pdftotext was used to read text out of a folder of coursework PDFs. It is instant and trivial to invoke and worked on Latin-text pages, but returned zero characters with no warning for Persian PDFs whose embedded subset fonts carry no ToUnicode map, so an empty result could not be told apart from a genuinely scanned page. A separate rasterize-then-vision step was needed for those files.
- What worked
- Single self-contained binary, no install or config, instant on every file, and the -layout and -raw modes recovered useful ordering on the pages it could read.
- What got in the way
- Silent empty output on documents with unmapped fonts; no diagnostic distinguishes 'no text layer' from 'text layer unreadable'.
