@exactjs/bun-adapter

Bun runtime adapter for eXact server endpoints and custom SSR page handlers.

Usage

import { createExactBunHandler } from '@exactjs/bun-adapter';

const exact = createExactBunHandler(exactRuntime);

Bun.serve({
	port: 3000,
	fetch: exact
});

Bun uses the Fetch API, so the adapter returns a standard Response while @exactjs/server remains responsible for protocol validation and dispatch. Buffered SSR responses pass their complete text to Bun's native Response. Produced SSR bodies use the streaming boundary.

Automatic request scheduling

The handler automatically trials bounded request-start batches under sustained load and retains them when native request capacity improves with responsive event-loop delay. When incoming demand limits throughput, a lower-lag policy can remain active while native responses keep draining and the event-loop thread has spare CPU. Runtimes without usable thread CPU accounting retain capacity-based trials. Sparse requests start immediately. Pass { adaptive: false } as the second factory argument to disable scheduling, or set maxBatchSize to change the default limit of 32 starts per callback.

For custom pages, wrap the complete Fetch dispatcher once:

import { createBunRequestHandler } from '@exactjs/bun-adapter';

Bun.serve({ fetch: createBunRequestHandler(renderPage) });

Forward both (request, server) when adding another callback around the handler. Scheduling uses Bun's host-wide pendingRequests counter, so route all HTTP requests through this Fetch dispatcher instead of combining it with Bun's native routes map. Calls without a server argument remain immediate. The outer handler owns scheduling when it dispatches to another eXact handler. Forward request.signal to SSR. String rendering inherits the adaptive policy; progressive rendering selects a cooperative work window at render entry and after pending component data settles. Both APIs use the same compiled components, and ready components continue synchronously. Initial Fetch admission still uses the host-wide adaptive controller, including on servers with mixed output modes. Responses are neither shared nor wrapped for scheduling; streaming and cancellation remain native.

Documentation | Source on GitHub