Compression

Deserve ships no compression middleware, and that gap is filled twice over before a request ever reaches application code.

Why It Is Not Built In

Compression belongs to the layers around the handler, not inside it. The runtime does it on the way out, and in production a proxy usually did it first. A middleware that re-runs the same work would only add cost and a chance to clash with the encoding those layers already set.

Two layers already cover it:

• The runtime - Deno.serve compresses response bodies on its own, in line with build on the platform.
• The network edge - most production apps sit behind a proxy like Cloudflare or nginx, where compression is the network layer's job, not the application's.

So a compress middleware earns its keep only on a bare localhost with no proxy in front, and even there the runtime has it handled.

What Deno Does on Its Own

Deno.serve reads the request Accept-Encoding header and compresses the response body when these hold true:

• The header advertises br for Brotli or gzip, and the quality value preference is respected.
• The response Content-Type is one of the compressible types.
• The response body is larger than 64 bytes.

When it compresses, the runtime sets Content-Encoding to the chosen scheme and adjusts the Vary header so caches key on the encoding. Nothing in the handler needs to ask for this.

A response is left uncompressed when any of these is present, which is the runtime stepping aside on purpose:

• A Content-Encoding header, meaning the body is already encoded.
• A Content-Range header, meaning a range request is in play.
• A Cache-Control: no-transform header, meaning no layer should rewrite the body.

Letting the Runtime Compress

The default path is to send a normal response and leave the encoding alone. A plain JSON or text body flows through, and the runtime compresses it when the client supports it.

export function GET(ctx: Context): Response {
  // Send plain, runtime compresses when able
  return ctx.send.json({ message: 'compressed by the runtime' })
}

Opting a Response Out

To keep one response uncompressed, set a header the runtime treats as a stop signal. A Cache-Control: no-transform tells both the runtime and any downstream proxy to leave the body untouched.

export function GET(ctx: Context): Response {
  // Block any layer from rewriting the body
  ctx.setHeader('Cache-Control', 'no-transform')
  return ctx.send.json({ message: 'sent verbatim' })
}

Setting headers through ctx.setHeader is the same path used everywhere else, so this opt-out reads like any other header.

Already-Encoded Bodies

A body that arrives pre-compressed, such as a stored .gz asset, carries its own Content-Encoding. The runtime sees that header and skips re-compression, so the bytes ship as-is. For files on disk this is handled by static serving, which sets the type and lets the runtime decide.

Behind a Proxy

In a deployment fronted by Cloudflare or nginx, the proxy negotiates compression with the client at the edge, often before the request reaches the origin. The origin can still send a normal response, and the layers agree on the encoding through the Accept-Encoding and Vary headers. The application stays out of it, which is the whole point.