Pathetic

Uses hand-rolled primitive binary heaps and compact coordinate encoding to eliminate object allocations, drastically reducing GC pressure and improving cache locality, as benchmarks show 4.5x faster performance for 1k nodes.

Implements Bloom filters for O(1) closed-set lookups, accelerating node checking compared to traditional O(n) approaches and handling thousands of concurrent requests efficiently.

Runs multiple heuristics like Manhattan and Octile in parallel, allowing choice between linear accuracy or squared speed for optimal pathfinding in real-time scenarios.

Fully asynchronous and concurrent architecture avoids main-thread stalls, enabling seamless handling of high-load pathfinding without performance degradation.