MIR sanity check improvements

[RalfJung]

This issue collect various proposals for improvements to the MIR sanity check so that they do not get lost:

• Function calls and DropAndReplace could test that given and expected type match (MIR sanity check: validate types on assignment #72796 (comment))
• SetDiscriminant could have some invariants checked (MIR sanity check: validate types on assignment #72796 (comment))

[RalfJung]

@oli-obk re: SetDiscriminant, allowing that only for enums seems curiously asymmetric with GetDiscriminant which works for all ADTs.

[oli-obk]

Right, so let's check that the local is an Adt and that the discriminant id is actually a valid id for that type.

[RalfJung]

Actually, does GetDiscriminant only work for ADTs or for all types? I forgot.

[oli-obk]

It works for all types (returning 0 for things that are neither Adt nor generator). That doesn't mean MIR building ever creates it and I don't think I've seen it before.

[ecstatic-morse]

Rvalue::Discriminant is also used for TyKind::Generators. edit: Whoops, this was already mentioned.

[RalfJung]

It works for all types (returning 0 for things that are neither Adt nor generator). That doesn't mean MIR building ever creates it and I don't think I've seen it before.

MIR building creates it for every match I think... or was it just every match of an ADT?