Explicitly document the size guarantees that Option makes.#75454

ltratt · 2020-08-12T14:12:28Z

Triggered by a discussion on wg-unsafe-code-guidelines about which layouts of Option<T> one can guarantee are optimised to a single pointer.

CC @RalfJung

Triggered by a discussion on wg-unsafe-code-guidelines about which layouts of `Option<T>` one can guarantee are optimised to a single pointer.

rust-highfive · 2020-08-12T14:12:32Z

r? @dtolnay

(rust_highfive has picked a reviewer for you, use r? to override)

RalfJung · 2020-08-12T14:31:33Z

library/core/src/option.rs

+//! Rust guarantees to optimise the following inner types such that an [`Option`] which contains
+//! them has the same size as a pointer:


Suggested change

//! Rust guarantees to optimise the following inner types such that an [`Option`] which contains

//! them has the same size as a pointer:

//! Rust guarantees to optimize the following types `T` such that an [`Option<T>`]

//! has the same size as a `T`:

Giving names to things helps. :) And we typically use US spelling, AFAIK.

Also maybe say that this means that one may transmute a T to an Option<T>?

Good point. Fixed in f5118a5.

RalfJung · 2020-08-12T14:31:53Z

library/core/src/option.rs

+//! * [`num::NonZero*`]
+//! * [`ptr::NonNull<T>`]
+//! * `#[repr(transparent)]` struct around one of the types in this list.
+//! * [`Box<T>`]


In terms of ordering I think it makes more sense to add this to the top or just below the references.

Agreed. Fixed in f5118a5.

RalfJung · 2020-08-12T14:34:11Z

Nominating for lang-team discussion.
@rust-lang/lang, I propose to guarnatee for a bunch of types that Option<T> has the same size as T, and that T may be transmuted to Option<T>. Most of this has actually already been decided in #60300, except for Box. So the FCP is basically about guaranteeing that Option<Box<T>> is subject to niche optimizations and has the same size as Box<T>.

RalfJung · 2020-08-12T14:38:36Z

library/core/src/option.rs

-//! * [`Box<T>`]
+//!
+//! For the above cases, it is guaranteed that one can use [`mem::transmute`]
+//! between `T` and `Option<T>` and vice versa.


No, not vice versa! If you transmute a None from Option<T> to T, you get UB.

Good point. Should there be a guarantee for non-None in such a case, or just rule out the reverse direction entirely?

Guarantee for non-None seems reasonable.

Fixed in 83f47aa.

library/core/src/option.rs

Co-authored-by: Ralf Jung <post@ralfj.de>

library/core/src/option.rs

Co-authored-by: Ralf Jung <post@ralfj.de>

RalfJung · 2020-08-13T05:59:30Z

@rust-lang/lang another point that is not entirely clear yet is function pointers -- the last FCP only covered extern "C" fn. I am sure we also want this guarantee for fn (aka extern "Rust" fn), but do we just want it for all ABIs or is there a chance some ABIs might not want a niche?

dtolnay · 2020-08-16T18:58:00Z

library/core/src/option.rs

+//! Rust guarantees to optimize the following types `<T>` such that an
+//! [`Option<T>`] has the same size as `T`:
+//!
+//! * [`Box<T>`]


This wording is pretty confusing to me, I think because it uses the same name T to mean two different things. For example this line is talking about what happens when T = Box<T> ... which is never true.

I struggled to find a good phrasing here; I agree the current phrasing isn't great.

Might something like:

//! Rust guarantees that the following types are optimised such that their size //! is identical whether or not they are surrounded by an `Option`:

be clearer?

Maybe just use Box<_> or Box<U>. Naming things is generally a good idea, so I'd not remove the T in the introduction.

Fixed in 68209c3.

dtolnay · 2020-08-16T18:59:33Z

Reassigning for reviewing accuracy after discussed by lang team.

r? @RalfJung

RalfJung · 2020-08-17T08:25:07Z

library/core/src/option.rs

-//! in Rust are stored as efficiently as any other pointer type.
+//! # Representation
+//!
+//! Rust guarantees to optimize the following types `<T>` such that an


Suggested change

//! Rust guarantees to optimize the following types `<T>` such that an

//! Rust guarantees to optimize the following types `T` such that an

I never saw <T> as notation for a type, not sure what that is supposed to indicate.

nikomatsakis · 2020-08-17T19:42:13Z

Discussed in rust-lang/lang meeting today:

We agree that this is a good idea and that we expect these representations to be guaranteed.
As far as ABI, we're not aware of any ABIs for which this property doesn't hold, and if a new ABI were added in the future, we could exempt it if necessary.

@rfcbot fcp merge

rfcbot · 2020-08-17T19:42:14Z

Team member @nikomatsakis has proposed to merge this. The next step is review by the rest of the tagged team members:

No concerns currently listed.

Once a majority of reviewers approve (and at most 2 approvals are outstanding), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

ltratt · 2020-09-02T10:41:05Z

That is certainly possible, though I am not sure if it is appropriate to make such a significant addition this late in the FCP process?

My tuppence worth: if we do add that guarantee (in some form) in, we might want to re-ask those who've signed the CFP off so far?

RalfJung · 2020-09-13T11:54:27Z

@cramertj @pnkfelix @withoutboats there's some FCP checkboxes waiting for you here. :)

rfcbot · 2020-09-15T23:41:46Z

🔔 This is now entering its final comment period, as per the review above. 🔔

rfcbot · 2020-09-25T23:45:17Z

The final comment period, with a disposition to merge, as per the review above, is now complete.

As the automated representative of the governance process, I would like to thank the author for their work and everyone else who contributed.

The RFC will be merged soon.

dtolnay · 2020-09-25T23:57:27Z

@bors r+

bors · 2020-09-25T23:57:28Z

📌 Commit 9bac577 has been approved by dtolnay

bors · 2020-09-26T09:26:07Z

⌛ Testing commit 9bac577 with merge 1671a2d6368f9890d7db0f6bf5f21226715e0c45...

@RalfJung

…es, r=dtolnay Explicitly document the size guarantees that Option makes. Triggered by a discussion on wg-unsafe-code-guidelines about which layouts of `Option<T>` one can guarantee are optimised to a single pointer. CC @RalfJung

RalfJung · 2020-09-26T09:32:20Z

Rolled up, yielding.
@bors retry rollup

Rollup of 12 pull requests Successful merges: - rust-lang#75454 (Explicitly document the size guarantees that Option makes.) - rust-lang#76631 (Add `x.py setup`) - rust-lang#77076 (Add missing code examples on slice iter types) - rust-lang#77093 (merge `need_type_info_err(_const)`) - rust-lang#77122 (Add `#![feature(const_fn_floating_point_arithmetic)]`) - rust-lang#77127 (Update mdBook) - rust-lang#77161 (Remove TrustedLen requirement from BuilderMethods::switch) - rust-lang#77166 (update Miri) - rust-lang#77181 (Add doc alias for pointer primitive) - rust-lang#77204 (Remove stray word from `ClosureKind::extends` docs) - rust-lang#77207 (Rename `whence` to `span`) - rust-lang#77211 (Remove unused #[allow(...)] statements from compiler/) Failed merges: - rust-lang#77170 (Remove `#[rustc_allow_const_fn_ptr]` and add `#![feature(const_fn_fn_ptr_basics)]`) r? `@ghost`

Explicitly document the size guarantees that Option makes.

73ada2d

Triggered by a discussion on wg-unsafe-code-guidelines about which layouts of `Option<T>` one can guarantee are optimised to a single pointer.

rust-highfive assigned dtolnay Aug 12, 2020

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Aug 12, 2020

RalfJung reviewed Aug 12, 2020

View reviewed changes

RalfJung added I-nominated T-lang Relevant to the language team labels Aug 12, 2020

Clarify and add guarantee about transmute.

f5118a5

RalfJung reviewed Aug 12, 2020

View reviewed changes

Be clear about the reverse transmute guarantees.

83f47aa

tesuji reviewed Aug 12, 2020

View reviewed changes

library/core/src/option.rs Outdated Show resolved Hide resolved

Be clearer about Some/None transmute.

f3d7196

RalfJung reviewed Aug 12, 2020

View reviewed changes

library/core/src/option.rs Outdated Show resolved Hide resolved

Change notation.

8cb8955

Co-authored-by: Ralf Jung <post@ralfj.de>

RalfJung reviewed Aug 12, 2020

View reviewed changes

library/core/src/option.rs Outdated Show resolved Hide resolved

Add Rust function pointers.

55802e3

Co-authored-by: Ralf Jung <post@ralfj.de>

dtolnay requested changes Aug 16, 2020

View reviewed changes

rust-highfive assigned RalfJung and unassigned dtolnay Aug 16, 2020

RalfJung reviewed Aug 17, 2020

View reviewed changes

ltratt added 2 commits August 17, 2020 09:31

Rename the types for clarity.

68209c3

Grammar tweak.

9bac577

RalfJung added S-waiting-on-team and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Aug 17, 2020

rfcbot added the final-comment-period In the final comment period and will be merged soon unless new substantive objections are raised. label Sep 15, 2020

rfcbot removed the proposed-final-comment-period Proposed to merge/close by relevant subteam, see T-<team> label. Will enter FCP once signed off. label Sep 15, 2020

dtolnay approved these changes Sep 25, 2020

View reviewed changes

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-team labels Sep 25, 2020

jonas-schievink mentioned this pull request Sep 26, 2020

Rollup of 13 pull requests #77212

Closed

RalfJung mentioned this pull request Sep 26, 2020

Rollup of 10 pull requests #77222

Closed

RalfJung mentioned this pull request Sep 26, 2020

Rollup of 12 pull requests #77224

Merged

bors merged commit 1e62382 into rust-lang:master Sep 26, 2020

rustbot added this to the 1.48.0 milestone Sep 26, 2020

spastorino removed the to-announce Announce this issue on triage meeting label Oct 1, 2020

RalfJung mentioned this pull request Mar 18, 2022

Document that Option<extern "abi" fn> discriminant elision applies for any ABI #95083

Merged

dtolnay self-assigned this Mar 24, 2024

		//! Rust guarantees to optimise the following inner types such that an [`Option`] which contains
		//! them has the same size as a pointer:

	//! Rust guarantees to optimize the following types `<T>` such that an
	//! Rust guarantees to optimize the following types `T` such that an

Uh oh!

Conversation

ltratt commented Aug 12, 2020

Uh oh!

rust-highfive commented Aug 12, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RalfJung commented Aug 12, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RalfJung commented Aug 13, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dtolnay commented Aug 16, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nikomatsakis commented Aug 17, 2020

Uh oh!

rfcbot commented Aug 17, 2020 • edited by cramertj Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ltratt commented Sep 2, 2020

Uh oh!

RalfJung commented Sep 13, 2020

Uh oh!

rfcbot commented Sep 15, 2020

Uh oh!

rfcbot commented Sep 25, 2020

Uh oh!

dtolnay commented Sep 25, 2020

Uh oh!

bors commented Sep 25, 2020

Uh oh!

bors commented Sep 26, 2020

Uh oh!

RalfJung commented Sep 26, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

11 participants

rfcbot commented Aug 17, 2020 •

edited by cramertj

Loading