@KhafraDev wdyt?

Sounds fine to me.

Thank you for your timely responses, @mcollina, @KhafraDev!

I think a utility function like this would make the most sense:

export function decompress(response: Response) {
  const decoders = []
  // ...populate the `codings` array based on the passed response.

  // Expose the entire pipeline since `pipeline` itself is likely
  // too low-level to export from Undici. These would also be called
  // together anyway. 
  return pipeline(response.body, ...decoders, () => {})
}

I'd love to hear your thoughts on whether response instance is enough to decide that. I can see request references in that code snippet above as well.

#3274 might solve this, I'd need to see some uses for a standalone decompress.

On the consumption side of things, it would be nice if this behavior was imported from Node.js to be consistent with whichever Undici version is used in a particular Node.js version. In other words, so it would always align with the Node.js version used by the consumer of Interceptors.

Granted, the decompression logic isn't supposed to change often. I'm mostly thinking bug fixes here.

@KhafraDev, would that allow us to use decompression without any sort of Undici interceptor being created? That'd be ideal. The functionality itself doesn't need an interceptor, it can be a utility function that deals with the body transformations based on the Content-Encoding header. Compressed body in, properly decompressed body out. Then, the same utility can be reused in the interceptor pull request you've mentioned, I'd hope.

I think we need to split compression and decompression into separate interceptors and add decompression parameters to request options the decompress: boolean, compress: boolean. Also enable them by default

import { request } from 'undici';

async function test(){
  const resp = await request('https://example.site/', {
    method: 'POST',
    body: `qwerty123`,
    
    decompress: true,
    compress: true
  });
  
  // decompressed response data
  return await resp.body.text();
}

OR

import { request, Agent, interceptors } from 'undici';

const dispatcher = new Agent().compose(interceptors.decompress()).compose(interceptors.compress(['gzip', 'br']));

async function test(){
  const resp = await request('https://example.site/', {
    method: 'POST',
    body: `qwerty123`,
    
    decompress: true,
    compress: true,
    dispatcher: dispatcher
  });
  
  // decompressed response data
  return await resp.body.text();
}

@PandaWorker, as long as the decompression is exposed as a standalone utility. This is what this proposal is about. The way we mean to use it in Interceptors (which are not Undici interceptors, pardon for the naming confusion) is with arbitrary response bodies. What I'm proposing here is likely more low-level than what you have in mind, so two may combine, in the end.

If the goal is to expose the decompression as utility #3274 won't be what might close this issue, as it attempts to expose it through an interceptor to be used with dispatcher.compose.

They are slightly interconnected but not necessarily the same (specially because this issue aims to have an utility that works with WHATWG Response object I assume). So a PR aiming just for that will be the best

@metcoder95, agreed. #3274 can reuse the utility function I am bringing up as a part of this suggestion. I'm glad to hear we are aligned on that!

If there are no objections, I can give this a go when I have a minute (likely next week). A code review and a bit of guidance will be appreciated.

Got excited by this, started with it sooner. The pull request is at #3423, extremely early.

Hey there :)
Can I help merge this in?

Hey, @mikicho 👋 Yes! I'm short on time these weeks, but I'm keeping a list of remaining todos on the pull request: #3423.

I could use some help with type-testing the newly added utilities and looking into this odd issue that somehow unrelated tests are failing now just because I moved the utilities under a different directory.