Calascibetta Romain

> @dinosaure This issue doesn't make Dream diverge any more on http/af, etc., than it already has — it's only about using separate repos for code that is already vendored...

> If not, then we can't use the opam-repository http/af for now. From your message, it seems that WebSocket upgrades are the last thing missing. Can you confirm? Yes, I...

If you want a fast `trie`, you should take a look on [`art`](https://github.com/dinosaure/art) which is a fast implementation of an Adaptive Radix Tree.

`x86_32` is supported only if it supports SSE2 (but OCaml-CI provides one which does not handle SSE2). `arm64` should be supported (at least, we have a mapping between SSE and...

Yes of course, the state of the repo still is experimental, but I will add copyright and hyperlink to SSE2NEON 👍

Done by b9bb207, the issue still is open because the ARM support is not yet available.

ARM64 is supported 🎉 (see #33). We will start to look about PPC64.

decompress: - [scope & clash of name](https://github.com/mirage/decompress/commit/b758b2eefdb0f612bdf6020652819232dff8fa91) - [scope & clash of name](https://github.com/mirage/decompress/commit/4091ea34c1db163095fd7ad085e89ef0b48e23e5)

So it seems that the `Hashtbl` `connects` keeps IDs forever and we don't remove them even if we properly close the connection. At one point so, we infinitely execute `bumpport`...

@balrajsingh it seems that with your patch, the "finalizer" of a connection (which `Hashtbl.remove` the connection from `t.connects`) is never called. It's why, at one point, `mirage-tcpip` loops infinitely on...