Some fun trivia.

TIL that some raw identifiers are not allowed when thinking of some edge-case inputs to feed to macros. (My initial reaction was confusion, since I thought the whole point was to allow any keyword to be valid as a raw identifier). Naturally, I then explored more of the possibilities. In particular, when it comes to lifetime identifiers, as far as I'm aware, 'r#static, 'r#crate, 'r#self, 'r#Self, 'r#super, and 'r#_ are the only invalid raw lifetimes.

This does not match the list of RESERVED_RAW_LIFETIMEs in the Rust Reference, which does not include 'r#static.

The precise wording of the reference is that "Unlike a normal lifetime, a raw lifetime may be any strict or reserved keyword except the ones listed above for RAW_LIFETIME". ("The ones listed above" are the ones in RESERVED_RAW_LIFETIME.) Since 'static is a weak keyword which is restricted for lifetimes (not a strict or reserved keyword), 'static then cannot be used for raw lifetimes either.

Sort of cool.

I also now know that derive(yoke::Yokeable) does not handle raw identifiers very well. This makes me somewhat worried about whether derive macros for unsafe traits, likely using syn to parse input without meticulously using Ident::unraw everywhere, manage to handle everything soundly, especially since you can introduce a lifetime as 'r#a and use it as 'a... that is, a lifetime can refer to a lifetime parameter even if they are not equal (as Idents). Also, even quote doesn't support raw lifetimes (though it does support raw identifiers in most places).

I'm very glad that 'r#static is not a valid lifetime; it places a limit on how unsuspecting derive macros for unsafe traits can be tricked.

5 posts - 3 participants

Read full topic