Position information for parsed regular expressions

[nikomatsakis]

It's unfortunate that a parsed regular expression Expr contains no offset information from the original string. It just means that if you have to later give an error message about some particular part you can't pinpoint with a lot of precision. (In my case, I want to report that certain features -- such as word boundaries -- are not supported.)

[BurntSushi]

That would definitely be very useful. I'd be fine adding this.

It would change the Expr type significantly, I think (which is fine). I don't have any experience building (or using) an AST with offset information inside of it. Do you have any advice on that front? What's convenient to use?

One approach I can think of is:

pub struct Spanned {
    expr: Expr,
    start: usize,
    end: usize,
}

And all of the recursive variants of Expr would then change to using Spanned instead.

Another approach would be to inline the offset information into every variant of Expr, but that seems more yucky.

[nikomatsakis]

The Spanned variant is what we do in the compiler. It's not great, but it's the usual thing. I usually use names like this:

pub struct Expr {
    start: usize,
    end: usize,
    kind: Box<ExprKind>
}

pub enum ExprKind {
    ...
}

but this has the disadvantage that matching becomes a bit more verbose (ExprKind::Foo instead of Expr::Foo).

[nikomatsakis]

This is one of those cases where having the ability to express common fields on enums would be really nice. But that'll be a while. :)

[BurntSushi]

Hah, right, your blog post is ringing a bell!

Your Expr/ExprKind variant seems nice. Usually I have a use regex_syntax::Expr::* in the function scope when matching anyway.