Docs / Rendering Pipeline & Cache

Rendering Pipeline & Cache

BarePDF’s rendering engine (crates/barepdf-render) uses demand-driven work and bounded caches to keep continuous scrolling responsive on large documents without allowing bitmap usage to grow with page count.

Demand-Driven Architecture

  1. Viewport Intersection: The layout engine (layout.rs) computes which page indices overlap the visible window viewport plus a 1-page prefetch margin.
  2. LRU Cache Query: The render scheduler checks the LRU bitmap cache for pre-rendered RGBA buffers matching the requested (PageIndex, RenderDimensions, ZoomFactor).
  3. Cache Hit: Bitmaps are immediately transferred to the Slint frame buffer for instant display.
  4. Cache Miss & Work Dispatch: Unrendered pages are sent to the single background PDFium actor. Control/visible work uses the high-priority channel; bounded background work never blocks the UI thread.

Generation Token Cancellation

When a user scrolls rapidly:

  • The UI increments the global ViewportGeneration token.
  • All queued render tasks, including visible work, check their document and generation identity before rasterization.
  • Stale render tasks (pages already scrolled past) cancel immediately, freeing CPU cores for newly visible pages.

Memory Budgeting

The raw RGBA cache uses a fixed 32MB byte budget. The application separately budgets Slint page images (16MB) and thumbnails (4MB), preventing duplicated raw/UI copies from growing with page count.