GenericAp can be converted to MemoryAp without checks.

Not all APs are memory APs, so I don't understand why no check is necessary here. Shouldn't the type be checked?

I only checked ArmCommunicationInterface<Initialized>::memory_interface and it had a runtime check. I'm not sure about other places, but if the caller is free to convert one AP type into another, then we need these checks anyway and we don't lose much on types, except for maybe a bit of API clarity.

Now we should check the type of the AP in every function which takes an AP address, which seems more cumbersome than using a specific type to indicate that we know it is a memory AP.

It might reflect the case that we didn't do that check before, so it's probably not worse now, but I would have preferred to keep different types to represent the reality that not all AP addresses point to memory APs.

I would agree if constructing a MemoryAp type included checking whether the AP is actually a memory AP. If we don't do this check at construction time, then it doesn't matter what type our API takes, there's opportunity for misuse.

Anyway, feel free to revert this PR if you feel that we should try another approach.

Actually please do not revert. The types and checks are added in the following pr.

EDIT: Give me a little moment, I’ll rebase #2683 and add links to the relevant parts for the types and the checks.

EDIT2: And for my humble opinion, as much as I’d like to avoid doing those checks several times at run time, this only happen when a new memory_interface is obtained, so it is not as bad as it looks.
Also, I’d like to have them cached somewhere so as not to have to do that check again but the complexity cost seemed to outweigh the simple check at instanciation time.

EDIT3:
@bugadani with regards to the grouping of imports, this is to (IMHO) simplify their readability.
There modules where things are imported from the same module by relative and absolute path. Considering that some module name appear several times in the tree (like memory) it becomes harder to figure where are things comming from.
Compounded whith re-exports (pub use) I pulled my hair a few times on this.
So definite +1 on defining 1 style and sticking with it. I didn’t know rustfmt could do that. If it does that’d be fantastic to enable (and limiting re-export but that’s yet another story).

EDIT4:
The link to the check done on MemoryAp instanciation: https://github.com/ithinuel/probe-rs/blob/add-adiv6-support-step-2/probe-rs/src/architecture/arm/ap/memory_ap/mod.rs#L179-L193