Tracking issue for fallback of diverging expressions to the never_type (!)

[Centril]

This issue is tracking the proposed change in fallback for "abrupt" expressions (part of the ! type RFC, tracking issue #35121).

Substeps

• create a fallback lint (create a lint for tracking "tainted" fallback around abrupt expressions and ! type #66173)
• ...

[bill-myers]

It seems to me this change is wrong as-is, since it causes a bare call to mem::zeroed() with unused result to panic, which is obviously absurd.

Based on https://hackmd.io/3VA7vpBFQZCZFjhVSh4O-A it seems to me that the problem is that the type of panic! and unimplemented! is an unbound inference variable, where in fact it should just be the "!" type.

With such a change, those expressions will have the "!" type even when fallback is to (), and that's the only argument in favor listed.

If fallback to "!" is still desired, then it DEFINITELY must not happen in places where it would cause code to panic that would not panic if the fallback was to "()" like simply calling mem::zeroed(): the thing that comes to mind is to add a "NotNever" or "Inhabited" trait bound to everything that uses unsafe code to generate arbitrary types like mem::zeroed, mem::uninitialized and mem::transmute, but not quite sure about this (especially because it seems it would be an incompatible change).

[Mark-Simulacrum]

Closing in favor of #35121 -- I don't think a separate tracking issue for components of never is particularly necessary. It's also somewhat likely that we're not going to end up changing fallback for never type-related things, regardless, due to the breaking changes and other complexity involved.