zpdf

Pure-Rust PDF parsing and rendering, with interchangeable CPU (tiny-skia) and GPU (wgpu) renderers whose output matches within <1% of pixels.

Features

• Pure Rust — zero C/C++ dependencies, fully safe.
• PDF parsing — header, traditional xref + xref/object streams + hybrid /XRefStm, trailer chains, lazy xref repair, object model, stream filters (Flate / LZW / ASCII85 / ASCIIHex / RunLength / DCT / CCITT G3-G4 + predictors) with corrupt-stream salvage.
• Encryption — RC4 (40/128-bit) and AES-128 / AES-256 (V5 R5/R6) standard security handler with crypt filters (empty user password).
• Content interpretation — graphics state, paths, clipping, text (incl. render modes and rise), inline & XObject images, Form XObjects (full resources, /BBox clip), axial/radial shadings, shading patterns, all 16 blend modes, dash patterns.
• Color — DeviceGray/RGB/CMYK, ICCBased (/N), Indexed, Lab, Separation/DeviceN via a full PDF function evaluator (types 0/2/3/4).
• Fonts — embedded TrueType, Type1, Type1C/CFF, CID/Type0 (Identity-H, /W, /CIDToGIDMap streams), Type3, the standard-14 fonts; encodings + /Differences; Quartz-subset recovery; /ToUnicode text extraction.
• Images — 1/2/4/8/16-bpc, /Decode, soft masks, stencil & color-key masks, Indexed palettes, CMYK JPEG; bilinear sampling with box-filter minification.
• Page geometry — CropBox-aware rendering, page-tree attribute inheritance (/Rotate, /Resources, boxes), page rotation.
• CPU rendering — tiny-skia backend, PNG output at any DPI.
• GPU rendering — wgpu backend (fills, strokes, clips, text, images, blend groups); matches the CPU renderer within <1% pixels.
• Tooling — CLI (info/render/text/compare/dump/debug-stream) and an interactive winit viewer.

Documentation

• docs/user-guide.md — the zpdf command-line tool.
• docs/library.md — using zpdf as a Rust library + architecture.
• docs/CHANGELOG.md — release notes.
• ROADMAP.md — development plan.

Quick start

# Inspect a document
cargo run -p zpdf-cli -- info document.pdf

# Render page 1 at 150 DPI (CPU)
cargo run -p zpdf-cli -- render document.pdf -p 1 -o out.png --dpi 150

# Render on the GPU (requires the `gpu` feature)
cargo run -p zpdf-cli --features gpu -- render document.pdf -p 1 -o gpu.png --backend wgpu

# Extract text, and compare two renders
cargo run -p zpdf-cli -- text document.pdf -p 1
cargo run -p zpdf-cli -- compare out.png gpu.png --out diff.png

# Interactive viewer (pan/zoom/page-flip)
cargo run -p zpdf-render-wgpu --example viewer -- document.pdf

Library usage

use zpdf::{ContentInterpreter, ImageCache, PdfDocument, RenderBackend};

let data = std::fs::read("document.pdf").map_err(zpdf::Error::Io)?;
let doc = PdfDocument::open(data)?;

let page = doc.page(0)?;                        // 0-based
let mut fonts = doc.load_page_fonts(&page);
let mut images = ImageCache::new();
let content = doc.page_content_bytes(&page)?;

let display_list = ContentInterpreter::new(page.effective_box()) // CropBox ∩ MediaBox
    .with_page_rotation(page.rotate)
    .with_fonts(&mut fonts)
    .with_document(doc.file(), &page.resources)
    .with_images(&mut images)
    .interpret(&content);

let mut renderer = zpdf::cpu::CpuRenderer::new()
    .with_fonts(&fonts)
    .with_images(&images);
let page_img = renderer.render_display_list(&display_list, 150.0 / 72.0)?; // 150 DPI
page_img.save_png("out.png")?;

Switch zpdf::cpu::CpuRenderer for zpdf::gpu::WgpuRenderer (with features = ["gpu-render"]) to render on the GPU — everything upstream is identical. See docs/library.md.

Architecture

13-crate workspace with a strict one-direction dependency flow. Render backends depend only on zpdf-display-list, never on the parser — parsing and rendering stay fully decoupled.

zpdf-core            Shared types: ObjectId, PdfObject, Matrix, Rect, Error, ParseLimits
  ├─ zpdf-parser     Lexer, xref/trailer, object & stream decoding, filters
  │   └─ zpdf-document   Catalog, page tree, resource inheritance, font loading
  │       └─ zpdf-content   Content-stream interpreter → DisplayList
  ├─ zpdf-font       Type1 / TrueType / CID / Type3 fonts, CMap, encodings
  ├─ zpdf-image      JPEG / Flate / CCITT / masks / palettes → RGBA
  ├─ zpdf-color      Device / Indexed / Lab color + PDF function evaluator
  ├─ zpdf-display-list   Flat RenderCommand sequence (the backend contract)
  ├─ zpdf-render          RenderBackend trait
  │   ├─ zpdf-render-cpu   tiny-skia backend
  │   └─ zpdf-render-wgpu  wgpu backend (+ viewer example)
  ├─ zpdf-cli        CLI tool
  └─ zpdf            Facade crate (re-exports; feature-gates cpu / gpu)

Supported PDF features

Dependencies

All pure Rust:

Development

cargo build                                   # CPU-only (default)
cargo build --features gpu-render             # include the wgpu backend
cargo test                                    # all tests
cargo test -p zpdf --features gpu-render      # + the GPU↔CPU acceptance harness
cargo clippy --workspace

Roadmap

See ROADMAP.md.

• Phase 1 — PDF parsing — done
• Phase 2 — Content interpretation + CPU rendering — done
• Phase 3 — wgpu GPU rendering — done
• Phase 4 — Advanced features — largely done (encryption incl. AES, shadings, blend modes, spot color, CropBox/rotation); remaining: soft masks, tiling-pattern cells, annotations, ICC profiles, JBIG2/JPX, CJK fallback

License

MIT