ACP: Negation functions for NonZeroI*

Proposal

Problem statement

Integer primitives have functions such as overflowing_neg() for numeric negation with controlled overflow. These functions are currently missing from the core::num::NonZeroI{8,16,32,64} types.

Motivation, use-cases

Non-zero integers are frequently used to represent error codes from low-level OS APIs, because these APIs use 0 to indicate non-error. Some APIs use signed error codes as a form of namespacing, such as internal/public (FreeBSD) or kernel/userspace (Linux).

For example, the FUSE protocol uses userspace error codes (positive non-zero) but negates them when sending an error frame to the kernel.

// old
let (neg, _) = err.get().overflowing_neg();
let wire_neg = unsafe { NonZeroI32::new_unchecked(neg) };

// new
let (wire_neg, _) = err.overflowing_neg();

Solution sketches

Since the primitive integer types already have *_neg() functions, and negation can't return zero for non-zero inputs, writing wrappers for the non-zero integers is straightforward.

pub const fn is_negative(self) -> bool {
    self.get().is_negative()
}

pub const fn checked_neg(self) -> Option<$Ty> {
    if let Some(result) = self.get().checked_neg() {
        return Some(unsafe { $Ty::new_unchecked(result) });
    }
    None
}

pub const fn overflowing_neg(self) -> ($Ty, bool) {
    let (result, overflow) = self.get().overflowing_neg();
    ((unsafe { $Ty::new_unchecked(result) }), overflow)
}

pub const fn saturating_neg(self) -> $Ty {
    if let Some(result) = self.checked_neg() {
        return result;
    }
    unsafe { $Ty::new_unchecked(<$Int>::MAX) }
}

pub const fn wrapping_neg(self) -> $Ty {
    let result = self.get().wrapping_neg();
    unsafe { $Ty::new_unchecked(result) }
}

Links and related work

rust-lang/rust#89065
rust-lang/rust#84186

What happens now?

This issue is part of the libs-api team API change proposal process. Once this issue is filed the libs-api team will review open proposals in its weekly meeting. You should receive feedback within a week or two.

Proposed implementation: jmillikin/upstream__rust@a807a2d

Docs screenshots:

How about trait Neg for the operator?

How about trait Neg for the operator?

It's easy to implement, but my understanding is that trait impls for two stable types can't be marked unstable due to rust-lang/rust#55436.

#[unstable(feature = "nonzero_negation_ops", issue = "none")]
#[rustc_const_unstable(feature = "const_ops", issue = "90080")]
impl const Neg for $Ty {
    type Output = Self;

    #[inline]
    fn neg(self) -> Self {
        unsafe { $Ty::new_unchecked(Neg::neg(self.get())) }
    }    
}

error: an `#[unstable]` annotation here has no effect
   --> library/core/src/num/nonzero.rs:877:13
[...]
    = note: `#[deny(ineffective_unstable_trait_impl)]` on by default
    = note: see issue #55436 <https://github.com/rust-lang/rust/issues/55436> for more information

That's not a deal-breaker if the trait implementation is sensible/desirable.

I agree that Neg would be useful, have no strong opinion on the exact details of the stabilization process, and would be happy to do whatever the Rust libs team recommends in this case.

Given that the signed nonzero types already have things like checked_abs and overflowing_abs and saturating_mul and such, adding things like checked_neg and saturating_neg seem well covered by existing precedent, so if you want to send a PR for the methods as unstable, you can r? @scottmcm on that and I'll approve it. (They probably just weren't done originally because they don't make sense on the unsigned versions.)

Trait impls can't go in unstable, though, so that at least should be a different PR, and I'll leave it for a real libs-api member to comment on it.

OK, I've filed PR rust-lang/rust#102341 to implement Neg for NonZeroI*.

To avoid merge conflicts, I'll send a second PR for the negation functions once that first one lands.

Traits take at least 10 days for FCP, so that PR will take much longer to land than one for unstable inherent methods.

Traits take at least 10 days for FCP, so that PR will take much longer to land than one for unstable inherent methods.

Interesting. OK, filed rust-lang/rust#102342 to add the new methods. Once one of them merges I'll rebase the second one.

Rollup merge of rust-lang#102342 - jmillikin:nonzero-negation, r=scot…

Rollup merge of #102342 - jmillikin:nonzero-negation, r=scottmcm

Rollup merge of rust-lang#102341 - jmillikin:nonzero-impl-neg, r=dtolnay

Rollup merge of rust-lang#111186 - jmillikin:nonzero-is-positive, r=d…

Auto merge of rust-lang#111044 - jmillikin:nonzero-negation, r=dtolnay

Auto merge of #111044 - jmillikin:nonzero-negation, r=dtolnay

Rollup merge of #111186 - jmillikin:nonzero-is-positive, r=dtolnay

Auto merge of #111044 - jmillikin:nonzero-negation, r=dtolnay

Navigation Menu

Repository navigation

ACP: Negation functions for `NonZeroI*` #105

Proposal

Problem statement

Motivation, use-cases

Solution sketches

Links and related work

What happens now?

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ACP: Negation functions for NonZeroI* #105

Description

Proposal

Problem statement

Motivation, use-cases

Solution sketches

Links and related work

What happens now?

Activity

jmillikin commented on Sep 18, 2022

cuviper commented on Sep 18, 2022

jmillikin commented on Sep 18, 2022

thomcc commented on Sep 18, 2022

jmillikin commented on Sep 18, 2022

scottmcm commented on Sep 26, 2022

jmillikin commented on Sep 27, 2022

scottmcm commented on Sep 27, 2022

jmillikin commented on Sep 27, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

ACP: Negation functions for `NonZeroI*` #105