some const_int_foo features were entirely removed

[Lokathor]

As part of the const project moving things into stable, some of the associated sub-features were recently removed entirely rather than being left in with a warning about them being old, which causes previously working code to now error out (example build log). Of course it's nightly and all, but it'd be better if they could be warnings instead of errors.

[varkor]

Const feature flags are handled a little differently to regular feature flags. Ideally, we'd handle these in the same way as regular library feature flags. For example, when const_int_wrapping was removed, this flag:

rust/src/libcore/num/mod.rs

Line 2887 in 0e1402d

#[rustc_const_unstable(feature = "const_int_wrapping")]

Should have been replaced by some sort of rustc_const_stable flag, e.g. #[rustc_const_stable(feature = "const_int_wrapping", since = "1.33.0")].

The problem is that rustc_const_stable doesn't currently exist. To fix this, rustc_const_stable would need to be added as an attribute and handled similarly to stable feature flags.

One could by searching for "rustc_const_unstable" in the compiler and working out which locations need to be modified to handle rustc_const_stable.lib_features.rs should contain most of the relevant code that will need to be modified to handle rustc_const_stable. I can provide more detailed mentoring instructions if someone wants to take this on.

[Centril]

I would personally prefer to simply warn on unknown features rather than introduce #[rustc_const_stable(...)] since that will need to be used a lot and thus will substantially add technical debt to the standard library.

While retaining #[stable(..)] is useful for the purposes of rustdoc documentation the same cannot be said for #[rustc_const_stable(...)] since it isn't shown... that is, unless we actually show it in rustdoc (which might be too much information to fit and show visually).

cc @QuietMisdreavus

[QuietMisdreavus]

It's possible to add a "const stability" banner like the regular stability banner and make rustdoc look it up based on that attribute. @rust-lang/rustdoc

[doctorn]

Hello! I'd be interested in working on this if nobody else already is

[Centril]

@doctorn Don't think anyone is working on this... The main issue seems to be that we haven't settled how we want to approach this since me and @varkor have a disagreement... and the rustdoc team hasn't weighed in except for @QuietMisdreavus ;)

[sjakobi]

The main issue seems to be that we haven't settled how we want to approach this since me and @varkor have a disagreement... and the rustdoc team hasn't weighed in except for @QuietMisdreavus ;)

May I suggest that the E-mentor and E-medium labels be removed until the approach has been settled?

[Mark-Simulacrum]

I suspect it likely doesn't matter too much at this point -- this is a pretty old issue, and I think we should const stability these days. So I'm going to go ahead and just close.