# pdf-slim

Losslessly shrink a PDF. The output renders pixel-for-pixel identically to the
input; any candidate that doesn't is discarded.

## How it works

`pdf-slim` builds several smaller candidates and keeps the smallest one that
passes verification. The input file is never modified.

1. **Structural** — `qpdf` object-stream generation and maximum-level flate
   recompression. Always render-safe; a modest win.
2. **mutool-subset** — `mutool clean` garbage-collects, deduplicates, deflates,
   and natively subsets fonts.
3. **font-resubset** — for any TrueType `FontFile2` stream still over a size
   threshold (the "Word embedded the whole font" case), the font is re-subset
   with `pyftsubset` against the document's actual character set. Glyph IDs are
   retained (`--retain-gids`) so CID / Identity-H instances sharing the file
   stay valid. The stream is spliced back, lengths are fixed with `fix-qdf`, and
   the result is recompressed.

Every candidate is verified by rendering each page of both the original and the
candidate at the same DPI and requiring byte-identical pixels. The smallest
verified candidate wins. If nothing wins by more than 1%, the input is copied
through unchanged, so repeated runs are stable.

## Usage

```
pdf-slim input.pdf [output.pdf] [--dpi 200] [--min-font-bytes 150000]
```

- `input.pdf` — the PDF to shrink.
- `output.pdf` — optional; defaults to `<input>-slim.pdf`. Refuses to overwrite
  the input.
- `--dpi` — verification render resolution (default `200`).
- `--min-font-bytes` — re-subset embedded TrueType fonts larger than this
  (default `150000`).

## Requirements

The following tools must be on `PATH`:

- `mutool` (MuPDF)
- `qpdf` (provides `qpdf` and `fix-qdf`)
- `pyftsubset` (fontTools)
- `python3`

## Ownership

Owned by SILO GROUP — [www.silogroup.org](https://www.silogroup.org).