Allow panicking in constants by oli-obk · Pull Request #2345 · rust-lang/rfcs (original) (raw)

@clarcharr it's helping the programmer for actively developed applications and libraries, but not everything has a maintainer.

There are situations even today when small backwards-incompatible coherence changes break existing unmaintained libraries (andybarron/app-dirs-rs#28 is my latest example).

This would be on a much larger scale, though. If we adopt const fns widely, which I assume is the plan, this would break any library or application which computes something incorrectly using compile-time data.

What if there's some bad regex::Regex::new hidden away in some part of a library that can't ever actually run? If regex makes new a const fn, and doesn't bump up from 1.0 in order to do that, that code will break. Even if the panic would never have been run at all, it could change to fail to compile.

@oli-obk Alright, that sounds reasonable to me.

Does this mean we make all compile-time panics lints which then forward the panic to runtime?