use const steal

[xoviat]

The steal function is already unsafe, so there's no need to mark taken to make sure that someone else doesn't take it.

[wt]

Doesn't the steal function need to mark the peripherals taken so that take will work as intended? Otherwise, using take after steal will get the peripherals when we don't know that there is exclusive access via take.

[xoviat]

Otherwise, using take after steal will get the peripherals when we don't know that there is exclusive access via take.

Yes, and this is the correct result. steal is already unsafe, therefore, if you call it, you cannot guarantee that other code somewhere else does not have a duplicate instance of the peripherals. It is therefore your job to make sure that this isn't a problem.

However, suppose you are writing unsafe code and you need to call cortex-m methods, while still allowing safe code to take them. There is no way to do so without using a transmute, because calling steal will impact take.

[adamgreig]

This is an annoying legacy thing from the early ecosystem days when cortex-m, svd2rust, and cortex-m-rtic (then -rtfm) were a bit more co-designed, but at the moment RTIC (and perhaps other things we don't know about) uses steal() and requires that it marks the peripherals as taken, so that future calls to take() fail. In other words, it's required that steal() impacts take() and other libraries rely on that for correctness.

I think the only option (short of a breaking change in 0.8 or 1.0 where we change the API completely) is to add a new method that is const unsafe and is documented to not affect take(). I can't find it right now, but there was a lot of back and forth on this topic a couple of years ago, mostly trying to bikeshed a suitable name for it.

[xoviat]

@adamgreig What about 'const_steal'?