further split up const_fn feature flag

[RalfJung]

This continues the work on splitting up const_fn into separate feature flags:

• const_fn_trait_bound for const fn with trait bounds
• const_fn_unsize for unsizing coercions in const fn (looks like only dyn unsizing is still guarded here)

I don't know if there are even any things left that const_fn guards... at least libcore and liballoc do not need it any more.

@oli-obk are you currently able to do reviews?

[rust-highfive]

r? @estebank

(rust-highfive has picked a reviewer for you, use r? to override)

[oli-obk]

Please move the const_fn feature from the active list to the removed list (different file)

[RalfJung]

@oli-obk That feature gate is still used in some places, and I don't always understand what that is for -- in particular in compiler/rustc_mir/src/const_eval/fn_queries.rs:81:25 and compiler/rustc_mir/src/transform/check_consts/validation.rs:445:59. So, I don't think I know how to fully remove const_fn.

[oli-obk]

replace the first one with true, all unstable functions have feature gates and all stable ones can be used without feature gates

the second one can likely just not check any feature gate, as it currently is testing for the feature gate being inactive, so there's no problem just removing the check

[RalfJung]

That works... now I have to remove this flag from >50 tests.
I know why I didn't originally want to remove this feature flag in this PR. ;)

[RalfJung]

Actually I take this back, something seems wrong: after doing what you proposed, the proc_macro crate builds without any const_fn_* feature even though it contains

impl<T: LambdaL> ScopedCell<T> {
    #[rustc_allow_const_fn_unstable(const_fn)]
    pub const fn new(value: <T as ApplyL<'static>>::Out) -> Self {
        ScopedCell(Cell::new(value))
    }
}

[RalfJung]

So, I got it all to build again, but I don't think this is right. I added a new feature gate for one last use of const_fn (const_fn_in_trait), but now that error shows up in loads of new places and also there's always two errors -- the feature gate error and a hard error.

Moreover, some errors in the min_const_fn file disappeared. I have no idea why.

[RalfJung]

Oh also this is now pointing to a commit of stdarch that is in my fork (to include rust-lang/stdarch#1140), I don't know if this will even build on CI...

[RalfJung]

This part of the diff seems wrong to me (this is also what happened in proc_macro).

[oli-obk]

I think the is may need its own test. The message is missing because other errors are reported first?

[RalfJung]

That makes no sense to me, the order of errors shouldn't change just because I got rid of a feature flag...

[oli-obk]

Actually I take this back, something seems wrong: after doing what you proposed, the proc_macro crate builds without any const_fn_* feature even though it contains

impl<T: LambdaL> ScopedCell<T> {
    #[rustc_allow_const_fn_unstable(const_fn)]
    pub const fn new(value: <T as ApplyL<'static>>::Out) -> Self {
        ScopedCell(Cell::new(value))
    }
}

It does require rustc_attrs though, so that's ok

[RalfJung]

@Mark-Simulacrum it seems to still use the old ctfe-stress-4.rs. I found a commit sha in src/ci/pgo.sh and updated that, let's hope that's right.
@bors r=oli-obk

[bors]

📌 Commit 49054c3 has been approved by oli-obk

[bors]

⌛ Testing commit 49054c3 with merge b56b175...

[bors]

☀️ Test successful - checks-actions
Approved by: oli-obk
Pushing b56b175 to master...

[rust-highfive]

📣 Toolstate changed by #84310!

Tested on commit b56b175.
Direct link to PR: #84310

💔 rls on windows: test-pass → build-fail (cc @Xanewok).
💔 rls on linux: test-pass → build-fail (cc @Xanewok).
💔 rustfmt on windows: test-pass → build-fail (cc @calebcartwright @topecongiro).
💔 rustfmt on linux: test-pass → build-fail (cc @calebcartwright @topecongiro).