explain why Rvalue::Len still exists

[RalfJung]

I just spent a bit of time trying to remove this until I realized why that's non-trivial. Let's document that for the next person. :)

[rustbot]

r? @cjgillot

rustbot has assigned @cjgillot.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

This PR changes MIR

cc @oli-obk, @RalfJung, @JakobDegen, @davidtwco, @celinval, @vakaras

[compiler-errors]

@bors r+ rollup

[bors]

📌 Commit 98f74b4 has been approved by compiler-errors

It is now in the queue for this repository.

[scottmcm]

I have a branch somewhere that removes it -- it's not that hard to change the emitting code to just do PtrMetadata -- but the bigger problem I hit was things like the dataflow framework. #126541 made some progress there by making GVN handle it, but as you found out, it doesn't seem trivial to nuke it.

(MIR building has enough types that when it needs a length for indexing it could emit the size directly for arrays or the right PtrMetadata for slices without much trouble.)

[RalfJung]

(MIR building has enough types that when it needs a length for indexing it could emit the size directly for arrays or the right PtrMetadata for slices without much trouble.)

That's what I tired, but it would have to add a temporary to hold a pointer to the place, to pass it to PtrMetadata -- and that seemed like a non-trivial-enough change that I stopped there.

[scottmcm]

The other complication here, dunno whether it's worth mentioning, is that PtrMetadata needs a wide pointer, so if the place is an array you also need to unsize it to slice.

BTW, are there [ST]B issues with adding a short-lived raw-ref to use PtrMetadata if the place is something like (*_2).1? Or is it just that it's a bit annoying to make the temporary?

[RalfJung]

The other complication here, dunno whether it's worth mentioning, is that PtrMetadata needs a wide pointer, so if the place is an array you also need to unsize it to slice.

Not really, as you said above -- if the place is an array we can just emit a constant instead.

BTW, are there [ST]B issues with adding a short-lived raw-ref to use PtrMetadata if the place is something like (*_2).1? Or is it just that it's a bit annoying to make the temporary?

There also shouldn't be any -- creating a raw pointer you never use should be effectively a NOP.