Thanks @oliverbarnes! Also, thanks for the code reviews @laserlemon and @mislav.

Hey guys, sorry for not responding sooner, I see this has already been implemented and merged.

I don't follow what's bad in memoizing the connection, though. from our exchange with @mislav, I understood it didn't matter:

It makes sense to pass connection options to connection even if it's going to be memoized, if these options aren't expected to change (such as SSL settings, endpoint, etc.).

The endpoint option actually is a request-specific option. Calling the search method hits an entirely different endpoint than fetching your user.

@laserlemon: One Faraday Connection per each endpoint is kept.

This isn't needed, however. If I were designing this, I would still have just one Connection object to the main endpoint, and change the hostname of the endpoint per each request that's not to the main one.

@mislav Maybe I'm misunderstanding you, but your link demonstrating that one connection is kept per endpoint is because of my very recent pull request.

Also, I agree with your design idea. I'd love it if the connection was just a one-off method and the Faraday middleware could switch the endpoint/hostname based on the incoming request. Unfortunately, how to do that isn't immediately coming to mind. But the endpoint option is the only real option remaining so it feels like we're very close!

@laserlemon: I didn't know you were responsible for it. I just looked at the master version, and assumed it was like that for longer.

Here, I implemented your wish. It wasn't that hard.

Cool! 👏