docuparse

High-performance document extraction library for Rust.

Extracts text and renders pages from PDF documents using the Pdfium engine — the same C++ library used by Google Chrome.

Features

• Native text extraction — extracts text directly from PDF text layers, no OCR required for digitally generated documents
• Page rendering — renders PDF pages to images at configurable DPI with optional annotation rendering and dimension clamping
• Automatic page classification — detects whether a page has a native text layer or is scanned
• PDF validation — magic byte checking, size guards, existence checks
• Document metadata — title, author, page count, file size, PDF version
• Zero intermediate files — everything in memory, no temp file I/O
• Pdfium singleton — initialised once, borrowed across your application

Installation

[dependencies]
docuparse = "0.0.1"

Docuparse links against Pdfium at runtime. You must provide the Pdfium shared library alongside your binary. Pre-built binaries are available from the pdfium-render releases.

Quick Start

use docuparse::{init_pdfium, PdfDocument};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let pdfium = init_pdfium()?;
    let doc = PdfDocument::open(pdfium, "document.pdf")?;

    println!("pages: {}", doc.page_count());
    println!("file:  {}", doc.metadata.file_size_display());

    for result in doc.extract_all_text_layers() {
        match result {
            Ok((page, Some(text))) => println!("page {}: {} chars", page + 1, text.len()),
            Ok((page, None))       => println!("page {}: scanned — no text layer", page + 1),
            Err(e)                 => eprintln!("page: error — {e}"),
        }
    }

    Ok(())
}

Page Rendering

use docuparse::{init_pdfium, PdfDocument, RenderConfig};

let pdfium = init_pdfium()?;
let doc = PdfDocument::open(pdfium, "document.pdf")?;

let config = RenderConfig::builder()
    .dpi(150)
    .with_annotations(true)
    .build();

let image = doc.render_page(0, &config)?;
image.save("page_1.png")?;

Batch Rendering

let images = doc.render_all_pages(&config)?;
for (i, image) in images.iter().enumerate() {
    image.save(format!("page_{:03}.png", i + 1))?;
}

Performance

Benchmarked on a representative mixed PDF document (native text layers + scanned pages) on Apple Silicon (M4):

Metric	Value
Text extraction (12 pages)	~8 ms
Per page	~680 µs
Memory RSS	~25 MB
Child processes spawned	0
Temp file I/O	none

Performance is dominated by the Pdfium C++ engine. The Rust wrapper contributes negligible overhead — dev and release builds measure identically, confirming zero Rust-level bottleneck.

Error Handling

All public functions return Result<T, PdfError>. Errors are fully typed via thiserror and compose cleanly with anyhow or eyre in application code:

use docuparse::PdfError;

match PdfDocument::open(pdfium, "file.pdf") {
    Err(PdfError::Validation(_)) => {
        eprintln!("not a valid PDF");
    }
    Err(e) => eprintln!("error: {e}"),
    Ok(doc) => { /* ... */ }
}

Roadmap

• OCR support via ONNX Runtime (feature flag)
• Image input (JPEG, PNG) alongside PDF
• Async API

License

MIT