Warn for unreachable code after panic

[giacomocavalieri]

This PR closes #3023

I tried copying Rust's approach where it tries to understand if the previous expression will always panic and raise a warning accordingly. I'm not sure about the error message, let me know if you think it can be improved somehow!

[lpil]

Lovely error messages! Looks so good!

One thing I don't like so much about this implementation is that it traverses the sub-tree for every single AST node, which is very wasteful. I think we could instead store whether the previously analysed expression is a panic on the ExprTyper in place of the bool that's used currently to prevent emitting multiple warnings. Then it's O(n) rather than O(n log n)

Let's remove the PanicKind too, saying the previous one panics is good enough I think.

[giacomocavalieri]

I think we could instead store whether the previously analysed expression is a panic on the ExprTyper in place of the bool that's used currently to prevent emitting multiple warnings.

I'm not entirely sure I see how that could be implemented, so I would have to change every infer_x function to update a field in the typer? For cases that wouldn't work right away because I have to check that all branches panic; I would have to type a branch, save the value of the field and go on with the others accumulating that result somewhere. Is this how it should work?

[lpil]

I think just the infer function and the ones for panic, anonymous functions, and case should be sufficient. In case I'd use a mutable variable as an accumulator to see if all clauses panic.

[giacomocavalieri]

I think just the infer function and the ones for panic, anonymous functions, and case should be sufficient. In case I'd use a mutable variable as an accumulator to see if all clauses panic.

Thanks for the help! I should have fixed it now

[lpil]

Super nice!! I've left two small notes

[lpil]

Delightful!! Thank you