Span of previous token when parsing fails

[racradi]

Imagine you have a Rust-like language that uses semicolons as terminators:

let x = 0
let y = 0;

Forgetting a semicolon is an easy mistake to make. I've experimented a bit with Chumsky, but it seems that it will always pinpoint the next token (let in this case) as the cause of the error. That's not wrong, because after all that token was supposed to be a semicolon, but I'm wondering if there's a way to get the span of the previous token (i.e. 0 in this case) so I can provide a better error message? Ideally I'd like to point at the 0 and say I expected a semicolon after it.

Thanks a lot for all the work you've put in to Chumsky, by the way. Writing a parser with it felt like a breeze.

[zesterer]

That's a difficult one, because there's not an inherent reason for the parser to report a missing semicolon instead of, say, "expected operator instead of let". One thing you could do is to add the following on to the end of your expression parser:

expr.then_ignore(just("let").not())

not by itself does not consume any tokens, it just checks for the presence of a pattern. So, this will effectively just cause the parser to emit an error if a let ever directly follows an expression without changing the functionality of the parser in other ways.

You can combine it with a .map_err to customise the error, such as by explicitly stating 'expecting a semicolon instead of a let' or something of the sort.