/**
 * @hidden
 * @packageDocumentation Benchmarks for all in-process {@link IOStream} implementations.
 *
 * ### Why async is slower than sync
 *
 * Every `await` suspension — even of an already-resolved Promise — forces the
 * JavaScript engine to schedule a **microtask** and resume the caller on the
 * next checkpoint.  In Node.js / V8 this costs roughly **60-70 ns per
 * `await`** (verified empirically in this environment).  For format parsers
 * that perform thousands of individual stream reads, the overhead accumulates:
 *
 * | reads per parse | extra latency |
 * |----------------:|--------------:|
 * |             100 |         ~7 µs |
 * |           1 000 |        ~70 µs |
 * |          10 000 |       ~700 µs |
 * |         100 000 |         ~7 ms |
 *
 * This is **inherent** to the async/await abstraction — it affects all stream
 * implementations equally because the overhead is in the `await` call itself,
 * not in the underlying I/O.  The sequential-read benchmark below makes this
 * visible: all implementations converge to the same throughput because 8 192
 * microtask suspensions per 1 MiB iteration dominate every other cost.
 *
 * The key mitigation: **read the largest possible block at once** in hot code
 * paths, reducing the total number of `await` calls.
 *
 * ### BlobStream cache behaviour
 *
 * `BlobStream` uses a **piece table** with per-segment fetch caching:
 * - First read of a segment issues `blob.slice().arrayBuffer()` and stores
 *   the resulting `Uint8Array` on the segment.
 * - All subsequent reads of the same range are served from the cache.
 * - When one `readBlock` spans multiple uncached segments, all
 *   `arrayBuffer()` calls are issued in parallel via `Promise.all`.
 *
 * The "warm" benchmarks below share a single pre-loaded stream across all
 * iterations (lazy-initialised on the first call).  The first iteration pays
 * the cold cost; all subsequent ones are served purely from the cache.
 *
 * Run with:
 * ```
 * npx vitest bench
 * ```
 */
export {};
//# sourceMappingURL=iostream.bench.d.ts.map