# DjVu module for johnmcm.com

The WebAssembly module behind johnmcm.com's DjVu Viewer (/open-file/djvu)
and DjVu to PDF converter (/djvu-to-pdf).

It is [djvu-rs](https://github.com/matyushkin/djvu-rs) 0.42.0 by Mikhail
Matyushkin, MIT licensed and written from the public DjVu v3 specification
(no DjVuLibre or other GPL code), compiled with wasm-pack. djvu-rs's own
browser bindings (WasmDocument, WasmPage: rendering and the text layer) are
used unchanged; its npm build lacks the PDF export and the bookmarks, so the
small crate in this folder (`src/lib.rs`) adds `export_pdf` and
`outline_json`.

This folder publishes exactly what is built:

- `Cargo.toml`, `Cargo.lock` and `src/lib.rs`: the crate, with djvu-rs pinned
  to 0.42.0. The Dockerfile downloads djvu-rs from crates.io and checks it
  against the checksum in Cargo.lock.
- `patches/pdf-text-and-links.patch`: the one change made to djvu-rs. Its PDF
  text layer was set in Helvetica with WinAnsi encoding, and every character
  outside ASCII became "?", so a French, Russian or Chinese book's PDF could
  not be searched or copied from. The patch sets that invisible text in an
  embedded, glyphless Type0 font (Identity-H, each character written as its
  UTF-16 code unit, with a ToUnicode map), as Tesseract and OCRmyPDF do, and
  writes non-ASCII bookmark titles as UTF-16. It also makes a link to another
  page of the book ("#12") go to that page, through a named destination per
  page; djvu-rs wrote it as a web link to "#12", which goes nowhere.
- `font/make_glyphless.py` and `font/glyphless.ttf`: the font the patch
  embeds, two empty glyphs half an em wide, made with fontTools for this site
  (public domain).
- `src/worker.mjs`: the worker the pages talk to.
- `Dockerfile` and `build.ps1`: the build, with the Rust image pinned by
  digest. Run `build.ps1` with Docker Desktop running; the module goes to
  `frontend/vendor/local/djvu@2/`.

The module is loaded as separate files by the page, so it can be replaced
with one built from modified source.