Replace the client-side PDF.js viewer with server-side rendering using PyMuPDF.
This avoids CSP worker restrictions and ES private field access issues that
affected Chromium browsers.
Changes:
- Add /api/v1/pdf-preview endpoint to MCP server (management.py)
- Add pdf-preview route and controller action in Astrolabe PHP backend
- Refactor PDFViewer.vue to display server-rendered PNG images
- Remove pdfjs-dist dependency and client-side PDF loading code
- Use @nextcloud/axios for CSRF token handling in PDFViewer
The server downloads the PDF via WebDAV, renders the requested page with
PyMuPDF at the specified scale, and returns a base64-encoded PNG image.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update psalm-baseline.xml to match renamed OauthController.php (lowercase 'a')
- Move AlertCircle import to top of PDFViewer.vue to satisfy ESLint import/first rule
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When viewing PDF chunks in semantic search, the PDF viewer failed with
"can't access private field" errors. This was caused by:
1. CSP blocks web workers (worker-src 'none'), forcing fake worker mode
2. Vite transforms ES private fields in the bundle, but the worker file
is untransformed, causing incompatible private field implementations
3. Vue's ref() wraps PDFDocumentProxy in a Proxy, which can't access
ES private fields
Fixed by:
- Loading pdfjs-dist externally via script tag (avoids Vite transform)
- Creating pdfjs-loader.mjs that imports pdf.mjs and sets window.pdfjsLib
- Using Util::addScript() for CSP-compliant script loading with nonces
- Using shallowRef() instead of ref() for pdfDoc to avoid Proxy wrapper
- Setting workerSrc at runtime using OC.linkTo() for correct app path
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Chris Coutinho