Can't shut down the EtherCAT bus after a state transition failure

[fpdotmonkey]

Hi! Thanks for opening an issue. Please provide some specific details to make it easier to debug any
problems.

Environment

• Version of ethercrab in use or git hash: 0daef10
• Operating system kind and version: 6.6.62+rpt-rpi-2712 Improve slave group iteration API, add example to docs #1 SMP PREEMPT Debian 1:6.6.62-1+rpt1 (2024-11-25) aarch64 GNU/Linux
• EtherCAT devices in use: irrelevant

Description of the problem/feature request/other

In the program I'm writing, I expect occasional failures to enter OP, generally because of unconfigured SDOs, which is fine for my application. In that event, I'd like to gracefully shutdown the EtherCAT bus. However, that's not possible. Observe,

let group = match group.into_op(&maindevice).await {
    Ok(group) => group,
    Err(err) => {
        group.into_init(&maindevice).await?;  // ERROR: group was moved above and can't be used anymore
        time_to_exit();
    }
}

The state transition failure, along with the borrow checker, make it so group cannot be interacted with after a state transition failure. So the question is how to solve this.

I have a couple thoughts, and I wonder what's the most right?

• Create a new group with MainDevice; I think it should work as-is, but it feels hacky
• Add a reference to a valid SubDeviceGroup in the Err type; this allows the cleanup, but makes handling Err possibly more complicated
• Have MainDevice shut down the bus on Drop

[jamwaffles]

Hey, thanks for writing this issue out. Another user of EtherCrab had some other desired improvements around better state transitions and I started a small experiment in #240 to address them, so it's clearly something that needs work.

To address your suggestions specifically, I would think the second option with something like Err((err, group)) would be the best approach, but that feels like a local maximum, if that makes sense. I'll have a think about what a good approach might be to relax the ownership a bit so failures, retries, etc can be done more ergonomically.