Use if-watch instead of local_ip_address

[macladson]

Proposed Changes

This is an alternative to #8659 which just uses if-watch which we already depend on transitively through libp2p.

[macladson]

Actually I think this doesn't work.
I think this uses if-addrs under the hood anyway:

https://github.com/libp2p/if-watch/blob/master/src/fallback.rs

[macladson]

I think this is initialized empty and so we would have to force a resync somehow

[mergify]

Some required checks have failed. Could you please take a look @macladson? 🙏

[ackintosh]

Thanks @macladson! It seems preferable to use if-addrs directly, since that's what if-watch uses under the hood.

[macladson]

Thanks @macladson! It seems preferable to use if-addrs directly, since that's what if-watch uses under the hood.

I actually misunderstood the feature paths, it only uses if-addrs for Windows, so this might still be worth doing if we can fix the PR

[ackintosh]

Let me clarify: if-watch uses if-addrs on Windows and macOS, but not on Linux. On Linux, if-watch uses rtnetlink instead, despite if-addrs supporting Linux.

I guess the reason is that if-watch wants to receive events for network interface up/down, whereas if-addrs only provides a snapshot.

In our case, we simply want to get a list of network interfaces, so using if-addrs directly seems straightforward.

[jxs]

Let me clarify: if-watch uses if-addrs on Windows and macOS, but not on Linux. On Linux, if-watch uses rtnetlink instead, despite if-addrs supporting Linux.

I guess the reason is that if-watch wants to receive events for network interface up/down, whereas if-addrs only provides a snapshot.

In our case, we simply want to get a list of network interfaces, so using if-addrs directly seems straightforward.

still, if we can avoid another dependency and just use if-watch as it's transitive would be nice, is it possible?

[macladson]

So the code was panicking since tokio::IfWatcher requires a runtime. Adding this either to the function itself or piping in an existing runtime from elsewhere feels way overkill to me.

I have switched to smol::IfWatcher which doesn't need an existing runtime, and I think it technically works now. The downside is, the smol feature added a bunch of extra deps so it totally defeats the purpose of this change.

So with that in mind, I am heavily leaning in favour of the original PR with if-addrs (#8659).

Unless anyone has a cleaner solution that I missed?