Expose IEEE802.15.4 address in Driver

[rubdos]

(I'm sorry for the PR-spam. I'm trying to get the "finished" things we have cleaned up, atomic and mergeable.)

This allows a net::Driver to be an IEEE802.15.4 device.

I wrote this in two stages.

1. Added ieee802154_address() to the trait
2. Merged ethernet_address() and ieee802154_address into hardware_address().

This allows the community (mainly @Dirbaio, cc @xoviat) to decide which way is better. HardwareAddress means we have a bunch of match-statements at runtime, for a property that's typically known at compile time (but that's the case anyway with smoltcp, currently), but it's a lot more compact and future proof. It's a (big-ish) breaking change, however.

CI will probably complain, but I'll clean that up after a decision on the above is taken :-)

[xoviat]

I'll take a look at this later if someone else doesn't. I want to clarify that the main thing I care about is that the implementation remains zero-copy, as is currently done in the runner implementation. I don't really have a strong opinion on changes that aren't performance-relevant.

[rubdos]

the implementation remains zero-copy

That's also a requirement for us, although not strictly in an initial phase. I don't think this change specifically is really concerning regarding zero-copy. Thanks for weighing in!

[xoviat]

Yeah I don't think this is performance relevant, I was just discussing in general.

[xoviat]

I just looked and I agree that hardware address is the away forward. Make the CI pass and we can merge.

[xoviat]

Re-export the hardware address type though

[Dirbaio]

embassy-net-driver should not depend on smoltcp. This is so smoltcp can be upgraded without requiring every driver crate to update. Can you copypaste the HardwareAddress enum instead?

[rubdos]

Can you copypaste the HardwareAddress enum instead?

IMO that loses part of the meaning of using HardwareAddress, since the whole purpose is to be able to use it as-is with smoltcp. It would be cool if we could just depend on smoltcp_wire instead, I suppose, although even that doesn't alleviate the concern.

I get the concern however, it felt pretty awkward to put smoltcp as a dependency.

Another option that I didn't mention in my OP, is to fix the hardware address type at compile time through a generic.

If after my plea here you still rather have the definition copied, I'll happily do that!

[rubdos]

I slept on it, and I think I agree. So I'll rewrite it and fix CI. If we're happy with the result, I'll squash the results such that Cargo.toml remains unchanged.

[rubdos]

Re-export the hardware address type though

I could reuse the one from embassy_net, I suppose, since it's reexported there already.

[xoviat]

You can make hardware address an enum with ethernet and ieee802154 variants. In any case, a driver should only need to have embassy-net-driver as a dependency in cargo.toml, nothing else.

[Dirbaio]

could you add an Ip variant with no contents, like in smoltcp? otherwise drivers with IP medium have to return a "dummy" hardware address which is a bit awkward.

(I know this was already "wrong" before, but now that you're changing this it would be the ideal time to fix it. Thank you! 🙏 )

[rubdos]

diff --git a/embassy-net/src/lib.rs b/embassy-net/src/lib.rs
index 746acae4..dbefb0ef 100644
--- a/embassy-net/src/lib.rs
+++ b/embassy-net/src/lib.rs
@@ -252,23 +252,7 @@ impl<D: Driver + 'static> Stack<D> {
         resources: &'static mut StackResources<SOCK>,
         random_seed: u64,
     ) -> Self {
-        #[cfg(any(feature = "medium-ethernet", feature = "medium-ieee802154"))]
-        let medium = device.capabilities().medium;
-
-        let hardware_addr = match medium {
-            #[cfg(feature = "medium-ethernet")]
-            Medium::Ethernet => to_smoltcp_hardware_address(device.hardware_address()),
-            #[cfg(feature = "medium-ip")]
-            Medium::Ip => HardwareAddress::Ip,
-            #[cfg(feature = "medium-ieee802154")]
-            Medium::Ieee802154 => to_smoltcp_hardware_address(device.hardware_address()),
-            #[allow(unreachable_patterns)]
-            _ => panic!(
-                "Unsupported medium {:?}. Make sure to enable it in embassy-net's Cargo features.",
-                medium
-            ),
-        };
-        let mut iface_cfg = smoltcp::iface::Config::new(hardware_addr);
+        let mut iface_cfg = smoltcp::iface::Config::new(to_smoltcp_hardware_address(device.hardware_address()));
         iface_cfg.random_seed = random_seed;

         let iface = Interface::new(

Including this, I suppose.

[Dirbaio]

yup! :D

[rubdos]

I suggest I still squash some things up here, but this is the general idea, right? :-)

[Dirbaio]

thank you!