Guarantee size and alignment of more integer primitives

[joshlf]

Follows up on #2200 (comment)

As an optional extension, we could consider guaranteeing equality between the alignments of:

• u32 and char
• u32 and f32
• u64 and f64

Those equalities strike me as less important, and I'm not sure if they're even valid (ie, whether they hold on all platforms).

[traviscross]

Thanks. I pushed a clarification and separated out the rules.

Clearly we want to guarantee this for fixed-width integers of the same size.

@rfcbot fcp merge lang

[rust-rfcbot]

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

• @joshtriplett
• @nikomatsakis
• @scottmcm
• @tmandry
• @traviscross

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!

cc @rust-lang/lang-advisors: FCP proposed for lang, please feel free to register concerns.
See this document for info about what commands tagged team members can give me.

[scottmcm]

Good to see this being specifically about for the stated N. I hope we don't need usize ever aligned differently from the uN of the same size, but that's less "well obviously", so sticking with just "the i vs u doesn't impact alignment" as the guarantee sounds good.

@rfcbot reviewed

[rust-rfcbot]

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

[rustbot]

This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[joshlf]

Now that the usize/isize language has merged (#2200), the diff in this PR is slightly changed. PTAL and confirm that the updated language and organization still looks good.

[programmerjake]

I created a minimal #![no_core] rust program that compiles when alignments match: it checks u32=char, u32=f32, and u64=f64

Source code

#![feature(intrinsics)]#![feature(rustc_attrs)]#![allow(internal_features)]#![feature(lang_items)]#![feature(no_core)]#![no_core]#[lang = "pointee_sized"]pubtraitPointeeSized{}#[lang = "meta_sized"]pubtraitMetaSized:PointeeSized{}#[lang = "sized"]pubtraitSized{}#[lang = "clone"]pubtraitClone:Sized{fnclone(&self) -> Self;}#[lang = "copy"]pubtraitCopy:Clone{}#[lang = "legacy_receiver"]pubtraitLegacyReceiver{}impl<T: ?Sized>LegacyReceiverfor&'_T{}#[rustc_nounwind]#[rustc_intrinsic]pubconstfnalign_of<T>() -> usize;const _:[();align_of::<u32>()] = [();align_of::<char>()];const _:[();align_of::<u32>()] = [();align_of::<f32>()];const _:[();align_of::<u64>()] = [();align_of::<f64>()];

I checked with rustc 1.94.0 with:

fortargetin`rustc --print=target-list`;do RUSTC_BOOTSTRAP=1 cargo check --target="$target"||echo"failed on $target";done

that gives failures on m68k-unknown-linux-gnu, m68k-unknown-none-elf, and powerpc64-ibm-aix

Compiler errors

 Checking align-test v0.0.0 (/tmp/align-test)
error[E0308]: mismatched types
--> src/lib.rs:35:36
|
35 | const _: [(); align_of::<u32>()] = [(); align_of::<f32>()];
| ----------------- ^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a size of 2, found one with a size of 4
| |
| help: consider specifying the actual array length: `4`
error[E0308]: mismatched types
--> src/lib.rs:36:36
|
36 | const _: [(); align_of::<u64>()] = [(); align_of::<f64>()];
| ----------------- ^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a size of 4, found one with a size of 8
| |
| help: consider specifying the actual array length: `8`
For more information about this error, try `rustc --explain E0308`.
error: could not compile `align-test` (lib) due to 2 previous errors
failed on m68k-unknown-linux-gnu
Checking align-test v0.0.0 (/tmp/align-test)
error[E0308]: mismatched types
--> src/lib.rs:35:36
|
35 | const _: [(); align_of::<u32>()] = [(); align_of::<f32>()];
| ----------------- ^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a size of 2, found one with a size of 4
| |
| help: consider specifying the actual array length: `4`
error[E0308]: mismatched types
--> src/lib.rs:36:36
|
36 | const _: [(); align_of::<u64>()] = [(); align_of::<f64>()];
| ----------------- ^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a size of 4, found one with a size of 8
| |
| help: consider specifying the actual array length: `8`
For more information about this error, try `rustc --explain E0308`.
error: could not compile `align-test` (lib) due to 2 previous errors
failed on m68k-unknown-none-elf
Checking align-test v0.0.0 (/tmp/align-test)
error[E0308]: mismatched types
--> src/lib.rs:36:36
|
36 | const _: [(); align_of::<u64>()] = [(); align_of::<f64>()];
| ----------------- ^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a size of 8, found one with a size of 4
| |
| help: consider specifying the actual array length: `4`
For more information about this error, try `rustc --explain E0308`.
error: could not compile `align-test` (lib) due to 1 previous error
failed on powerpc64-ibm-aix

[programmerjake]

so, I think we can guarantee that u32 and char have the same alignment, but not integers and the same-sized floating-point types.

[RalfJung]

We already guarantee that u32 and char are ABI-compatible, that requires their alignment to be the same. We test for that as well.

And yeah float types are definitely different from integer types in various ways, we should not guarantee anything there.

[RalfJung]

FWIW we also guarantee that usize is ABI-compatible with the equal-sized unsigned int, and similar for isize (and we test for that). So unless we want to take that back, the ship has sailed for targets where that wouldn't be true.

[joshlf]

Okay, updated to guarantee size/alignment equality for:

• usize, isize, uN, and iN
• uN and iN
• char and u32

[rust-rfcbot]

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.

[traviscross]

We're talking about this one on the lang-docs call. @joshlf, you had added some commits expanding the guarantee based on feedback after the lang team had reviewed and approved this. If you don't mind, let's handle those in a separate PR so that we can cleanly ensure lang-team signoff. I've rolled this back to the commit the lang team reviewed and rebased (and backed up your other work; let me know if you need that). Given the completed FCP, let's merge that.