Bun Performance

Define the workload and claim: startup, CPU cost, retention, throughput, or latency. Measure a baseline before optimizing. Select the changed path from profiles or measured workload frequency.

Read [measurement and memory][ref-1] for CPU/heap commands, JSC warm-up, retainers, and RSS. Read [native APIs and concurrency][ref-2] for files, byte ownership, streaming, workers, and bounded queues.

Remove measured redundant work, allocations, copies, or expensive calls. Maintain error, ordering, encoding, lifetime, streaming, and cancellation contracts. Account for backing-store retention when replacing copies with views.

Compare repeated before/after runs with identical runtime, build, inputs, and hardware conditions. Run correctness checks for changed behavior. Report distributions, errors, and measurement scope. Label unmeasured proposals as hypotheses.