The never type and diverging type variables

[lrh2000]

Previously, the never type will be replaced by a diverging type variable (generally) only if some coercion occurs. It can cause inconsistent behaviors like

return.foo();      //~ ERROR no method named `foo` found for type `!`
{ return }.foo();  //~ ERROR type annotations needed

let a = (return, );          // The type is `(!, )`.
let a = ({ return }, );      // The type is `(_, )`.
let a: (_, ) = (return, );   // The type is `(_, )`.

With the first commit, the never type will be replaced by a diverging type variable just at the end of the type check for every expression, even if no coercion occurs. Thus the problems above get solved and the consistency should be improved.

Then, another problem is we'll issue too many "type annotations needed". They are issued when types must be known at some point but the resolution failed, even if the type variables are just some diverging ones. A typical example is { return }.foo().

With the second commit, the information about diverging is recorded in the unification table, so that we can check whether performing the fallback affects other non-diverging type variables. If it doesn't, we will safely perform the fallback and we won't issue "type annotations needed" anymore.

As a result, "type annotations needed" will be issued for

let a = return;
{ if true { a } else { return } }.foo();

but not for

let a: ! = return;
{ if true { a } else { return } }.foo();

cc @nikomatsakis
cc @Mark-Simulacrum
Discussed at https://rust-lang.zulipchat.com/#narrow/stream/259160-t-lang.2Fproject-never-type/topic/Never.20type.20in.20tuple .

[rust-highfive]

r? @petrochenkov

(rust-highfive has picked a reviewer for you, use r? to override)

[petrochenkov]

r? @nikomatsakis

[bors]

☔ The latest upstream changes (presumably #84107) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #85290) made this pull request unmergeable. Please resolve the merge conflicts.

[lovishpuri]

Resolved conflicts in PR: #85558

[nikomatsakis]

Thanks for the PR @lrh2000 (and for the rebase, @lovishpuri!), sorry I haven't had time to go over it in detail yet.

[JohnCSimon]

ping from triage:
@lrh2000 Returning to you to address merge conflicts and switch back to S-waiting-on-review thanks,

[lrh2000]

Closed due to inactivity and no positive feedback (so I'm not sure whether it is on the right road).