Backport pr #1388 to v0.6

[Thomasdezeeuw]

Pr: #1388
Issue: #1386

[Thomasdezeeuw]

Mio 0.6.x itself is fine, but the problem is in the net2 crate: deprecrated/net2-rs#105.

[faern]

So when the net2 PR is merged and net2 0.2.36 is released. Let's bump mio so it says it requires at least net2 = 0.2.36? That way people will get it even if they later do cargo update -p mio in their projects.

Would be great if we could also do the same for tokio. Get a tokio 0.2.23 patch release published that forces the use of this newer mio.

My own experience is that selectively upgrading some dependencies is somewhat common. And we would like this patch to propagate as fast as possible.

[Thomasdezeeuw]

So when the net2 PR is merged and net2 0.2.36 is released. Let's bump mio so it says it requires at least net2 = 0.2.36? That way people will get it even if they later do cargo update -p mio in their projects.

Yes that sounds good.

Would be great if we could also do the same for tokio. Get a tokio 0.2.23 patch release published that forces the use of this newer mio.

I think we can do that, but I want some more prs merged before releasing Mio both 0.7.x and 0.6x.

[faern]

I'm thinking that maybe mio 0.6 is not completely without guilt. Or I have made a mistake. Because I try to run the mio test suite under my toolchain built on rust-lang/rust#78802. Without patching net2 I get invalid pointer access (to be expected) but if I build it on top of the net2 code in deprecrated/net2-rs#106 I still get test failures on Windows.

For example test_tcp::peek fails on

poll.register(&s, Token(1), Ready::readable(), PollOpt::edge()).unwrap();

with error kind AddrNetAvailable and the message "The requested address is not valid in its context.".

I have not yet been able to figure out why. I'm not at all knowledgeable in the mio codebase really.

[faern]

I don't think it's the net2 PR that is faulty. I have added tests to it where it prints out the c::sockaddr struct content as a byte slice. And with the PR code the results are identical to master for both IPv4 and IPv6 🤷

[faern]

Ah.. miow also does the same memory layout assumption: https://github.com/yoshuawuyts/miow/blob/c86373961b97eba692374435e93b81871af3f611/src/net.rs#L459-L470

I'll try to get PRs for that up during the weekend.

[faern]

This can be closed, since #1405 is merged and published.