Performance

All benchmarks run on Apple M3 Pro using mitata. Numbers will vary on your hardware -- run them yourself to get a feel.

Operations

Lookups are O(1) regardless of registry size. Transforms scale linearly with chain depth at roughly ~140ns per hop. The graph is built once at registry creation, so path resolution doesn't rebuild anything.

Tradeoffs

Every feature has a cost. Here's what each one adds.

Hooks with empty callbacks add about 19% -- the timing hooks (onTransform, onStep) add performance.now() calls and per-step object allocations. When no timing hooks are registered, doba skips all of that entirely. The pipe builder is about 2x slower than a hand-written migration function because it runs each step (rename, drop, add, map) as a separate object spread. If that matters in your hot path, use bare functions. For most cases the ergonomics are worth it.

Providing an explicit path skips BFS entirely and saves about 52% on longer chains (2.0us vs 4.2us at 25 hops). Worth doing if you're calling the same transform repeatedly and already know the route.

Scaling

Edge cases

Errors fail fast. An unknown schema returns in ~103ns without touching the graph. A missing path is slower (~668ns) because BFS has to exhaust the search before giving up.

Recursive structures

These measure the cost of migrating complex nested data through a single transform. The time is dominated by the migration function itself, not doba's overhead.

Run it yourself

bun run bench              # core operations
bun tests/features.bench.ts   # feature overhead, explain, pipe, hooks
bun tests/stress.bench.ts     # extreme scale, recursive structures