macos-vision 1.0.2 → 1.1.0

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## [1.1.0](https://github.com/woladi/macos-vision/compare/v1.0.3...v1.1.0) (2026-04-09)
+
+### Features
+
+* add PDF support via sips rasterization ([a48bf17](https://github.com/woladi/macos-vision/commit/a48bf17579a6df11aed6eadbde4fa5041ccaa981))
+
+## [1.0.3](https://github.com/woladi/macos-vision/compare/v1.0.2...v1.0.3) (2026-04-08)
+
+### Reverts
+
+* remove socket.ignore field — worsens supply chain risk score ([a1827ad](https://github.com/woladi/macos-vision/commit/a1827ad489220ebb7a2e8c85632945fe969438db))
+
 ## [1.0.2](https://github.com/woladi/macos-vision/compare/v1.0.1...v1.0.2) (2026-04-08)
 
 ## [1.0.1](https://github.com/woladi/macos-vision/compare/v0.3.1...v1.0.1) (2026-04-08)

package/dist/index.d.ts

@@ -11,6 +11,8 @@ export interface VisionBlock {
     height: number;
     /** OCR transcription confidence, 0–1 */
     confidence: number;
+    /** 0-based page index. Present only when the source was a PDF. Absent for images. */
+    page?: number;
 }
 export interface OcrOptions {
     /** Return plain text (default) or structured blocks with coordinates */

package/dist/index.js

@@ -1,20 +1,93 @@
 import { execFile } from 'child_process';
 import { promisify } from 'util';
-import { resolve, dirname } from 'path';
+import { resolve, dirname, extname, join } from 'path';
 import { fileURLToPath } from 'url';
+import { tmpdir } from 'os';
+import { open, mkdir, readdir, rm } from 'fs/promises';
 const execFileAsync = promisify(execFile);
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const BIN_PATH = resolve(__dirname, '../bin/vision-helper');
 const BINARY_TIMEOUT_MS = 30_000;
+const SIPS_TIMEOUT_MS = 60_000;
 async function run(flag, imagePath) {
     const { stdout } = await execFileAsync(BIN_PATH, [flag, resolve(imagePath)], {
         timeout: BINARY_TIMEOUT_MS,
     });
     return stdout;
 }
+// ─── PDF helpers ─────────────────────────────────────────────────────────────
+/**
+ * Returns true if the file at `filePath` is a PDF.
+ * Uses extension as a fast path; falls back to magic bytes (`%PDF`) for
+ * files whose extension does not match their actual content.
+ */
+async function isPdf(filePath) {
+    if (extname(filePath).toLowerCase() === '.pdf')
+        return true;
+    let fh;
+    try {
+        fh = await open(filePath, 'r');
+        const buf = Buffer.alloc(4);
+        await fh.read(buf, 0, 4, 0);
+        return buf[0] === 0x25 && buf[1] === 0x50 && buf[2] === 0x44 && buf[3] === 0x46;
+    }
+    finally {
+        await fh?.close();
+    }
+}
+/**
+ * Rasterizes a PDF to PNG files in `outDir` using macOS `sips`.
+ * Returns sorted list of absolute PNG paths (order = page order).
+ *
+ * sips names single-page output `{basename}.png` and multi-page output
+ * `{basename}-1.png`, `{basename}-2.png`, etc. The numeric sort handles both.
+ */
+async function rasterizePdf(pdfPath, outDir) {
+    await execFileAsync('sips', ['-s', 'format', 'png', '--resampleHeight', '2000', pdfPath, '--out', outDir], { timeout: SIPS_TIMEOUT_MS });
+    const entries = await readdir(outDir);
+    const pngs = entries.filter((n) => n.toLowerCase().endsWith('.png'));
+    pngs.sort((a, b) => {
+        const numA = parseInt(a.match(/-(\d+)\.png$/i)?.[1] ?? '0', 10);
+        const numB = parseInt(b.match(/-(\d+)\.png$/i)?.[1] ?? '0', 10);
+        return numA - numB;
+    });
+    return pngs.map((n) => join(outDir, n));
+}
+/**
+ * Full PDF OCR pipeline: rasterize → per-page OCR → merge results.
+ * Temporary PNG files are always cleaned up in the `finally` block.
+ */
+async function ocrPdf(pdfPath, format) {
+    const outDir = join(tmpdir(), `macos-vision-${globalThis.crypto.randomUUID()}`);
+    await mkdir(outDir, { recursive: true });
+    try {
+        const pages = await rasterizePdf(pdfPath, outDir);
+        if (format === 'blocks') {
+            const all = [];
+            for (let i = 0; i < pages.length; i++) {
+                const blocks = (await ocr(pages[i], { format: 'blocks' }));
+                all.push(...blocks.map((b) => ({ ...b, page: i })));
+            }
+            return all;
+        }
+        const texts = [];
+        for (let i = 0; i < pages.length; i++) {
+            texts.push((await ocr(pages[i])));
+        }
+        return texts.join('\n\n--- Page Break ---\n\n');
+    }
+    finally {
+        await rm(outDir, { recursive: true, force: true });
+    }
+}
 export async function ocr(imagePath, options = {}) {
     const absPath = resolve(imagePath);
     const { format = 'text' } = options;
+    // ── PDF fast-path: rasterize via sips, then OCR each page ────────────────
+    if (await isPdf(absPath)) {
+        return ocrPdf(absPath, format);
+    }
+    // ── Existing image path (unchanged) ──────────────────────────────────────
     if (format === 'blocks') {
         const { stdout } = await execFileAsync(BIN_PATH, ['--json', absPath], {
             timeout: BINARY_TIMEOUT_MS,

package/dist/layout.d.ts

@@ -85,6 +85,14 @@ export declare function sortBlocksByReadingOrder(blocks: LayoutBlock[]): LayoutB
  * (`paragraphId`) using simple bounding-box heuristics. All other block types
  * are placed into the sorted sequence by their top-left coordinate.
  *
+ * **Multi-page PDFs**: `VisionBlock` items from PDF OCR carry an optional `page` field (0-based).
+ * Because all coordinates are page-local (0–1 relative to each page), mixing blocks from
+ * different pages produces meaningless geometry. Pre-filter by page before calling inferLayout:
+ * ```ts
+ * const pageBlocks = pdfBlocks.filter(b => b.page === 0);
+ * const layout = inferLayout({ textBlocks: pageBlocks });
+ * ```
+ *
  * @example
  * ```ts
  * const blocks  = await ocr('page.png', { format: 'blocks' });

package/dist/layout.js

@@ -114,6 +114,14 @@ export function sortBlocksByReadingOrder(blocks) {
  * (`paragraphId`) using simple bounding-box heuristics. All other block types
  * are placed into the sorted sequence by their top-left coordinate.
  *
+ * **Multi-page PDFs**: `VisionBlock` items from PDF OCR carry an optional `page` field (0-based).
+ * Because all coordinates are page-local (0–1 relative to each page), mixing blocks from
+ * different pages produces meaningless geometry. Pre-filter by page before calling inferLayout:
+ * ```ts
+ * const pageBlocks = pdfBlocks.filter(b => b.page === 0);
+ * const layout = inferLayout({ textBlocks: pageBlocks });
+ * ```
+ *
  * @example
  * ```ts
  * const blocks  = await ocr('page.png', { format: 'blocks' });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "macos-vision",
-  "version": "1.0.2",
+  "version": "1.1.0",
   "description": "Apple Vision OCR & image analysis for Node.js — native, fast, offline, no API keys",
   "author": "Adrian Wolczuk",
   "license": "MIT",
@@ -48,11 +48,6 @@
       "eslint --fix"
     ]
   },
-  "socket": {
-    "ignore": {
-      "installScripts": true
-    }
-  },
   "os": [
     "darwin"
   ],