๐Ÿ“ฆ bradfitz / exp-httpclient

experimental new HTTP client API for #golang (WIP)

โ˜… 586 stars โ‘‚ 12 forks ๐Ÿ‘ 586 watching
๐Ÿ“ฅ Clone https://github.com/bradfitz/exp-httpclient.git
HTTPS git clone https://github.com/bradfitz/exp-httpclient.git
SSH git clone git@github.com:bradfitz/exp-httpclient.git
CLI gh repo clone bradfitz/exp-httpclient
Brad Fitzpatrick Brad Fitzpatrick Link slides from problem page 53da77b 7 years ago ๐Ÿ“ History
๐Ÿ“‚ master View all commits โ†’
๐Ÿ“ http
๐Ÿ“ httpclient
๐Ÿ“„ problems.md
๐Ÿ“„ README.md
๐Ÿ“„ README.md

Rethinking Go's HTTP client

This repository explores redesigning the API for the Go language's net/http Client and Transport.

Initially, though, it collects problems with the current API. The actual solution has not yet been designed.

FAQ

What's wrong with Go's HTTP client?

See the list of problems.

Or see the overview presentation for the problems in a different format.

What about the Server?

This repo does not aim to address the server side of the net/http package. The server half is in better shape than the client, and it's also easier to fix the client half without fragmenting the ecosystem. Changing the Server interface needs to be done much more carefully.

But even long term, it's almost certainly best for the client and server to live in separate packages. They might share some types & code from shared HTTP package(s).

Who's leading this effort?

Brad Fitzpatrick, @bradfitz. I've owned the net/http package for over 8 years and have plenty of gripes about it. I welcome all input. If we're going to finally change it, we should get it right, so there's no need to rush this process.

Contributing

This repo is temporary and doesn't accept PRs and issues are disabled. It will move at some point to Go's repos with Go's bots and policies.

For now, discuss at https://github.com/golang/go/issues/23707

What's the plan?

Roughly:

  • Iterate on the API & godoc repeatedly until it looks right (with a fake, panic("TODO")-only implementation)
  • Discuss, revise.
  • Add a temporary implementation (likely inefficient), wrapping the existing net/http Client.
  • Port code to use it. See if we're still happy.
  • Discuss, revise.
  • Copy net/http and golang.org/x/net/http2 code into httpclient (likely several packages).
  • Benchmark, tune, revise API as needed.
  • Redo the "legacy" net/http and golang.org/x/net/http2 client APIs as wrappers around httpclient
Of course, this is all up for debate.