Wifi device refactor

[MabezDev]

Currently, the WifiDevice Is heavily tied to the smoltcp stack. This PR implements embedded_svc::wifi::Wifi on the raw device instead of on the stack to allow usage of other stacks (i.e embassy_net). I have included an implementation that just calls the inner device functions for compatibility but I can remove that if needed. Finally, I added the internal error codes from esp-idf and reworked the error types slightly. I'm sure we could get Bingen to generate this for us, but we only really need it for a few things so I just hand-rolled it for now.

[bjoernQ]

The static_ip example needs an update just like the dhcp example

Unfortunately, these changes seem to trigger mis-compilations for me

• ESP32-S3: BLE example crashes with IllegalInstruction - opt-level 1 fixes that
• ESP32-S3: DHCP example gets stuck in WifiScan - opt-level 1 fixes that
• ESP32-S3: DHCP example crashes with IllegalInstruction when awaiting the IP address - with all opt-levels

And one totally irrelevant nit-picky thing: Maybe the commit message of the last commit is a bit mis-leading?

[MabezDev]

Went a bit overboard and overhauled the errors in the wifi module to use a result and fixed the examples. I can't reproduce any of those miscompilations though :(. Maybe try cargo update? That's what fixed my issues earlier.

[bjoernQ]

Very nice improvements to the error handling

Unfortunately, BLE doesn't compile anymore

Even cargo update and cargo clean don't make it work for me 😢

On S3 in opt-level 3 the dhcp and static_ip examples just freeze at the wifi scan. Oddly enough the coex example gets the scan done but never connects

On S2 in opt-level 2 the dhcp example runs into an illegal instruction while the static_ip example works.

With opt-level 1 everything seems to work on S3.

But on S2 the dhcp example still crashs.

I think needed opt-level 1 is not a big deal but I am a bit concerned I never can make the dhcp example work on S2

Maybe @jessebraham can also give this a try

[MabezDev]

Unfortunately, BLE doesn't compile anymore

Oops fixed.

On S3 in opt-level 3 the dhcp and static_ip examples just freeze at the wifi scan. Oddly enough the coex example gets the scan done but never connects

That was the behaviour I saw yesterday then a cargo update magically fixed it 🤷

I can reproduce the S2 issue, which goes away with lto = 'off' under the release profile. Which by the way, is different to lto = false see here: https://doc.rust-lang.org/cargo/reference/profiles.html#lto

[MabezDev]

To be honest, I think we should be setting LTO = false (easy to get confused 🤦 ) LTO = 'off' always for the time being (on esp-hal too), until we know the codegeneration from LLVM is more reliable, there are a few buggy optimizations that LTO may be applying here even when we don't want it to.

[MabezDev]

Out of curiosity I also just retested #112 (comment) (S2 static IP opt-level = 3) with lto = 'off' and it now works...

Disabling LTO completely is obviously not the true fix, but until it's fixed in the backend I don't really see any reason to have it enabled.

[bjoernQ]

Great - lto = 'off makes it work for me - only BLE example on C2 fails with that but we can just state that in the README like we did before with the opt-levels .... So, if you don't mind to change that this LGTM

[bjoernQ]

Given that #116 is already in this LGTM

[jessebraham]

Changes look good, I only tried a few examples (using different chips) but they all worked.