sema.type_of_expr() cannot resolve non tailexpr never

[A4-Tacks]

rust-analyzer version: rust-analyzer 1.90.0-nightly (9748d87 2025-07-21)

rustc version: rustc 1.90.0-nightly (9748d87 2025-07-21)

editor or extension: VIM-9.1 coc-rust-analyzer

code snippet to reproduce:

This code cannot applicable Convert match to let-else

fn main() {
    let _x$0 = match 2 {
        it @ 0..5 => it,
        _ => {
            return; // not in tail_expr
        },
    };
}

But tailexpr can applicable

fn main() {
    let _x$0 = match 2 {
        it @ 0..5 => it,
        _ => {
            return // is tail_expr
        },
    };
}

Should type_of_expr handle the never type of non tailexpr in BlockExpr

[ChayimFriedman2]

So: unrelated to IDE, it seems our ! handling is incorrect. If I change blocks to use never instead of new_maybe_never_var(), a bunch of blocks change from () to !, and that change is correct on edition 2024, but not on edition 2021. So this has something to do with the never type fallback and its change, which I will not even pretend to understand, but I don't think we ever tried to model them correctly.

[ShoyuVanilla]

I once have tried - two years ago or so - to introduce rustc's never type fallback into rust-analyzer but gave up as it needed to port some graph search logics in rustc that quite hard to implement on rust-analyzer (I can't remember the details 😅). And I just lost interest after 2024 edition as we don't need it in 2024 edition.
But as I revisit my comment on the issue, it seems that maybe we could try it now with next-solver

[ChayimFriedman2]

Worst case, we can implement only the 2024 handling (which we don't currently). It's anyway planned to switch all editions to it after some time AFAIK.

[A4-Tacks]

Before everything becomes complicated, can we first implement a simple pattern:

• if a block does not have tail-expr and the last statement of a block is never,
  then the block is never