commit 5c3c104ba0f843b9c46a921c8d6c2bd6c068acd2
author deadbeef <deadbeef@webrtc.org> Fri Aug 04 15:01:57 2017 -0700
committer Commit Bot <commit-bot@chromium.org> Fri Aug 04 22:01:57 2017 +0000
tree 62cba79b42be521599e06be10f029f8e64d3b404
parent d48f56de1f06d62fcb46bb9f20bd5f336e80aae9

Make Port (and subclasses) fully "Network"-based, instead of IP-based.

For ICE, we want sockets that are bound to specific network interfaces,
rather than to specific IP addresses. So, a while ago, we added a
"Network" class that gets passed into the Port constructor, in
addition to the IP address as before.

But we never finished the job of removing the IP address field, such that
a Port only guarantees something about the network interface it's
associated with, and not the specific IP address it ends up with.

This CL does that, and as a consequence, if a port ends up bound to
an IP address other than the "best" one (returned by Network::GetBestIP),
this *won't* be treated as an error.

This is relevant to Android, where even though we pass an IP address
into "Bind" as a way of identifying the network, the socket actually
gets bound using "android_setsocknetwork", which doesn't provide any
guarantees about the IP address. So, if a network interface has multiple
IPv6 addresses (for instance), we may not correctly predict the one
the OS will choose, and that's ok.

This CL also moves "SetAlternateLocalAddress" from VirtualSocket to
VirtualSocketServer, which makes for much more readable test code.

The next step, if there is one, is to pass along the Network class all
the way to SocketServer::Bind. Then the socket server could do smart
things with the network information. We could even stick a platform-
specific network handle in the Network object, such that the socket
server could use it for the binding, or for "sendmsg", for example.
See bug 7026 for more context about the sendmsg idea.

BUG=webrtc:7715

Review-Url: https://codereview.webrtc.org/2989303002
Cr-Commit-Position: refs/heads/master@{#19251}