.stderr);\n process.stdout.write = (() => true) as typeof process.stdout.write;\n process.stderr.write = (() => true) as typeof process.stderr.write;\n }\n suppressDepth++;\n try {\n return await fn();\n } finally {\n suppressDepth--;\n if (suppressDepth === 0) {\n if (savedStdoutWrite) process.stdout.write = savedStdoutWrite;\n if (savedStderrWrite) process.stderr.write = savedStderrWrite;\n savedStdoutWrite = null;\n savedStderrWrite = null;\n }\n }\n}\n\n/** Pierces an active `withSuppressedStdio` region SYNCHRONOUSLY to let this\n * module's own onProgress/onWarning calls through. No-op outside any suppressed region. */\nfunction withRealStdio<T>(fn: () => T): T {\n if (suppressDepth === 0) return fn();\n const realOut = savedStdoutWrite;\n const realErr = savedStderrWrite;\n if (realOut) process.stdout.write = realOut;\n if (realErr) process.stderr.write = realErr;\n try {\n return fn();\n } finally {\n process.stdout.write = (() => true) as typeof process.stdout.write;\n process.stderr.write = (() => true) as typeof process.stderr.write;\n }\n}\n\n/** Recognize words on pages [first..last] of a PDF (renders at `dpi`,\n * default OCR_RENDER_DPI). Progress via `onProgress`; diagnostics via `onWarning` — this module never writes to stdout/stderr itself. */\nexport async function recognizePdf(pdfPath: string, first: number, last: number, { dpi = OCR_RENDER_DPI, onProgress, onWarning }: { dpi?: number; onProgress?: (e: OcrProgressEvent) => void; onWarning?: (message: string) => void } = {}): Promise<RecognizeResult> {\n // tesseract.js writes its decompressed traineddata cache here by default,\n // dropping a model file into the user's cwd — redirect it under tmpdir() instead.\n const cachePath = path.join(os.tmpdir(), 'pdf-to-md-engine');\n mkdirSync(cachePath, { recursive: true });\n\n const words: PageWord[] = [];\n const emptyPages: number[] = [];\n const msPerPage = new Map<number, number>();\n\n // ONE continuous suppressed region for the whole run; onProgress/onWarning\n // below go through `withRealStdio`, the only sanctioned way out of it.\n await withSuppressedStdio(async () => {\n // OEM 1 (LSTM only, tesseract.js's own default): measured 99.0% recall\n // on the hard benchmark pages (2026-08-16).\n const worker = await createWorker('eng', 1, {\n langPath: TESSDATA_DIR,\n cachePath,\n });\n try {\n // NOT optional: tesseract.js's own default PSM (SINGLE_BLOCK) measured\n // 3-8 percentage points worse than AUTO on two-column pages.\n await worker.setParameters({ tessedit_pageseg_mode: PSM.AUTO });\n\n const loadingTask = openPdfForRender(pdfPath);\n try {\n const pdf = await awaitPdfOpen(loadingTask, pdfPath);\n for (let pn = first; pn <= Math.min(last, pdf.numPages); pn++) {\n const t0 = Date.now();\n const { png, width, height } = await renderLoadedPage(pdf, pn, dpi);\n const { data } = await worker.recognize(png, {}, { blocks: true });\n let pageWords: Word[] = wordsFromPage(data, width, height);\n\n // Adaptive second pass: re-recognizes the SAME rendered PNG, so\n // this costs one extra recognize() call only on chart-like pages.\n if (shouldRetrySparse(pageWords)) {\n const autoFraction = proseFraction(pageWords);\n const autoWords = pageWords;\n // setParameters on the shared worker is reversible, so one\n // in-process worker can alternate PSM per page instead of paying for a second worker.\n await worker.setParameters({ tessedit_pageseg_mode: PSM.SPARSE_TEXT });\n const sparse = await worker.recognize(png, {}, { blocks: true });\n const sparseWords = wordsFromPage(sparse.data, width, height);\n await worker.setParameters({ tessedit_pageseg_mode: PSM.AUTO });\n\n if (sparseWords.length > autoWords.length) {\n pageWords = sparseWords;\n const autoCount = autoWords.length;\n const sparseCount = sparseWords.length;\n withRealStdio(() => onWarning?.(`p${pn}: chart-like page (prose fraction ${autoFraction.toFixed(2)}) — retried with sparse-text segmentation, ${autoCount} -> ${sparseCount} words`));\n }\n }\n\n const ms = Date.now() - t0;\n for (const w of pageWords) words.push({ page: pn, ...w });\n if (pageWords.length === 0) emptyPages.push(pn);\n msPerPage.set(pn, ms);\n const wordCount = pageWords.length;\n withRealStdio(() => onProgress?.({ type: 'page', page: pn, words: wordCount, ms }));\n }\n } finally {\n await loadingTask.destroy();\n }\n } finally {\n await worker.terminate();\n }\n });\n\n return { words, emptyPages, msPerPage };\n}\n"],"names":["mkdirSync","os","path","createWorker","PSM","foldTypographic","awaitPdfOpen","openPdfForRender","renderLoadedPage","resolvePackagePath","OCR_RENDER_DPI","SPARSE_RETRY_PROSE_FRACTION","proseFraction","words","length","filter","w","test","text","shouldRetrySparse","TESSDATA_DIR","join","dirname","wordsFromPage","page","width","height","pageWords","block","blocks","para","paragraphs","line","lines","x0","y0","x1","y1","bbox","push","x","y","h","confidence","suppressDepth","savedStdoutWrite","savedStderrWrite","withSuppressedStdio","fn","process","stdout","write","bind","stderr","withRealStdio","realOut","realErr","recognizePdf","pdfPath","first","last","dpi","onProgress","onWarning","cachePath","tmpdir","recursive","emptyPages","msPerPage","Map","worker","langPath","setParameters","tessedit_pageseg_mode","AUTO","loadingTask","pdf","pn","Math","min","numPages","t0","Date","now","png","data","recognize","autoFraction","autoWords","SPARSE_TEXT","sparse","sparseWords","autoCount","sparseCount","toFixed","ms","set","wordCount","type","destroy","terminate"],"mappings":"AAAA,yEAAyE;AACzE,EAAE;AACF,4EAA4E;AAC5E,gFAAgF;AAChF,EAAE;AACF,4EAA4E;AAC5E,SAASA,SAAS,QAAQ,UAAU;AACpC,OAAOC,QAAQ,UAAU;AACzB,OAAOC,UAAU,YAAY;AAE7B,SAASC,YAAY,EAAEC,GAAG,QAAQ,eAAe;AAEjD,SAASC,eAAe,QAAQ,cAAc;AAC9C,SAASC,YAAY,EAAEC,gBAAgB,QAAQ,iBAAiB;AAChE,SAASC,gBAAgB,QAAQ,eAAe;AAChD,SAASC,kBAAkB,QAAQ,gBAAgB;AAEnD,wEAAwE;AACxE,gFAAgF;AAChF,OAAO,MAAMC,iBAAiB,IAAI;AAElC,8EAA8E;AAC9E,4GAA4G;AAC5G,OAAO,MAAMC,8BAA8B,IAAI;AAE/C,6IAA6I,GAC7I,SAASC,cAAcC,KAAyB;IAC9C,IAAIA,MAAMC,MAAM,KAAK,GAAG,OAAO;IAC/B,OAAOD,MAAME,MAAM,CAAC,CAACC,IAAM,cAAcC,IAAI,CAACD,EAAEE,IAAI,GAAGJ,MAAM,GAAGD,MAAMC,MAAM;AAC9E;AAEA;kGACkG,GAClG,OAAO,SAASK,kBAAkBN,KAAyB;IACzD,OAAOA,MAAMC,MAAM,GAAG,KAAKF,cAAcC,SAASF;AACpD;AAEA,0EAA0E;AAC1E,kIAAkI;AAClI,MAAMS,eAAelB,KAAKmB,IAAI,CAACnB,KAAKoB,OAAO,CAACb,mBAAmB,yCAAyC;AAcxG,8NAA8N,GAC9N,SAASc,cAAcC,IAAmB,EAAEC,KAAa,EAAEC,MAAc;QAEnDF;IADpB,MAAMG,YAAoB,EAAE;IAC5B,KAAK,MAAMC,UAASJ,eAAAA,KAAKK,MAAM,cAAXL,0BAAAA,eAAe,EAAE,CAAE;YAClBI;QAAnB,KAAK,MAAME,SAAQF,oBAAAA,MAAMG,UAAU,cAAhBH,+BAAAA,oBAAoB,EAAE,CAAE;gBACtBE;YAAnB,KAAK,MAAME,SAAQF,cAAAA,KAAKG,KAAK,cAAVH,yBAAAA,cAAc,EAAE,CAAE;oBACnBE;gBAAhB,KAAK,MAAMhB,MAAKgB,cAAAA,KAAKnB,KAAK,cAAVmB,yBAAAA,cAAc,EAAE,CAAE;oBAChC,MAAM,EAAEE,EAAE,EAAEC,EAAE,EAAEC,EAAE,EAAEC,EAAE,EAAE,GAAGrB,EAAEsB,IAAI;oBACjCX,UAAUY,IAAI,CAAC;wBACb,kEAAkE;wBAClE,2DAA2D;wBAC3DrB,MAAMb,gBAAgBW,EAAEE,IAAI;wBAC5BsB,GAAGN,KAAKT;wBACRgB,GAAG,AAACf,CAAAA,SAASW,EAAC,IAAKX;wBACnBV,GAAG,AAACoB,CAAAA,KAAKF,EAAC,IAAKT;wBACfiB,GAAG,AAACL,CAAAA,KAAKF,EAAC,IAAKT;wBACf,wDAAwD;wBACxD,qEAAqE;wBACrEiB,YAAY3B,EAAE2B,UAAU;oBAC1B;gBACF;YACF;QACF;IACF;IACA,OAAOhB;AACT;AAEA,4EAA4E;AAC5E,2EAA2E;AAC3E,EAAE;AACF,mEAAmE;AACnE,sHAAsH;AACtH,IAAIiB,gBAAgB;AACpB,IAAIC,mBAAuD;AAC3D,IAAIC,mBAAuD;AAC3D,eAAeC,oBAAuBC,EAAoB;IACxD,IAAIJ,kBAAkB,GAAG;QACvBC,mBAAmBI,QAAQC,MAAM,CAACC,KAAK,CAACC,IAAI,CAACH,QAAQC,MAAM;QAC3DJ,mBAAmBG,QAAQI,MAAM,CAACF,KAAK,CAACC,IAAI,CAACH,QAAQI,MAAM;QAC3DJ,QAAQC,MAAM,CAACC,KAAK,GAAI,IAAM;QAC9BF,QAAQI,MAAM,CAACF,KAAK,GAAI,IAAM;IAChC;IACAP;IACA,IAAI;QACF,OAAO,MAAMI;IACf,SAAU;QACRJ;QACA,IAAIA,kBAAkB,GAAG;YACvB,IAAIC,kBAAkBI,QAAQC,MAAM,CAACC,KAAK,GAAGN;YAC7C,IAAIC,kBAAkBG,QAAQI,MAAM,CAACF,KAAK,GAAGL;YAC7CD,mBAAmB;YACnBC,mBAAmB;QACrB;IACF;AACF;AAEA;yFACyF,GACzF,SAASQ,cAAiBN,EAAW;IACnC,IAAIJ,kBAAkB,GAAG,OAAOI;IAChC,MAAMO,UAAUV;IAChB,MAAMW,UAAUV;IAChB,IAAIS,SAASN,QAAQC,MAAM,CAACC,KAAK,GAAGI;IACpC,IAAIC,SAASP,QAAQI,MAAM,CAACF,KAAK,GAAGK;IACpC,IAAI;QACF,OAAOR;IACT,SAAU;QACRC,QAAQC,MAAM,CAACC,KAAK,GAAI,IAAM;QAC9BF,QAAQI,MAAM,CAACF,KAAK,GAAI,IAAM;IAChC;AACF;AAEA;uIACuI,GACvI,OAAO,eAAeM,aAAaC,OAAe,EAAEC,KAAa,EAAEC,IAAY,EAAE,EAAEC,MAAMnD,cAAc,EAAEoD,UAAU,EAAEC,SAAS,EAAuG,GAAG,CAAC,CAAC;IACxO,0EAA0E;IAC1E,kFAAkF;IAClF,MAAMC,YAAY9D,KAAKmB,IAAI,CAACpB,GAAGgE,MAAM,IAAI;IACzCjE,UAAUgE,WAAW;QAAEE,WAAW;IAAK;IAEvC,MAAMrD,QAAoB,EAAE;IAC5B,MAAMsD,aAAuB,EAAE;IAC/B,MAAMC,YAAY,IAAIC;IAEtB,2EAA2E;IAC3E,uEAAuE;IACvE,MAAMtB,oBAAoB;QACxB,uEAAuE;QACvE,4CAA4C;QAC5C,MAAMuB,SAAS,MAAMnE,aAAa,OAAO,GAAG;YAC1CoE,UAAUnD;YACV4C;QACF;QACA,IAAI;YACF,uEAAuE;YACvE,6DAA6D;YAC7D,MAAMM,OAAOE,aAAa,CAAC;gBAAEC,uBAAuBrE,IAAIsE,IAAI;YAAC;YAE7D,MAAMC,cAAcpE,iBAAiBmD;YACrC,IAAI;gBACF,MAAMkB,MAAM,MAAMtE,aAAaqE,aAAajB;gBAC5C,IAAK,IAAImB,KAAKlB,OAAOkB,MAAMC,KAAKC,GAAG,CAACnB,MAAMgB,IAAII,QAAQ,GAAGH,KAAM;oBAC7D,MAAMI,KAAKC,KAAKC,GAAG;oBACnB,MAAM,EAAEC,GAAG,EAAE3D,KAAK,EAAEC,MAAM,EAAE,GAAG,MAAMlB,iBAAiBoE,KAAKC,IAAIhB;oBAC/D,MAAM,EAAEwB,IAAI,EAAE,GAAG,MAAMf,OAAOgB,SAAS,CAACF,KAAK,CAAC,GAAG;wBAAEvD,QAAQ;oBAAK;oBAChE,IAAIF,YAAoBJ,cAAc8D,MAAM5D,OAAOC;oBAEnD,gEAAgE;oBAChE,kEAAkE;oBAClE,IAAIP,kBAAkBQ,YAAY;wBAChC,MAAM4D,eAAe3E,cAAce;wBACnC,MAAM6D,YAAY7D;wBAClB,2DAA2D;wBAC3D,sFAAsF;wBACtF,MAAM2C,OAAOE,aAAa,CAAC;4BAAEC,uBAAuBrE,IAAIqF,WAAW;wBAAC;wBACpE,MAAMC,SAAS,MAAMpB,OAAOgB,SAAS,CAACF,KAAK,CAAC,GAAG;4BAAEvD,QAAQ;wBAAK;wBAC9D,MAAM8D,cAAcpE,cAAcmE,OAAOL,IAAI,EAAE5D,OAAOC;wBACtD,MAAM4C,OAAOE,aAAa,CAAC;4BAAEC,uBAAuBrE,IAAIsE,IAAI;wBAAC;wBAE7D,IAAIiB,YAAY7E,MAAM,GAAG0E,UAAU1E,MAAM,EAAE;4BACzCa,YAAYgE;4BACZ,MAAMC,YAAYJ,UAAU1E,MAAM;4BAClC,MAAM+E,cAAcF,YAAY7E,MAAM;4BACtCwC,cAAc,IAAMS,sBAAAA,gCAAAA,UAAY,CAAC,CAAC,EAAEc,GAAG,kCAAkC,EAAEU,aAAaO,OAAO,CAAC,GAAG,2CAA2C,EAAEF,UAAU,IAAI,EAAEC,YAAY,MAAM,CAAC;wBACrL;oBACF;oBAEA,MAAME,KAAKb,KAAKC,GAAG,KAAKF;oBACxB,KAAK,MAAMjE,KAAKW,UAAWd,MAAM0B,IAAI,CAAC;wBAAEf,MAAMqD;wBAAI,GAAG7D,CAAC;oBAAC;oBACvD,IAAIW,UAAUb,MAAM,KAAK,GAAGqD,WAAW5B,IAAI,CAACsC;oBAC5CT,UAAU4B,GAAG,CAACnB,IAAIkB;oBAClB,MAAME,YAAYtE,UAAUb,MAAM;oBAClCwC,cAAc,IAAMQ,uBAAAA,iCAAAA,WAAa;4BAAEoC,MAAM;4BAAQ1E,MAAMqD;4BAAIhE,OAAOoF;4BAAWF;wBAAG;gBAClF;YACF,SAAU;gBACR,MAAMpB,YAAYwB,OAAO;YAC3B;QACF,SAAU;YACR,MAAM7B,OAAO8B,SAAS;QACxB;IACF;IAEA,OAAO;QAAEvF;QAAOsD;QAAYC;IAAU;AACxC"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/errors.ts"],"sourcesContent":["// Library code throws; only src/cli.ts prints and exits.\n\n/** The closed set of error codes every `PdfToMdError` carries — switch on\n * `code` instead of parsing `message`. */\nexport type PdfToMdErrorCode =\n /** input missing/unreadable/not a PDF */\n | 'PDF_OPEN'\n /** malformed page selection (validated before any work) */\n | 'PAGE_RANGE'\n /** rendering an opened, in-range page to PNG failed */\n | 'PAGE_RENDER'\n /** the OCR engine crashed twice on one page (message names it) */\n | 'OCR_PAGE_FAILED'\n /** malformed word-dump input (--words-json / extractOcr words) */\n | 'WORDS_INPUT'\n /** malformed or insufficient evidence passed to auditWords */\n | 'AUDIT_INPUT'\n /** malformed Analysis handed to toMarkdown/toText (e.g. an out-of-range\n * wordIndexes entry) — validated fail-fast, whole object checked before any output. */\n | 'ANALYSIS_INPUT';\n\n/** The error every entry point throws — never a bare `Error`. */\nexport class PdfToMdError extends Error {\n code: PdfToMdErrorCode;\n\n constructor(code: PdfToMdErrorCode, message: string) {\n super(message);\n this.name = 'PdfToMdError';\n this.code = code;\n }\n}\n"],"names":["PdfToMdError","Error","code","message","name"],"mappings":"AAAA,yDAAyD;AAEzD;wCACwC,GAkBxC,+DAA+D,GAC/D,OAAO,MAAMA,qBAAqBC;IAGhC,YAAYC,IAAsB,EAAEC,OAAe,CAAE;QACnD,KAAK,CAACA;QACN,IAAI,CAACC,IAAI,GAAG;QACZ,IAAI,CAACF,IAAI,GAAGA;IACd;AACF"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/errors.ts"],"sourcesContent":["// Library code throws; only src/cli.ts prints and exits.\n\n/** The closed set of error codes every `PdfToMdError` carries — switch on\n * `code` instead of parsing `message`. */\nexport type PdfToMdErrorCode =\n /** input missing/unreadable/not a PDF */\n | 'PDF_OPEN'\n /** malformed page selection (validated before any work) */\n | 'PAGE_RANGE'\n /** rendering an opened, in-range page to PNG failed */\n | 'PAGE_RENDER'\n /** the OCR engine crashed twice on one page (message names it) */\n | 'OCR_PAGE_FAILED'\n /** malformed word-dump input (--words-json / extractOcr words) */\n | 'WORDS_INPUT'\n /** malformed or insufficient evidence passed to auditWords */\n | 'AUDIT_INPUT'\n /** malformed Analysis handed to toMarkdown/toText (e.g. an out-of-range\n * wordIndexes entry) — validated fail-fast, whole object checked before any output. */\n | 'ANALYSIS_INPUT';\n\n/** The error every entry point throws — never a bare `Error`. */\nexport class PdfToMdError extends Error {\n code: PdfToMdErrorCode;\n\n constructor(code: PdfToMdErrorCode, message: string) {\n super(message);\n this.name = 'PdfToMdError';\n this.code = code;\n }\n}\n"],"names":["PdfToMdError","Error","code","message","name"],"mappings":"AAAA,yDAAyD;AAEzD;wCACwC,GAkBxC,+DAA+D,GAC/D,OAAO,MAAMA,qBAAqBC;IAGhC,YAAYC,IAAsB,EAAEC,OAAe,CAAE;QACnD,KAAK,CAACA;QACN,IAAI,CAACC,IAAI,GAAG;QACZ,IAAI,CAACF,IAAI,GAAGA;IACd;AACF"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/extract.ts"],"sourcesContent":["// The library layer for PDF -> markdown conversion: extractText/extractOcr/\n// pdfToMarkdown each take plain args, return plain data, and throw PdfToMdError — no process.exit, no console.*, no file writing.\n//\n// Warnings are DATA: every notice is returned in `warnings` verbatim and also\n// delivered incrementally via `onWarning` — the exact strings are load-bearing (the test suite asserts them).\nimport { buildOcrAnalysisPages } from './analyze.ts';\nimport { collect } from './collect.ts';\nimport { dictAvailable, dictUnavailableWarning } from './dict.ts';\nimport type { OcrProgressEvent } from './engines/tesseract.ts';\nimport type { PageWord, Tuning } from './geometry.ts';\nimport { renderPagesToMarkdown } from './render-analysis.ts';\nimport { AUTO_OCR_CHARS_PER_PAGE, buildOcrReport } from './report.ts';\nimport { buildTextAnalysis, selectCollectedPages } from './text-analysis.ts';\nimport type { Analysis, AnalysisPage, CollectResult, Report } from './types.ts';\nimport { validatePages } from './validate.ts';\n\nexport type { OcrProgressEvent } from './engines/tesseract.ts';\n\n// A plain-JS caller (or `.ts` with `as any`) gets no compile-time check on\n// `pages`, so it's validated once at runtime — see src/validate.ts's `validatePages`.\n\n/** One recognized word with its page — the OCR replay input contract (same\n * shape as `geometry.ts`'s `PageWord`, re-exported under this module's public name). */\nexport type OcrWordInput = PageWord;\n\n/** A 1-based, inclusive page range — the shape every entry point's `pages`\n * option takes (the CLI's own `--pages N[-M]` parses into this). */\nexport interface PageSelection {\n first: number;\n /** inclusive; defaults to `first` */\n last?: number;\n}\n\n/** extractText/extractOcr/pdfToMarkdown's common return shape. */\nexport interface ExtractResult {\n markdown: string;\n /** per-document and per-page QA metrics — see `Report` (`src/types.ts`) */\n report: Report;\n /** every notice also delivered incrementally via `onWarning`, collected\n * here in full (hybrid pages, auto-OCR-fallback, empty OCR pages, sparse-retry) */\n warnings: string[];\n}\n\ninterface CommonOptions {\n /** convert only this page range (1-based, inclusive); defaults to the\n * whole document */\n pages?: PageSelection;\n /** delivered as each warning happens (also returned, in full, as\n * `warnings` on the resolved result) */\n onWarning?: (message: string) => void;\n}\n\n/** Options for {@link extractText}. */\nexport interface ExtractTextOptions extends CommonOptions {\n /** emit a `### pN` marker before each page's content (output formatting\n * only — `report.pageStats` is always populated regardless) */\n pageMarkers?: boolean;\n /** source PDF text runs entering layout, delivered once for the selected\n * final result. Runs retain measured bounds and are not fabricated words. */\n onWords?: (words: OcrWordInput[]) => void;\n}\n\n/** Options for {@link extractOcr}. */\nexport interface ExtractOcrOptions extends CommonOptions {\n /** override any subset of the OCR-path geometry defaults — see\n * DEFAULT_TUNING (`src/geometry.ts`) for each knob's measured default */\n tuning?: Partial<Tuning>;\n /** page-render resolution for engine recognition (default 288, measured;\n * >~400 measured WORSE on ordinary pages — not a \"bigger is safer\" knob) */\n dpi?: number;\n /** marks report.ocrFallback — set by pdfToMarkdown's auto-fallback */\n fallback?: boolean;\n /** one event per page as recognition finishes (timing/count only — engine\n * chatter and the sparse-retry notice arrive via `onWarning` instead) */\n onProgress?: (e: OcrProgressEvent) => void;\n /** every recognized word entering the geometry, delivered once after\n * recognition — the severity-1 audit instrument (--debug-words) */\n onWords?: (words: OcrWordInput[]) => void;\n}\n\n/** Options for {@link pdfToMarkdown}. */\nexport interface PdfToMarkdownOptions extends ExtractTextOptions, Omit<ExtractOcrOptions, 'fallback'> {\n /** force the OCR path even when a text layer exists */\n ocr?: boolean;\n /** never fall back to OCR, even for image-only pages */\n noOcr?: boolean;\n}\n\n// One collector per entry point: collects into `warnings`, streams via\n// onWarning, and emits the dictionary-unavailable notice exactly once up front.\nfunction warnCollector(onWarning?: (message: string) => void): { warnings: string[]; warn: (m: string) => void } {\n const warnings: string[] = [];\n const warn = (m: string) => {\n warnings.push(m);\n onWarning?.(m);\n };\n if (!dictAvailable()) warn(dictUnavailableWarning());\n return { warnings, warn };\n}\n\n/** Extract embedded text with a QA report and warnings; never runs OCR.\n * @param pdfPath path to the PDF file\n * @param opts see {@link ExtractTextOptions}\n */\nexport async function extractText(pdfPath: string, opts: ExtractTextOptions = {}): Promise<ExtractResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n const range = opts.pages ? { first: opts.pages.first, last: opts.pages.last ?? opts.pages.first } : {};\n const collected = await collect(pdfPath, range);\n const text = buildTextAnalysis(collected, opts.pageMarkers ?? false);\n textNotices(collected, new Set(), warn);\n opts.onWords?.(text.words);\n return { markdown: text.markdown, report: text.analysis.report, warnings };\n}\n\n/**\n * OCR extraction. Input is either a PDF (pages render at `dpi` and the\n * engine recognizes them, imported lazily) or an already-recognized word dump (the offline replay input — see `--words-json`).\n *\n * @param input `{ pdfPath }` to recognize, or `{ words }` to replay a dump\n * @param opts see {@link ExtractOcrOptions}\n */\nexport async function extractOcr(input: { pdfPath: string } | { words: OcrWordInput[] }, opts: ExtractOcrOptions = {}): Promise<ExtractResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n\n // Shared with analyze(pdf, {path:'ocr'}) — see that module's header for\n // why this makes the two byte-identical by construction, not by hand.\n const { pages, words } = await buildOcrAnalysisPages(input, opts, warn);\n const md = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n\n const report = buildOcrReport({\n md,\n fallback: opts.fallback,\n // Confidence is on `words` — see report.ts's low-confidence detector for\n // how it handles a confidence-less replay honestly.\n words,\n });\n return { markdown: md, report, warnings };\n}\n\n/** A page needs OCR only when its image signal accompanies an unusable text\n * layer. The caller supplies the image part of that decision. */\nfunction needsPageOcr(chars: number): boolean {\n return chars < AUTO_OCR_CHARS_PER_PAGE;\n}\n\nfunction textNotices(collected: CollectResult, ocrPages: Set<number>, warn: (message: string) => void): void {\n for (const page of collected.pages) {\n if (page.textChars !== 0 || ocrPages.has(page.page)) continue;\n if (page.hasImage) warn(`p${page.page}: no extractable text; page has image content and OCR was not selected`);\n else warn(`p${page.page}: no extractable text; the page may be blank or contain only vector content`);\n }\n for (const page of collected.pages) {\n if (!page.largeImage || ocrPages.has(page.page)) continue;\n warn(`p${page.page}: text layer used but page carries large image(s); if it has baked-in text this output won't have it — compare (don't blindly replace) against --pages ${page.page} --ocr, which can itself introduce errors`);\n }\n}\n\nfunction analysisWords(pages: AnalysisPage[]): OcrWordInput[] {\n return pages.flatMap((page) => page.words.map((word) => ({ page: page.page, text: word.text, ...word.box, confidence: word.confidence })));\n}\n\ninterface AutoAnalysisResult {\n analysis: Analysis;\n markdown: string;\n warnings: string[];\n}\n\n/** Shared automatic route for the library's Markdown result and the CLI's\n * raw/plain-text formats. A page changes producer only when it has image\n * evidence and an unusable text layer, so a blank or vector-only page never\n * becomes speculative OCR input. */\nexport async function analyzePdfAuto(pdfPath: string, opts: PdfToMarkdownOptions = {}): Promise<AutoAnalysisResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n const range = opts.pages ? { first: opts.pages.first, last: opts.pages.last ?? opts.pages.first } : {};\n\n if (opts.ocr) {\n const { pages, words } = await buildOcrAnalysisPages({ pdfPath }, { ...opts, onWords: undefined }, warn);\n const markdown = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n const report = buildOcrReport({ md: markdown, words });\n const analysis: Analysis = { outputVersion: report.outputVersion, report, pages };\n opts.onWords?.(words);\n return { analysis, markdown, warnings };\n }\n\n const collected = await collect(pdfPath, range);\n const ocrPageNumbers = new Set(!opts.noOcr ? collected.pages.filter((page) => page.hasImage && needsPageOcr(page.textChars)).map((page) => page.page) : []);\n if (!ocrPageNumbers.size) {\n const text = buildTextAnalysis(collected, opts.pageMarkers ?? false);\n textNotices(collected, ocrPageNumbers, warn);\n opts.onWords?.(text.words);\n return { analysis: text.analysis, markdown: text.markdown, warnings };\n }\n\n const textPageNumbers = new Set(collected.pages.filter((page) => !ocrPageNumbers.has(page.page)).map((page) => page.page));\n const text = textPageNumbers.size ? buildTextAnalysis(selectCollectedPages(collected, textPageNumbers), true) : undefined;\n const textByPage = new Map(text?.analysis.pages.map((page) => [page.page, page]));\n const ocrByPage = new Map<number, AnalysisPage>();\n const ocrWords: OcrWordInput[] = [];\n const ocrRanges: Array<{ first: number; last: number }> = [];\n for (const page of collected.pages) {\n if (!ocrPageNumbers.has(page.page)) continue;\n warn(`p${page.page}: sparse text layer with image content; using OCR for this page`);\n const lastRange = ocrRanges.at(-1);\n if (lastRange && lastRange.last + 1 === page.page) lastRange.last = page.page;\n else ocrRanges.push({ first: page.page, last: page.page });\n }\n for (const range of ocrRanges) {\n const result = await buildOcrAnalysisPages({ pdfPath }, { ...opts, pages: range, onWords: undefined }, warn);\n for (const page of result.pages) ocrByPage.set(page.page, page);\n ocrWords.push(...result.words);\n }\n textNotices(collected, ocrPageNumbers, warn);\n const pages = collected.pages.map((page) => ocrByPage.get(page.page) ?? (textByPage.get(page.page) as AnalysisPage));\n const markdown = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n const pagePaths = new Map(pages.map((page) => [page.page, page.path]));\n const report = buildOcrReport({\n md: markdown,\n fallback: true,\n words: ocrWords,\n pagePaths,\n hybridPages: text?.analysis.report.hybridPages,\n });\n if (text) {\n report.bodyFontHeight = text.analysis.report.bodyFontHeight;\n report.headingSizes = text.analysis.report.headingSizes;\n report.listItems = text.analysis.report.listItems;\n report.joinedPairs = text.analysis.report.joinedPairs;\n report.closedHyphens = text.analysis.report.closedHyphens;\n report.hybridPages = text.analysis.report.hybridPages;\n }\n const analysis: Analysis = { outputVersion: report.outputVersion, report, pages };\n opts.onWords?.(analysisWords(pages));\n return { analysis, markdown, warnings };\n}\n\n/**\n * Extract each page from embedded text, using OCR only when the page has\n * image content and fewer than `AUTO_OCR_CHARS_PER_PAGE` text characters.\n *\n * @param pdfPath path to the PDF file\n * @param opts see {@link PdfToMarkdownOptions}\n */\nexport async function pdfToMarkdown(pdfPath: string, opts: PdfToMarkdownOptions = {}): Promise<ExtractResult> {\n const result = await analyzePdfAuto(pdfPath, opts);\n return { markdown: result.markdown, report: result.analysis.report, warnings: result.warnings };\n}\n"],"names":["buildOcrAnalysisPages","collect","dictAvailable","dictUnavailableWarning","renderPagesToMarkdown","AUTO_OCR_CHARS_PER_PAGE","buildOcrReport","buildTextAnalysis","selectCollectedPages","validatePages","warnCollector","onWarning","warnings","warn","m","push","extractText","pdfPath","opts","pages","range","first","last","collected","text","pageMarkers","textNotices","Set","onWords","words","markdown","report","analysis","extractOcr","input","md","forcePageMarkers","fallback","needsPageOcr","chars","ocrPages","page","textChars","has","hasImage","largeImage","analysisWords","flatMap","map","word","box","confidence","analyzePdfAuto","ocr","undefined","outputVersion","ocrPageNumbers","noOcr","filter","size","textPageNumbers","textByPage","Map","ocrByPage","ocrWords","ocrRanges","lastRange","at","result","set","get","pagePaths","path","hybridPages","bodyFontHeight","headingSizes","listItems","joinedPairs","closedHyphens","pdfToMarkdown"],"mappings":"AAAA,4EAA4E;AAC5E,kIAAkI;AAClI,EAAE;AACF,8EAA8E;AAC9E,8GAA8G;AAC9G,SAASA,qBAAqB,QAAQ,eAAe;AACrD,SAASC,OAAO,QAAQ,eAAe;AACvC,SAASC,aAAa,EAAEC,sBAAsB,QAAQ,YAAY;AAGlE,SAASC,qBAAqB,QAAQ,uBAAuB;AAC7D,SAASC,uBAAuB,EAAEC,cAAc,QAAQ,cAAc;AACtE,SAASC,iBAAiB,EAAEC,oBAAoB,QAAQ,qBAAqB;AAE7E,SAASC,aAAa,QAAQ,gBAAgB;AA0E9C,uEAAuE;AACvE,gFAAgF;AAChF,SAASC,cAAcC,SAAqC;IAC1D,MAAMC,WAAqB,EAAE;IAC7B,MAAMC,OAAO,CAACC;QACZF,SAASG,IAAI,CAACD;QACdH,sBAAAA,gCAAAA,UAAYG;IACd;IACA,IAAI,CAACZ,iBAAiBW,KAAKV;IAC3B,OAAO;QAAES;QAAUC;IAAK;AAC1B;AAEA;;;CAGC,GACD,OAAO,eAAeG,YAAYC,OAAe,EAAEC,OAA2B,CAAC,CAAC;QAGlBA,kBAElBA;QAE1CA;IANAT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IACvD,MAAMS,QAAQF,KAAKC,KAAK,GAAG;QAAEE,OAAOH,KAAKC,KAAK,CAACE,KAAK;QAAEC,IAAI,GAAEJ,mBAAAA,KAAKC,KAAK,CAACG,IAAI,cAAfJ,8BAAAA,mBAAmBA,KAAKC,KAAK,CAACE,KAAK;IAAC,IAAI,CAAC;IACrG,MAAME,YAAY,MAAMtB,QAAQgB,SAASG;IACzC,MAAMI,OAAOjB,kBAAkBgB,YAAWL,oBAAAA,KAAKO,WAAW,cAAhBP,+BAAAA,oBAAoB;IAC9DQ,YAAYH,WAAW,IAAII,OAAOd;KAClCK,gBAAAA,KAAKU,OAAO,cAAZV,oCAAAA,mBAAAA,MAAeM,KAAKK,KAAK;IACzB,OAAO;QAAEC,UAAUN,KAAKM,QAAQ;QAAEC,QAAQP,KAAKQ,QAAQ,CAACD,MAAM;QAAEnB;IAAS;AAC3E;AAEA;;;;;;CAMC,GACD,OAAO,eAAeqB,WAAWC,KAAsD,EAAEhB,OAA0B,CAAC,CAAC;IACnHT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IAEvD,wEAAwE;IACxE,sEAAsE;IACtE,MAAM,EAAEQ,KAAK,EAAEU,KAAK,EAAE,GAAG,MAAM7B,sBAAsBkC,OAAOhB,MAAML;IAClE,MAAMsB,KAAK/B,sBAAsBe,OAAO;QAAEiB,kBAAkB;QAAMX,aAAa;IAAK;IAEpF,MAAMM,SAASzB,eAAe;QAC5B6B;QACAE,UAAUnB,KAAKmB,QAAQ;QACvB,yEAAyE;QACzE,oDAAoD;QACpDR;IACF;IACA,OAAO;QAAEC,UAAUK;QAAIJ;QAAQnB;IAAS;AAC1C;AAEA;+DAC+D,GAC/D,SAAS0B,aAAaC,KAAa;IACjC,OAAOA,QAAQlC;AACjB;AAEA,SAASqB,YAAYH,SAAwB,EAAEiB,QAAqB,EAAE3B,IAA+B;IACnG,KAAK,MAAM4B,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAIsB,KAAKC,SAAS,KAAK,KAAKF,SAASG,GAAG,CAACF,KAAKA,IAAI,GAAG;QACrD,IAAIA,KAAKG,QAAQ,EAAE/B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,sEAAsE,CAAC;aACxG5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,2EAA2E,CAAC;IACtG;IACA,KAAK,MAAMA,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAI,CAACsB,KAAKI,UAAU,IAAIL,SAASG,GAAG,CAACF,KAAKA,IAAI,GAAG;QACjD5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,uJAAuJ,EAAEA,KAAKA,IAAI,CAAC,yCAAyC,CAAC;IAClO;AACF;AAEA,SAASK,cAAc3B,KAAqB;IAC1C,OAAOA,MAAM4B,OAAO,CAAC,CAACN,OAASA,KAAKZ,KAAK,CAACmB,GAAG,CAAC,CAACC,OAAU,CAAA;gBAAER,MAAMA,KAAKA,IAAI;gBAAEjB,MAAMyB,KAAKzB,IAAI;gBAAE,GAAGyB,KAAKC,GAAG;gBAAEC,YAAYF,KAAKE,UAAU;YAAC,CAAA;AACxI;AAQA;;;kCAGkC,GAClC,OAAO,eAAeC,eAAenC,OAAe,EAAEC,OAA6B,CAAC,CAAC;QAGvBA;QA0D5DA;IA5DAT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IACvD,MAAMS,QAAQF,KAAKC,KAAK,GAAG;QAAEE,OAAOH,KAAKC,KAAK,CAACE,KAAK;QAAEC,IAAI,GAAEJ,mBAAAA,KAAKC,KAAK,CAACG,IAAI,cAAfJ,8BAAAA,mBAAmBA,KAAKC,KAAK,CAACE,KAAK;IAAC,IAAI,CAAC;IAErG,IAAIH,KAAKmC,GAAG,EAAE;YAKZnC;QAJA,MAAM,EAAEC,KAAK,EAAEU,KAAK,EAAE,GAAG,MAAM7B,sBAAsB;YAAEiB;QAAQ,GAAG;YAAE,GAAGC,IAAI;YAAEU,SAAS0B;QAAU,GAAGzC;QACnG,MAAMiB,WAAW1B,sBAAsBe,OAAO;YAAEiB,kBAAkB;YAAMX,aAAa;QAAK;QAC1F,MAAMM,SAASzB,eAAe;YAAE6B,IAAIL;YAAUD;QAAM;QACpD,MAAMG,WAAqB;YAAEuB,eAAexB,OAAOwB,aAAa;YAAExB;YAAQZ;QAAM;SAChFD,iBAAAA,KAAKU,OAAO,cAAZV,qCAAAA,oBAAAA,MAAeW;QACf,OAAO;YAAEG;YAAUF;YAAUlB;QAAS;IACxC;IAEA,MAAMW,YAAY,MAAMtB,QAAQgB,SAASG;IACzC,MAAMoC,iBAAiB,IAAI7B,IAAI,CAACT,KAAKuC,KAAK,GAAGlC,UAAUJ,KAAK,CAACuC,MAAM,CAAC,CAACjB,OAASA,KAAKG,QAAQ,IAAIN,aAAaG,KAAKC,SAAS,GAAGM,GAAG,CAAC,CAACP,OAASA,KAAKA,IAAI,IAAI,EAAE;IAC1J,IAAI,CAACe,eAAeG,IAAI,EAAE;YACkBzC;YAE1CA;QAFA,MAAMM,OAAOjB,kBAAkBgB,YAAWL,oBAAAA,KAAKO,WAAW,cAAhBP,+BAAAA,oBAAoB;QAC9DQ,YAAYH,WAAWiC,gBAAgB3C;SACvCK,iBAAAA,KAAKU,OAAO,cAAZV,qCAAAA,oBAAAA,MAAeM,KAAKK,KAAK;QACzB,OAAO;YAAEG,UAAUR,KAAKQ,QAAQ;YAAEF,UAAUN,KAAKM,QAAQ;YAAElB;QAAS;IACtE;IAEA,MAAMgD,kBAAkB,IAAIjC,IAAIJ,UAAUJ,KAAK,CAACuC,MAAM,CAAC,CAACjB,OAAS,CAACe,eAAeb,GAAG,CAACF,KAAKA,IAAI,GAAGO,GAAG,CAAC,CAACP,OAASA,KAAKA,IAAI;IACxH,MAAMjB,OAAOoC,gBAAgBD,IAAI,GAAGpD,kBAAkBC,qBAAqBe,WAAWqC,kBAAkB,QAAQN;IAChH,MAAMO,aAAa,IAAIC,IAAItC,iBAAAA,2BAAAA,KAAMQ,QAAQ,CAACb,KAAK,CAAC6B,GAAG,CAAC,CAACP,OAAS;YAACA,KAAKA,IAAI;YAAEA;SAAK;IAC/E,MAAMsB,YAAY,IAAID;IACtB,MAAME,WAA2B,EAAE;IACnC,MAAMC,YAAoD,EAAE;IAC5D,KAAK,MAAMxB,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAI,CAACqC,eAAeb,GAAG,CAACF,KAAKA,IAAI,GAAG;QACpC5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,+DAA+D,CAAC;QACnF,MAAMyB,YAAYD,UAAUE,EAAE,CAAC,CAAC;QAChC,IAAID,aAAaA,UAAU5C,IAAI,GAAG,MAAMmB,KAAKA,IAAI,EAAEyB,UAAU5C,IAAI,GAAGmB,KAAKA,IAAI;aACxEwB,UAAUlD,IAAI,CAAC;YAAEM,OAAOoB,KAAKA,IAAI;YAAEnB,MAAMmB,KAAKA,IAAI;QAAC;IAC1D;IACA,KAAK,MAAMrB,SAAS6C,UAAW;QAC7B,MAAMG,SAAS,MAAMpE,sBAAsB;YAAEiB;QAAQ,GAAG;YAAE,GAAGC,IAAI;YAAEC,OAAOC;YAAOQ,SAAS0B;QAAU,GAAGzC;QACvG,KAAK,MAAM4B,QAAQ2B,OAAOjD,KAAK,CAAE4C,UAAUM,GAAG,CAAC5B,KAAKA,IAAI,EAAEA;QAC1DuB,SAASjD,IAAI,IAAIqD,OAAOvC,KAAK;IAC/B;IACAH,YAAYH,WAAWiC,gBAAgB3C;IACvC,MAAMM,QAAQI,UAAUJ,KAAK,CAAC6B,GAAG,CAAC,CAACP;YAASsB;gBAAAA,iBAAAA,UAAUO,GAAG,CAAC7B,KAAKA,IAAI,eAAvBsB,4BAAAA,iBAA6BF,WAAWS,GAAG,CAAC7B,KAAKA,IAAI;;IACjG,MAAMX,WAAW1B,sBAAsBe,OAAO;QAAEiB,kBAAkB;QAAMX,aAAa;IAAK;IAC1F,MAAM8C,YAAY,IAAIT,IAAI3C,MAAM6B,GAAG,CAAC,CAACP,OAAS;YAACA,KAAKA,IAAI;YAAEA,KAAK+B,IAAI;SAAC;IACpE,MAAMzC,SAASzB,eAAe;QAC5B6B,IAAIL;QACJO,UAAU;QACVR,OAAOmC;QACPO;QACAE,WAAW,EAAEjD,iBAAAA,2BAAAA,KAAMQ,QAAQ,CAACD,MAAM,CAAC0C,WAAW;IAChD;IACA,IAAIjD,MAAM;QACRO,OAAO2C,cAAc,GAAGlD,KAAKQ,QAAQ,CAACD,MAAM,CAAC2C,cAAc;QAC3D3C,OAAO4C,YAAY,GAAGnD,KAAKQ,QAAQ,CAACD,MAAM,CAAC4C,YAAY;QACvD5C,OAAO6C,SAAS,GAAGpD,KAAKQ,QAAQ,CAACD,MAAM,CAAC6C,SAAS;QACjD7C,OAAO8C,WAAW,GAAGrD,KAAKQ,QAAQ,CAACD,MAAM,CAAC8C,WAAW;QACrD9C,OAAO+C,aAAa,GAAGtD,KAAKQ,QAAQ,CAACD,MAAM,CAAC+C,aAAa;QACzD/C,OAAO0C,WAAW,GAAGjD,KAAKQ,QAAQ,CAACD,MAAM,CAAC0C,WAAW;IACvD;IACA,MAAMzC,WAAqB;QAAEuB,eAAexB,OAAOwB,aAAa;QAAExB;QAAQZ;IAAM;KAChFD,gBAAAA,KAAKU,OAAO,cAAZV,oCAAAA,mBAAAA,MAAe4B,cAAc3B;IAC7B,OAAO;QAAEa;QAAUF;QAAUlB;IAAS;AACxC;AAEA;;;;;;CAMC,GACD,OAAO,eAAemE,cAAc9D,OAAe,EAAEC,OAA6B,CAAC,CAAC;IAClF,MAAMkD,SAAS,MAAMhB,eAAenC,SAASC;IAC7C,OAAO;QAAEY,UAAUsC,OAAOtC,QAAQ;QAAEC,QAAQqC,OAAOpC,QAAQ,CAACD,MAAM;QAAEnB,UAAUwD,OAAOxD,QAAQ;IAAC;AAChG"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/extract.ts"],"sourcesContent":["// The library layer for PDF -> markdown conversion: extractText/extractOcr/\n// pdfToMarkdown each take plain args, return plain data, and throw PdfToMdError — no process.exit, no console.*, no file writing.\n//\n// Warnings are DATA: every notice is returned in `warnings` verbatim and also\n// delivered incrementally via `onWarning` — the exact strings are load-bearing (the test suite asserts them).\nimport { buildOcrAnalysisPages } from './analyze.ts';\nimport { collect } from './collect.ts';\nimport { dictAvailable, dictUnavailableWarning } from './dict.ts';\nimport type { OcrProgressEvent } from './engines/tesseract.ts';\nimport type { PageWord, Tuning } from './geometry.ts';\nimport { renderPagesToMarkdown } from './render-analysis.ts';\nimport { AUTO_OCR_CHARS_PER_PAGE, buildOcrReport } from './report.ts';\nimport { buildTextAnalysis, selectCollectedPages } from './text-analysis.ts';\nimport type { Analysis, AnalysisPage, CollectResult, Report } from './types.ts';\nimport { validatePages } from './validate.ts';\n\nexport type { OcrProgressEvent } from './engines/tesseract.ts';\n\n// A plain-JS caller (or `.ts` with `as any`) gets no compile-time check on\n// `pages`, so it's validated once at runtime — see src/validate.ts's `validatePages`.\n\n/** One recognized word with its page — the OCR replay input contract (same\n * shape as `geometry.ts`'s `PageWord`, re-exported under this module's public name). */\nexport type OcrWordInput = PageWord;\n\n/** A 1-based, inclusive page range — the shape every entry point's `pages`\n * option takes (the CLI's own `--pages N[-M]` parses into this). */\nexport interface PageSelection {\n first: number;\n /** inclusive; defaults to `first` */\n last?: number;\n}\n\n/** extractText/extractOcr/pdfToMarkdown's common return shape. */\nexport interface ExtractResult {\n markdown: string;\n /** per-document and per-page QA metrics — see `Report` (`src/types.ts`) */\n report: Report;\n /** every notice also delivered incrementally via `onWarning`, collected\n * here in full (hybrid pages, auto-OCR-fallback, empty OCR pages, sparse-retry) */\n warnings: string[];\n}\n\ninterface CommonOptions {\n /** convert only this page range (1-based, inclusive); defaults to the\n * whole document */\n pages?: PageSelection;\n /** delivered as each warning happens (also returned, in full, as\n * `warnings` on the resolved result) */\n onWarning?: (message: string) => void;\n}\n\n/** Options for {@link extractText}. */\nexport interface ExtractTextOptions extends CommonOptions {\n /** emit a `### pN` marker before each page's content (output formatting\n * only — `report.pageStats` is always populated regardless) */\n pageMarkers?: boolean;\n /** source PDF text runs entering layout, delivered once for the selected\n * final result. Runs retain measured bounds and are not fabricated words. */\n onWords?: (words: OcrWordInput[]) => void;\n}\n\n/** Options for {@link extractOcr}. */\nexport interface ExtractOcrOptions extends CommonOptions {\n /** override any subset of the OCR-path geometry defaults — see\n * DEFAULT_TUNING (`src/geometry.ts`) for each knob's measured default */\n tuning?: Partial<Tuning>;\n /** page-render resolution for engine recognition (default 288, measured;\n * >~400 measured WORSE on ordinary pages — not a \"bigger is safer\" knob) */\n dpi?: number;\n /** marks report.ocrFallback — set by pdfToMarkdown's auto-fallback */\n fallback?: boolean;\n /** one event per page as recognition finishes (timing/count only — engine\n * chatter and the sparse-retry notice arrive via `onWarning` instead) */\n onProgress?: (e: OcrProgressEvent) => void;\n /** every recognized word entering the geometry, delivered once after\n * recognition — the severity-1 audit instrument (--debug-words) */\n onWords?: (words: OcrWordInput[]) => void;\n}\n\n/** Options for {@link pdfToMarkdown}. */\nexport interface PdfToMarkdownOptions extends ExtractTextOptions, Omit<ExtractOcrOptions, 'fallback'> {\n /** force the OCR path even when a text layer exists */\n ocr?: boolean;\n /** never fall back to OCR, even for image-only pages */\n noOcr?: boolean;\n}\n\n// One collector per entry point: collects into `warnings`, streams via\n// onWarning, and emits the dictionary-unavailable notice exactly once up front.\nfunction warnCollector(onWarning?: (message: string) => void): { warnings: string[]; warn: (m: string) => void } {\n const warnings: string[] = [];\n const warn = (m: string) => {\n warnings.push(m);\n onWarning?.(m);\n };\n if (!dictAvailable()) warn(dictUnavailableWarning());\n return { warnings, warn };\n}\n\n/** Extract embedded text with a QA report and warnings; never runs OCR.\n * @param pdfPath path to the PDF file\n * @param opts see {@link ExtractTextOptions}\n */\nexport async function extractText(pdfPath: string, opts: ExtractTextOptions = {}): Promise<ExtractResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n const range = opts.pages ? { first: opts.pages.first, last: opts.pages.last ?? opts.pages.first } : {};\n const collected = await collect(pdfPath, range);\n const text = buildTextAnalysis(collected, opts.pageMarkers ?? false);\n textNotices(collected, new Set(), warn);\n opts.onWords?.(text.words);\n return { markdown: text.markdown, report: text.analysis.report, warnings };\n}\n\n/**\n * OCR extraction. Input is either a PDF (pages render at `dpi` and the\n * engine recognizes them, imported lazily) or an already-recognized word dump (the offline replay input — see `--words-json`).\n *\n * @param input `{ pdfPath }` to recognize, or `{ words }` to replay a dump\n * @param opts see {@link ExtractOcrOptions}\n */\nexport async function extractOcr(input: { pdfPath: string } | { words: OcrWordInput[] }, opts: ExtractOcrOptions = {}): Promise<ExtractResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n\n // Shared with analyze(pdf, {path:'ocr'}) — see that module's header for\n // why this makes the two byte-identical by construction, not by hand.\n const { pages, words } = await buildOcrAnalysisPages(input, opts, warn);\n const md = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n\n const report = buildOcrReport({\n md,\n fallback: opts.fallback,\n // Confidence is on `words` — see report.ts's low-confidence detector for\n // how it handles a confidence-less replay honestly.\n words,\n });\n return { markdown: md, report, warnings };\n}\n\n/** A page needs OCR only when its image signal accompanies an unusable text\n * layer. The caller supplies the image part of that decision. */\nfunction needsPageOcr(chars: number): boolean {\n return chars < AUTO_OCR_CHARS_PER_PAGE;\n}\n\nfunction textNotices(collected: CollectResult, ocrPages: Set<number>, warn: (message: string) => void): void {\n for (const page of collected.pages) {\n if (page.textChars !== 0 || ocrPages.has(page.page)) continue;\n if (page.hasImage) warn(`p${page.page}: no extractable text; page has image content and OCR was not selected`);\n else warn(`p${page.page}: no extractable text; the page may be blank or contain only vector content`);\n }\n for (const page of collected.pages) {\n if (!page.largeImage || ocrPages.has(page.page)) continue;\n warn(`p${page.page}: text layer used but page carries large image(s); if it has baked-in text this output won't have it — compare (don't blindly replace) against --pages ${page.page} --ocr, which can itself introduce errors`);\n }\n}\n\nfunction analysisWords(pages: AnalysisPage[]): OcrWordInput[] {\n return pages.flatMap((page) => page.words.map((word) => ({ page: page.page, text: word.text, ...word.box, confidence: word.confidence })));\n}\n\ninterface AutoAnalysisResult {\n analysis: Analysis;\n markdown: string;\n warnings: string[];\n}\n\n/** Shared automatic route for the library's Markdown result and the CLI's\n * raw/plain-text formats. A page changes producer only when it has image\n * evidence and an unusable text layer, so a blank or vector-only page never\n * becomes speculative OCR input. */\nexport async function analyzePdfAuto(pdfPath: string, opts: PdfToMarkdownOptions = {}): Promise<AutoAnalysisResult> {\n validatePages(opts.pages);\n const { warnings, warn } = warnCollector(opts.onWarning);\n const range = opts.pages ? { first: opts.pages.first, last: opts.pages.last ?? opts.pages.first } : {};\n\n if (opts.ocr) {\n const { pages, words } = await buildOcrAnalysisPages({ pdfPath }, { ...opts, onWords: undefined }, warn);\n const markdown = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n const report = buildOcrReport({ md: markdown, words });\n const analysis: Analysis = { outputVersion: report.outputVersion, report, pages };\n opts.onWords?.(words);\n return { analysis, markdown, warnings };\n }\n\n const collected = await collect(pdfPath, range);\n const ocrPageNumbers = new Set(!opts.noOcr ? collected.pages.filter((page) => page.hasImage && needsPageOcr(page.textChars)).map((page) => page.page) : []);\n if (!ocrPageNumbers.size) {\n const text = buildTextAnalysis(collected, opts.pageMarkers ?? false);\n textNotices(collected, ocrPageNumbers, warn);\n opts.onWords?.(text.words);\n return { analysis: text.analysis, markdown: text.markdown, warnings };\n }\n\n const textPageNumbers = new Set(collected.pages.filter((page) => !ocrPageNumbers.has(page.page)).map((page) => page.page));\n const text = textPageNumbers.size ? buildTextAnalysis(selectCollectedPages(collected, textPageNumbers), true) : undefined;\n const textByPage = new Map(text?.analysis.pages.map((page) => [page.page, page]));\n const ocrByPage = new Map<number, AnalysisPage>();\n const ocrWords: OcrWordInput[] = [];\n const ocrRanges: Array<{ first: number; last: number }> = [];\n for (const page of collected.pages) {\n if (!ocrPageNumbers.has(page.page)) continue;\n warn(`p${page.page}: sparse text layer with image content; using OCR for this page`);\n const lastRange = ocrRanges.at(-1);\n if (lastRange && lastRange.last + 1 === page.page) lastRange.last = page.page;\n else ocrRanges.push({ first: page.page, last: page.page });\n }\n for (const range of ocrRanges) {\n const result = await buildOcrAnalysisPages({ pdfPath }, { ...opts, pages: range, onWords: undefined }, warn);\n for (const page of result.pages) ocrByPage.set(page.page, page);\n ocrWords.push(...result.words);\n }\n textNotices(collected, ocrPageNumbers, warn);\n const pages = collected.pages.map((page) => ocrByPage.get(page.page) ?? (textByPage.get(page.page) as AnalysisPage));\n const markdown = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n const pagePaths = new Map(pages.map((page) => [page.page, page.path]));\n const report = buildOcrReport({\n md: markdown,\n fallback: true,\n words: ocrWords,\n pagePaths,\n hybridPages: text?.analysis.report.hybridPages,\n });\n if (text) {\n report.bodyFontHeight = text.analysis.report.bodyFontHeight;\n report.headingSizes = text.analysis.report.headingSizes;\n report.listItems = text.analysis.report.listItems;\n report.joinedPairs = text.analysis.report.joinedPairs;\n report.closedHyphens = text.analysis.report.closedHyphens;\n report.hybridPages = text.analysis.report.hybridPages;\n }\n const analysis: Analysis = { outputVersion: report.outputVersion, report, pages };\n opts.onWords?.(analysisWords(pages));\n return { analysis, markdown, warnings };\n}\n\n/**\n * Extract each page from embedded text, using OCR only when the page has\n * image content and fewer than `AUTO_OCR_CHARS_PER_PAGE` text characters.\n *\n * @param pdfPath path to the PDF file\n * @param opts see {@link PdfToMarkdownOptions}\n */\nexport async function pdfToMarkdown(pdfPath: string, opts: PdfToMarkdownOptions = {}): Promise<ExtractResult> {\n const result = await analyzePdfAuto(pdfPath, opts);\n return { markdown: result.markdown, report: result.analysis.report, warnings: result.warnings };\n}\n"],"names":["buildOcrAnalysisPages","collect","dictAvailable","dictUnavailableWarning","renderPagesToMarkdown","AUTO_OCR_CHARS_PER_PAGE","buildOcrReport","buildTextAnalysis","selectCollectedPages","validatePages","warnCollector","onWarning","warnings","warn","m","push","extractText","pdfPath","opts","pages","range","first","last","collected","text","pageMarkers","textNotices","Set","onWords","words","markdown","report","analysis","extractOcr","input","md","forcePageMarkers","fallback","needsPageOcr","chars","ocrPages","page","textChars","has","hasImage","largeImage","analysisWords","flatMap","map","word","box","confidence","analyzePdfAuto","ocr","undefined","outputVersion","ocrPageNumbers","noOcr","filter","size","textPageNumbers","textByPage","Map","ocrByPage","ocrWords","ocrRanges","lastRange","at","result","set","get","pagePaths","path","hybridPages","bodyFontHeight","headingSizes","listItems","joinedPairs","closedHyphens","pdfToMarkdown"],"mappings":"AAAA,4EAA4E;AAC5E,kIAAkI;AAClI,EAAE;AACF,8EAA8E;AAC9E,8GAA8G;AAC9G,SAASA,qBAAqB,QAAQ,eAAe;AACrD,SAASC,OAAO,QAAQ,eAAe;AACvC,SAASC,aAAa,EAAEC,sBAAsB,QAAQ,YAAY;AAGlE,SAASC,qBAAqB,QAAQ,uBAAuB;AAC7D,SAASC,uBAAuB,EAAEC,cAAc,QAAQ,cAAc;AACtE,SAASC,iBAAiB,EAAEC,oBAAoB,QAAQ,qBAAqB;AAE7E,SAASC,aAAa,QAAQ,gBAAgB;AA0E9C,uEAAuE;AACvE,gFAAgF;AAChF,SAASC,cAAcC,SAAqC;IAC1D,MAAMC,WAAqB,EAAE;IAC7B,MAAMC,OAAO,CAACC;QACZF,SAASG,IAAI,CAACD;QACdH,sBAAAA,gCAAAA,UAAYG;IACd;IACA,IAAI,CAACZ,iBAAiBW,KAAKV;IAC3B,OAAO;QAAES;QAAUC;IAAK;AAC1B;AAEA;;;CAGC,GACD,OAAO,eAAeG,YAAYC,OAAe,EAAEC,OAA2B,CAAC,CAAC;QAGlBA,kBAElBA;QAE1CA;IANAT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IACvD,MAAMS,QAAQF,KAAKC,KAAK,GAAG;QAAEE,OAAOH,KAAKC,KAAK,CAACE,KAAK;QAAEC,IAAI,GAAEJ,mBAAAA,KAAKC,KAAK,CAACG,IAAI,cAAfJ,8BAAAA,mBAAmBA,KAAKC,KAAK,CAACE,KAAK;IAAC,IAAI,CAAC;IACrG,MAAME,YAAY,MAAMtB,QAAQgB,SAASG;IACzC,MAAMI,OAAOjB,kBAAkBgB,YAAWL,oBAAAA,KAAKO,WAAW,cAAhBP,+BAAAA,oBAAoB;IAC9DQ,YAAYH,WAAW,IAAII,OAAOd;KAClCK,gBAAAA,KAAKU,OAAO,cAAZV,oCAAAA,mBAAAA,MAAeM,KAAKK,KAAK;IACzB,OAAO;QAAEC,UAAUN,KAAKM,QAAQ;QAAEC,QAAQP,KAAKQ,QAAQ,CAACD,MAAM;QAAEnB;IAAS;AAC3E;AAEA;;;;;;CAMC,GACD,OAAO,eAAeqB,WAAWC,KAAsD,EAAEhB,OAA0B,CAAC,CAAC;IACnHT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IAEvD,wEAAwE;IACxE,sEAAsE;IACtE,MAAM,EAAEQ,KAAK,EAAEU,KAAK,EAAE,GAAG,MAAM7B,sBAAsBkC,OAAOhB,MAAML;IAClE,MAAMsB,KAAK/B,sBAAsBe,OAAO;QAAEiB,kBAAkB;QAAMX,aAAa;IAAK;IAEpF,MAAMM,SAASzB,eAAe;QAC5B6B;QACAE,UAAUnB,KAAKmB,QAAQ;QACvB,yEAAyE;QACzE,oDAAoD;QACpDR;IACF;IACA,OAAO;QAAEC,UAAUK;QAAIJ;QAAQnB;IAAS;AAC1C;AAEA;+DAC+D,GAC/D,SAAS0B,aAAaC,KAAa;IACjC,OAAOA,QAAQlC;AACjB;AAEA,SAASqB,YAAYH,SAAwB,EAAEiB,QAAqB,EAAE3B,IAA+B;IACnG,KAAK,MAAM4B,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAIsB,KAAKC,SAAS,KAAK,KAAKF,SAASG,GAAG,CAACF,KAAKA,IAAI,GAAG;QACrD,IAAIA,KAAKG,QAAQ,EAAE/B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,sEAAsE,CAAC;aACxG5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,2EAA2E,CAAC;IACtG;IACA,KAAK,MAAMA,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAI,CAACsB,KAAKI,UAAU,IAAIL,SAASG,GAAG,CAACF,KAAKA,IAAI,GAAG;QACjD5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,uJAAuJ,EAAEA,KAAKA,IAAI,CAAC,yCAAyC,CAAC;IAClO;AACF;AAEA,SAASK,cAAc3B,KAAqB;IAC1C,OAAOA,MAAM4B,OAAO,CAAC,CAACN,OAASA,KAAKZ,KAAK,CAACmB,GAAG,CAAC,CAACC,OAAU,CAAA;gBAAER,MAAMA,KAAKA,IAAI;gBAAEjB,MAAMyB,KAAKzB,IAAI;gBAAE,GAAGyB,KAAKC,GAAG;gBAAEC,YAAYF,KAAKE,UAAU;YAAC,CAAA;AACxI;AAQA;;;kCAGkC,GAClC,OAAO,eAAeC,eAAenC,OAAe,EAAEC,OAA6B,CAAC,CAAC;QAGvBA;QA0D5DA;IA5DAT,cAAcS,KAAKC,KAAK;IACxB,MAAM,EAAEP,QAAQ,EAAEC,IAAI,EAAE,GAAGH,cAAcQ,KAAKP,SAAS;IACvD,MAAMS,QAAQF,KAAKC,KAAK,GAAG;QAAEE,OAAOH,KAAKC,KAAK,CAACE,KAAK;QAAEC,IAAI,GAAEJ,mBAAAA,KAAKC,KAAK,CAACG,IAAI,cAAfJ,8BAAAA,mBAAmBA,KAAKC,KAAK,CAACE,KAAK;IAAC,IAAI,CAAC;IAErG,IAAIH,KAAKmC,GAAG,EAAE;YAKZnC;QAJA,MAAM,EAAEC,KAAK,EAAEU,KAAK,EAAE,GAAG,MAAM7B,sBAAsB;YAAEiB;QAAQ,GAAG;YAAE,GAAGC,IAAI;YAAEU,SAAS0B;QAAU,GAAGzC;QACnG,MAAMiB,WAAW1B,sBAAsBe,OAAO;YAAEiB,kBAAkB;YAAMX,aAAa;QAAK;QAC1F,MAAMM,SAASzB,eAAe;YAAE6B,IAAIL;YAAUD;QAAM;QACpD,MAAMG,WAAqB;YAAEuB,eAAexB,OAAOwB,aAAa;YAAExB;YAAQZ;QAAM;SAChFD,iBAAAA,KAAKU,OAAO,cAAZV,qCAAAA,oBAAAA,MAAeW;QACf,OAAO;YAAEG;YAAUF;YAAUlB;QAAS;IACxC;IAEA,MAAMW,YAAY,MAAMtB,QAAQgB,SAASG;IACzC,MAAMoC,iBAAiB,IAAI7B,IAAI,CAACT,KAAKuC,KAAK,GAAGlC,UAAUJ,KAAK,CAACuC,MAAM,CAAC,CAACjB,OAASA,KAAKG,QAAQ,IAAIN,aAAaG,KAAKC,SAAS,GAAGM,GAAG,CAAC,CAACP,OAASA,KAAKA,IAAI,IAAI,EAAE;IAC1J,IAAI,CAACe,eAAeG,IAAI,EAAE;YACkBzC;YAE1CA;QAFA,MAAMM,OAAOjB,kBAAkBgB,YAAWL,oBAAAA,KAAKO,WAAW,cAAhBP,+BAAAA,oBAAoB;QAC9DQ,YAAYH,WAAWiC,gBAAgB3C;SACvCK,iBAAAA,KAAKU,OAAO,cAAZV,qCAAAA,oBAAAA,MAAeM,KAAKK,KAAK;QACzB,OAAO;YAAEG,UAAUR,KAAKQ,QAAQ;YAAEF,UAAUN,KAAKM,QAAQ;YAAElB;QAAS;IACtE;IAEA,MAAMgD,kBAAkB,IAAIjC,IAAIJ,UAAUJ,KAAK,CAACuC,MAAM,CAAC,CAACjB,OAAS,CAACe,eAAeb,GAAG,CAACF,KAAKA,IAAI,GAAGO,GAAG,CAAC,CAACP,OAASA,KAAKA,IAAI;IACxH,MAAMjB,OAAOoC,gBAAgBD,IAAI,GAAGpD,kBAAkBC,qBAAqBe,WAAWqC,kBAAkB,QAAQN;IAChH,MAAMO,aAAa,IAAIC,IAAItC,iBAAAA,2BAAAA,KAAMQ,QAAQ,CAACb,KAAK,CAAC6B,GAAG,CAAC,CAACP,OAAS;YAACA,KAAKA,IAAI;YAAEA;SAAK;IAC/E,MAAMsB,YAAY,IAAID;IACtB,MAAME,WAA2B,EAAE;IACnC,MAAMC,YAAoD,EAAE;IAC5D,KAAK,MAAMxB,QAAQlB,UAAUJ,KAAK,CAAE;QAClC,IAAI,CAACqC,eAAeb,GAAG,CAACF,KAAKA,IAAI,GAAG;QACpC5B,KAAK,CAAC,CAAC,EAAE4B,KAAKA,IAAI,CAAC,+DAA+D,CAAC;QACnF,MAAMyB,YAAYD,UAAUE,EAAE,CAAC,CAAC;QAChC,IAAID,aAAaA,UAAU5C,IAAI,GAAG,MAAMmB,KAAKA,IAAI,EAAEyB,UAAU5C,IAAI,GAAGmB,KAAKA,IAAI;aACxEwB,UAAUlD,IAAI,CAAC;YAAEM,OAAOoB,KAAKA,IAAI;YAAEnB,MAAMmB,KAAKA,IAAI;QAAC;IAC1D;IACA,KAAK,MAAMrB,SAAS6C,UAAW;QAC7B,MAAMG,SAAS,MAAMpE,sBAAsB;YAAEiB;QAAQ,GAAG;YAAE,GAAGC,IAAI;YAAEC,OAAOC;YAAOQ,SAAS0B;QAAU,GAAGzC;QACvG,KAAK,MAAM4B,QAAQ2B,OAAOjD,KAAK,CAAE4C,UAAUM,GAAG,CAAC5B,KAAKA,IAAI,EAAEA;QAC1DuB,SAASjD,IAAI,IAAIqD,OAAOvC,KAAK;IAC/B;IACAH,YAAYH,WAAWiC,gBAAgB3C;IACvC,MAAMM,QAAQI,UAAUJ,KAAK,CAAC6B,GAAG,CAAC,CAACP;YAASsB;gBAAAA,iBAAAA,UAAUO,GAAG,CAAC7B,KAAKA,IAAI,eAAvBsB,4BAAAA,iBAA6BF,WAAWS,GAAG,CAAC7B,KAAKA,IAAI;;IACjG,MAAMX,WAAW1B,sBAAsBe,OAAO;QAAEiB,kBAAkB;QAAMX,aAAa;IAAK;IAC1F,MAAM8C,YAAY,IAAIT,IAAI3C,MAAM6B,GAAG,CAAC,CAACP,OAAS;YAACA,KAAKA,IAAI;YAAEA,KAAK+B,IAAI;SAAC;IACpE,MAAMzC,SAASzB,eAAe;QAC5B6B,IAAIL;QACJO,UAAU;QACVR,OAAOmC;QACPO;QACAE,WAAW,EAAEjD,iBAAAA,2BAAAA,KAAMQ,QAAQ,CAACD,MAAM,CAAC0C,WAAW;IAChD;IACA,IAAIjD,MAAM;QACRO,OAAO2C,cAAc,GAAGlD,KAAKQ,QAAQ,CAACD,MAAM,CAAC2C,cAAc;QAC3D3C,OAAO4C,YAAY,GAAGnD,KAAKQ,QAAQ,CAACD,MAAM,CAAC4C,YAAY;QACvD5C,OAAO6C,SAAS,GAAGpD,KAAKQ,QAAQ,CAACD,MAAM,CAAC6C,SAAS;QACjD7C,OAAO8C,WAAW,GAAGrD,KAAKQ,QAAQ,CAACD,MAAM,CAAC8C,WAAW;QACrD9C,OAAO+C,aAAa,GAAGtD,KAAKQ,QAAQ,CAACD,MAAM,CAAC+C,aAAa;QACzD/C,OAAO0C,WAAW,GAAGjD,KAAKQ,QAAQ,CAACD,MAAM,CAAC0C,WAAW;IACvD;IACA,MAAMzC,WAAqB;QAAEuB,eAAexB,OAAOwB,aAAa;QAAExB;QAAQZ;IAAM;KAChFD,gBAAAA,KAAKU,OAAO,cAAZV,oCAAAA,mBAAAA,MAAe4B,cAAc3B;IAC7B,OAAO;QAAEa;QAAUF;QAAUlB;IAAS;AACxC;AAEA;;;;;;CAMC,GACD,OAAO,eAAemE,cAAc9D,OAAe,EAAEC,OAA6B,CAAC,CAAC;IAClF,MAAMkD,SAAS,MAAMhB,eAAenC,SAASC;IAC7C,OAAO;QAAEY,UAAUsC,OAAOtC,QAAQ;QAAEC,QAAQqC,OAAOpC,QAAQ,CAACD,MAAM;QAAEnB,UAAUwD,OAAOxD,QAAQ;IAAC;AAChG"}