NavSearch

class NavSearch

import { NavSearch } from '@driftengine/nav';

In depth

Reused rather than constructed per search, because the scratch is proportional to the graph and a world with a dozen agents re-pathing on a schedule would otherwise allocate a dozen arrays the size of the map every time somebody changed their mind. One of these per thread of control; a search is not re-entrant and says so.

The route is a function of the graph and the two endpoints, and of nothing else. Two runs, two machines, a replay: same nodes out. That takes more than avoiding randomness — a heap pops ties in whatever order it happens to hold them, and two equally good routes through a symmetric world are exactly the case a road network produces constantly. So ties break on the node index, which is arbitrary but stable, and the sequence is frozen the same way the engine's RNG is.

The heuristic is straight-line distance, which is admissible precisely because buildNavGraph refuses an edge cheaper than its own span. That refusal is what makes the first route this returns the cheapest one rather than merely a good one, and it is why it is a build error rather than a note.

Constructor

constructor(graph: NavGraph)

Accessors

Methods

find(from: number, to: number, outPath: Uint32Array): number