TcpListener::bind/TcpStream::connect (&str, Int), &str and SocketAddr{...} overrides

[thomaslee]

cc @alexcrichton

In #13919 we changed TcpStream::connect to accept a host and a port parameter instead of a somewhat "uncomfortable" SocketAddr. There were some reasonable complaints following the merge that basically said sometimes you might want a SocketAddr for various different reasons, and that carting strings around everywhere might be painful in certain situations.

@huonw and @zargony suggested it might be better for TcpStream::connect and TcpListener::bind to use traits so that they can operate with various different types of arguments. I gave that a shot, and I think the results are really nice.

This change allows the following:

// (&str, Int)
let conn1 = try!(TcpStream::connect(("mozilla.org", 80)))

// &str (I think this is what @alexcrichton originally wanted the API to look like prior to #13919)
let conn2 = try!(TcpStream::connect("mozilla.org:80"))

// SocketAddr{...} (this is the old pre-#13919 API)
let conn3 = try!(TcpStream::connect(SocketAddr{ip: Ipv4Addr(63, 254, 215, 20), port: 80}))

It would also be trivial to add support for e.g. (IpAddr, Int) tuples too, but I didn't want to get too carried away :)

I think this PR needs some work before it gets merged. In particular, connect_timeout hasn't been modified, and I'm not sure that we want two distinct traits (ToBindSocketAddr and ToSocketAddrs) for very similar use-cases. Figure it's at a point where it'll probably drive some discussion, though.

[alexcrichton]

While it seems that this trait should indeed be private, I don't think it should be tagged with doc(hidden). Knowing the implementors of this trait will be very important for discoverability to what you can pass to connect.

[alexcrichton]

This looks quite interesting, thanks for working on this! If we go with this change, can you lift the relevant trait into the net module and use it with UDP as well?

[thomaslee]

Absolutely, makes sense. I'll hold off on the other proposed changes here until I get a go-ahead. I think all the changes you're proposing here make sense, though I guess we'll want to figure out what to do with (&'a str, Int) etc.

[zargony]

I like this. I think it's important to be able to use connect and bind without the overhead of a string and running the parser. Nice work! :)

[alexcrichton]

Closing due to inactivity, but feel free to reopen with a rebase!

We've started to grow generics like BytesContainer in more locations, so I would think that this would be ok to merge if my comments were addressed.