Collection of assumptions about MIR semantics

[RalfJung]

Let's use this issue to collect assumptions that rustc is making about MIR semantics. Usually those assumptions occur as an optimization.

• Generator optimization: Overlap locals that never have storage live at the same time rust#60187 assumes that after a StorageDead+StorageLive, the address of a local is allowed to differ -- so old pointers (created before the StorageDead) may not be used any more. Cc What about: StorageDead/StorageLive? #129
• Don't store locals that have been moved from in generators rust#61922 assumes that if the address of a local has not been taken, after moving out there is just no way to access that storage again. It can't be accessed through existing pointers because none have been created, and new pointers cannot be created because the move checker would prevent that.

[RalfJung]

I feel for issues such as this, a wiki is a better fit. Do we want to use the wiki of this project for that?

[Lokathor]

Well, "yes that sounds good, but maybe wait until there's more than 2 bullet points"

[JakobDegen]

Closing for backlog bonanza. Issue has had no updates since 2019. If someone wants to actually maintain a list of these things, feel free to open a new issue