pdfrum-markdown

Markdown, and layout-preserving plain text, from an interpreted page. Two readings of the same page, chosen by what the file offers: a tagged PDF (ISO 32000-1 §14.7) names its headings, paragraphs, lists, tables and figures in the structure tree, and an untagged one is read by typography instead — body size against heading scale, list markers, monospace runs as code fences, running headers dropped.

use pdfrum_markdown::{Block, render};

let md = render(&[
    Block::Heading { level: 1, text: "Results".into() },
    Block::Paragraph("Ninety-nine percent agreement.".into()),
]);
assert_eq!(md, "# Results\n\nNinety-nine percent agreement.\n");

The pipeline is three named stages, and each is callable on its own: [page_lines] extracts and groups drawn glyphs into [Line]s, [page_blocks] turns those into [Block]s (through the structure tree when there is one, [heuristics] when there is not), and [render()] writes CommonMark. [page_markdown] is the three composed; [page_layout] stops after the first and keeps the page's own geometry as spacing instead.

[document_blocks] is the whole-document entry point and is not a loop over [page_blocks]: repeated header and footer bands are only detectable across pages, so a line that recurs in the same band on page after page is dropped there and could not be dropped one page at a time.

Images are indices, never bytes. Block::Image carries the image's position in the page's drawing order — the order the facade's Page::images lists them in — so [render_with_images] can link each one wherever the caller actually wrote the file.

From the facade: Page::markdown, feature markdown.

Part of pdfrum. #![forbid(unsafe_code)].

MIT OR Apache-2.0