use PlaceRef abstraction more consistently

[RalfJung]

We have quite a bit of code that needs to work with the projections that make up a Place or PlaceRef. Much of it works with the projections array directly, which is rather verbose. We now have better APIs that avoid having to "break the abstraction" of Place/PlaceRef: we have Place::iter_projections and PlaceRef::last_projection.

It would be good to clean things up by using these higher-level APIs consistently. To do that, you'll need to find places where the lower-level API is used. One good way to find these places is to grep for Place::ty_from -- this helper function is only useful when a PlaceRef has been broken apart into its constituents, so it is a sign that some higher-level APIs can be used. In almost all cases, the code will either "do something with the last projection (if any)", in which case it should use PlaceRef::last_projection, or it will iterate all the projections, in which case it should use Place::iter_projections (usually iteration starts with the outermost projection, i.e., you should use place.iter_projections().rev()).

You can see #80624 for some examples for the kinds of changes that are needed to perform this cleanup.

[RalfJung]

I just realized one can write place_ref.ty(...) instead of PlaceRef::ty, so cleaning this up could be done as part of the same task or separately.

[oliviacrain]

@RalfJung I went ahead and did some of this work for the rustc_mir crate in #80865. If I'm not completely off-base with these changes, I can go ahead and take a deeper look at this.

[RalfJung]

@oliviacrain

@RalfJung I went ahead and did some of this work for the rustc_mir crate in #80865. If I'm not completely off-base with these changes, I can go ahead and take a deeper look at this.

That MR has been merged; do you still plan to push the same kind of change through the rest of the compiler?

[RalfJung]

Another thing I just noticed: visit_projection of the MIR visitor takes a Local and a projection list separately; that can become a single PlaceRef I think (but I didn't check all use sites).

[henryboisdequin]

Another thing I just noticed: visit_projection of the MIR visitor takes a Local and a projection list separately; that can become a single PlaceRef I think (but I didn't check all use sites).

Working on it in #82091 :)

[RalfJung]

@henryboisdequin wrote

on Zulip I and a couple of others discussed that it would be impossible to replace &mut Place with &mut PlaceRef in a mutable visitor

If you could leave a summary of the problem here, that'd be good. Note that I was specifically talking about visit_projection; I am not suggesting to change visit_place in any way. So I am confused about why you want to replace Place by PlaceRef anywhere. The mutable vistitor does not have visit_projection; it has process_projection instead which however doesn't even take a Local... I am not entirely sure what is going on there.