Refactor: {Lvalue,Rvalue,Operand}::ty only need the locals' types, not the full &Mir

[RalfJung]

I am writing code that needs to call these ty methods while mutating MIR -- which is impossible with the current API.

Even with the refactoring the situation is not great: I am cloning the local_decls and then passing the clone to the ty methods. I have to clone because Mir::basic_blocks_mut borrows the entire Mir including the local_decls. But even that is better than not being able to get these types at all...

Cc @nikomatsakis

[rust-highfive]

r? @eddyb

(rust_highfive has picked a reviewer for you, use r? to override)

[RalfJung]

A way to avoid the cloning may be to wrap basic_blocks and cache into their own type; then Mir::basic_blocks_mut would only have to borrow these two bits.

The refactoring in this PR however would still be needed.

EDIT: I saw the tidy errors, but I'm going to wait for feedback if this approach will be accepted at all before fixing them.

[arielb1]

You can allow passing the MIR to .ty using AsRef<LocalDecls> to make the code cleaner:

impl<'tcx> AsRef<LocalDecls<'tcx>> for LocalDecls<'tcx> { /* .. */ }
impl<'tcx> AsRef<LocalDecls<'tcx>> for Mir<'tcx> { /* .. */ }

    fn ty<'a, 'gcx, D: AsRef<LocalDecls>>(&self, local_decls: D, tcx: TyCtxt<'a, 'gcx, 'tcx>) -> LvalueTy<'tcx> {
        self.ty_(local_decls.as_ref(), tcx)
    }

    fn ty_<'a, 'gcx>(&self, local_decls: &LocalDecls<'tcx>, tcx: TyCtxt<'a, 'gcx, 'tcx>) -> LvalueTy<'tcx> {
        // ..
    }

[arielb1]

A way to avoid the cloning may be to wrap basic_blocks and cache into their own type; then Mir::basic_blocks_mut would only have to borrow these two bits.

The cloning does not look like such big of a deal. We need to figure out a "builder" interface for MIR, but that's for the future.

[RalfJung]

You can allow passing the MIR to .ty using AsRef to make the code cleaner:

For this I have to wrap LocalDelcs into a newtype, or else coherence checks kick in.

Would that be all right, or too hevay-weight?

[nikomatsakis]

@RalfJung in which crate are you adding the impls?

[RalfJung]

@nikomatsakis in librustc.

[RalfJung]

I made a new trait for this overloading.

[RalfJung]

Should I be annotating these fn local_decls with #[inline]? Seems silly for them to ever not be inlined.

[RalfJung]

Oh, and another thought: I could monomorphize the big functions taking HasLocalDecls (change them to something private that takes &LocalDelcs), and only make small shims polymorphic. That reduces code size. Is it worth it though?

[nikomatsakis]

@bors r+

[bors]

📌 Commit 0bbc315 has been approved by nikomatsakis

[bors]

⌛ Testing commit 0bbc315 with merge 23ecebd...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: nikomatsakis
Pushing 23ecebd to master...