Fix deallocation span for locals freed by become tail calls

[LaneAsade]

This PR fixes #4863

The problem was due to the fact that before_memory_deallocation used current_user_relevant_span to record where an allocation was freed, which produced wrong spans for the locals deallocated by a become tail call.

The solution was to introduce a new field called tail_call_span on MiriMachine which records become expression's span in before_terminator if a TailCall terminator is detected and then use that span in before_memory_deallocation instead of the current one.

[rustbot]

Thank you for contributing to Miri! A reviewer will take a look at your PR, typically within a week or two.
Please remember to not force-push to the PR branch except when you need to rebase due to a conflict or when the reviewer asks you for it.

[RalfJung]

Thanks for looking into this, and sorry for the slow review!

However, I don't think this logic is correct. Once you set tail_call_span, it will never be unset, so it becomes the deallocation span of all allocations henceforth. That's clearly wrong. And this will be hard to fix since you really only want to affect the deallocations that happen while the TailCall terminator is executed, not anything that happens after.

I think this will require work in the core interpreter. Here on the Miri side, the best we can do are hard-to-get-right and, worse, hard-to-maintain hacks.

@rustbot author

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.