GeoRepair

This crate is experimental. The API is actively evolving — expect breaking changes between 0.x releases. Core algorithms, I/O backends, and feature flags are all subject to change as we improve correctness and performance.

OGC geometry repair and validation for Rust. Built-in I/O for WKB, WKT, and a custom binary batch format — no extra dependencies required. Detects and fixes invalid GIS geometries (self-intersections, unclosed rings, degenerate shapes, NaN coordinates, and more) using algorithms selected by geometry type.

The Structure strategy (default) mirrors GEOS's ST_MakeValid algorithm: planar graph extraction, face walking, and winding-number assembly. The Arrange strategy uses CDT-based repair as a robust fallback for complex topologies. Passes 2490/2490 GEOS XML validation tests, with parallel batch performance roughly on par with GEOS on 1.58M data set polygons (full dataset 0.92×, validation 3.3×).

See the full documentation for quick-start examples, validation rules, CRS support, I/O backends, Python bindings, C FFI, and known limitations.

Performance

Real-world dataset (1,578,988 polygons)

Structure batch on a production GIS dataset. GEOS linked via conda-forge (MSVC, serial internally, no LTO). i5-12400F (6C/12T), mimalloc. All columns are parallel batch (Rayon 12 threads); serial included for reference.

Synthetic benchmarks (CoordSeq direct — no WKT overhead)

Structure strategy, i5-12400F (6C/12T). GEOS benchmark uses CoordSeq direct construction, NOT WKT — this is a fair comparison. Serial column is single-threaded; parallel column uses Rayon 12-thread batch.

Stars indicate improvement: *** ≥100× · ** 10–100× · * 1–10× · (blank) <1× (GEOS faster)

Arrange pipeline (CDT fallback):

Notes:

• conda-forge libgeos on Windows is MSVC, serial per-call, no LTO. "par batch" = many GEOS C calls run concurrently via Rayon (12 threads).
• Synthetic benchmarks use CoordSeq direct GEOS construction (no WKT). WKT round-trip inflated earlier GEOS numbers by 10-120×. Real-world benchmark always used CoordSeq (fair).
• GeoRepair parallel speedup is limited on sub-5µs geometries (Rayon overhead dominates). Real-world batches of 1.58M polygons show better throughput scaling (2-3× vs serial).
• GeoRepair wins big on invalid polygon repair and MultiPolygon unification (C++ quadratic algorithms hit their worst case). GEOS wins on large valid polygons where C++ optimization and mature JTS codebase dominate.

Run benchmarks

# Real-world dataset benchmark (system GEOS — conda-forge)
cargo bench --features bench-geos-system,arrange,structure,parallel,simd,io-shp --bench real_world

# Real-world dataset benchmark (static GEOS — built from source)
cargo bench --features bench-geos,arrange,structure,parallel,simd,io-shp --bench real_world

# Synthetic benchmarks with serial + parallel columns (no GEOS)
cargo bench --features arrange,structure,parallel,simd --bench bench

# Synthetic benchmarks with GEOS comparison
cargo bench --features bench-geos-system,arrange,structure,parallel,simd --bench bench

# Criterion microbenchmarks
cargo bench --features bench-criterion --bench criterion

Features

License

Apache-2.0