Changelog

v0.3.3

Released 2026-05-06.

This release reduces allocation overhead in the rasterization hot paths.

• line rasterization now precomputes contiguous coordinate and offset buffers before entering the Numba kernels

• polygon rasterization uses the same precomputed coordinate layout, which avoids repeated geometry unpacking

• both line and polygon rasterization prefilter inputs against the target bounding box earlier, so fewer geometries reach the kernels

• the Numba line traversal loop now advances grid crossings incrementally instead of recomputing the next crossing from scratch on every step

• polygon rasterization reuses scratch buffers inside the exact and hybrid engines, which cuts per-geometry allocation churn

The net effect is lower memory churn and better throughput when rasterizing many geometries or larger inputs.