PdfPecker
Docs: How it works
Download PDF

How it works

The source paths below are in the PdfPecker repository under src/PdfPecker/.

Markdown is first turned into HTML: Markdown/BlockParser finds the block structure line by line, following the CommonMark reference algorithm, Markdown/InlineParser handles emphasis (delimiter stack) and links (bracket stack), and Markdown/HtmlRenderer writes the HTML, which then takes the same path as any other document:

  1. Html/HtmlParser builds a DOM in a single pass, applying the browser rules for implied end tags.
  2. Css/StyleResolver cascades a user-agent stylesheet, presentational attributes, document stylesheets and inline styles. Rules are indexed by id, class and tag, and elements with identical inputs share one computed style. For a tagged PDF, Pdf/StructureTree then maps the elements to structure types and gives every text node, image and list marker a slot in that tree.
  3. Layout/LayoutEngine lays out blocks, inline text (line breaking, justification), lists and tables (automatic column widths), paginating as it goes, and produces a flat list of draw operations. Inline content with right-to-left text gets its embedding levels from Text/Bidi (UAX #9), and each line is put in visual order after it is broken. Fonts/ComplexShaper shapes Arabic and Hebrew runs with the font's GSUB and GPOS tables (Fonts/OpenTypeLayout).
  4. Pdf/PdfDocumentWriter writes PDF 1.4: content streams, fonts, image XObjects, link annotations, the bookmark outline and the cross-reference table. A tagged PDF's content is wrapped in marked-content sequences for its slots (or as artifacts), and the structure tree and its parent tree are written last. Fonts/FontSubsetter and Fonts/CffSubsetter cut embedded fonts down to the glyphs used (with their composite components and the CFF subroutines they call).

The renderer (Rendering/) works the other way round: Parsing/PdfFile reads the object graph, PageRenderer interprets content streams, Fonts/ turns glyphs into outlines (cached as anti-aliased bitmaps), Images/ decodes images, and Graphics/Rasterizer fills paths with exact area coverage before PngEncoder, JpegEncoder or WebPEncoder writes the result.

The third-party fonts, font data and code built into the library are listed, with their licenses, in THIRD-PARTY-NOTICES.md.

Edit this page on GitHub