prompt-a-thon

Files and conversion

Participant files land in UPLOAD_FOLDER. The web app and the converter do not share a directory for the conversion itself. The web process sends DOCX bytes over a Unix socket. The daemon, running as promptathon-convert, writes only inside its own temp directory and returns PDF bytes.

Protocol:

CONVERT <job_id> <byte_count>\n
<docx bytes>

Reply is OK <job_id> <byte_count> plus PDF bytes, or ERROR <job_id> <message>.

The daemon runs LibreOffice Writer headless (--convert-to pdf:writer_pdf_Export) with a 90 second timeout and a small concurrency limit. pandoc is not part of this path. WeasyPrint is not part of this path. WeasyPrint is used for certificate PDFs.

After conversion, or for a PDF uploaded directly, the app:

rejects tiny or oversized files and files that do not start with %PDF
rewrites the PDF with pypdf when it can, because some generators trip qpdf
opens it with pikepdf, strips actions that can launch code, and clears document metadata

Preview helpers in viewer_workers/ can read CSV, DOCX, and XLSX for admin viewing. They are separate from the submission store.

The admin server page reports whether prompt-a-thon-convert is active and whether the socket accepts a connection. Gunicorn writes logs/error.log and logs/access.log. The Flask logger also writes logs/app.log. The server page shows error.log when that file exists, otherwise app.log.

This site is open source. Improve this page.