macros: add support for `exhaustive` attribute by dybucc · Pull Request #5390 · rust-lang/libc

dybucc · 2026-08-08T07:59:21Z

Description

Attempts to take the first steps in closing #4080. This should
allow us to add a private field for each of the records declared within
s_with_default or s_no_extra_traits_with_default.

This should in turn enforce those records to be initialized field-by-field in
downstream crates, while avoiding the lint that currently pops up when using a
struct marked non_exhaustive in FFI contexts.

Checklist

Relevant tests in libc-test/semver have been updated
Commit messages permalink to headers for added or changed API
Placeholder or unstable values like *LAST or *MAX have the standard
doc comment
Tested locally (cargo test -p libc-test --target mytarget); especially
relevant for platforms that may not be checked in CI

@rustbot label +stable-nominated

tgross35 · 2026-08-25T08:35:45Z

Could you base this on top of #5372? It looks like there may be some overlap. Feel free to leave a review there as well, if you have any suggestions.

dybucc · 2026-08-29T10:25:32Z

@rustbot author

rustbot · 2026-08-29T10:25:36Z

Reminder, once the PR becomes ready for a review, use @rustbot ready.

dybucc · 2026-08-29T16:03:18Z

I think it's done. I've created a new patchset based off of #5372 ¹,
but I haven't yet dropped the commits in this patch. Once that PR
merges, I'll drop the ones in this patch and cherry pick the ones from
the above patch.

@rustbot ready

https://github.com/dybucc/libc/commits/feature/default-macro-cfg/ ↩

tgross35

@rustbot author

Needs the updates after #4080. Would you mind also renaming s_with_default! to s2! and similar for s_no_extra_traits_with_default? Since with this PR they'll be doing more than just adding Default.

Please also include this in their doc comments.

View changes since this review

dybucc · 2026-09-01T11:53:30Z

@rustbot ready

tgross35

Looks pretty great to me! Nothing major here

View changes since this review

tgross35 · 2026-09-07T04:27:32Z

-/// Emit a struct with the given derive attributes plus a generated `Default` impl.
+/// Emit a struct with the given derive attributes plus a generated `Default`
+/// impl. Ensure that the record has an additional private field added to
+/// replicate `#[non_exhaustive]`, unless it is annotated with `#[exhaustive]`.
 ///
-/// Fields default to `Default::default()`. A field whose default can't be derived must carry
-/// `#[custom_default(EXPR)]` as its *first* attribute, and `EXPR` is used instead.
+/// Fields default to `Default::default()`. A field whose default can't be
+/// derived must carry `#[custom_default(EXPR)]` as its *first* attribute, and
+/// `EXPR` is used instead.
 ///
-/// This works by scanning each field for `#[custom_default]` attributes. If one exists, the
-/// attribute's contents are added to `processed_field_defaults` and will be used in the expansion
-/// for `Default`. If it does not exist, `Default::default()` is used instead. In either case, the
-/// field is added to `processed_fields` with `#[custom_default]` stripped if necessary, and
+/// This works by scanning each field for `#[custom_default]` attributes. If one
+/// exists, the attribute's contents are added to `processed_field_defaults` and
+/// will be used in the expansion for `Default`. If it does not exist,
+/// `Default::default()` is used instead. In either case, the field is added to
+/// `processed_fields` with `#[custom_default]` stripped if necessary, and
 /// `struct_with_default` is invoked again with the remaining fields.
 ///
-/// Attributes are split into `cfg_attrs` and `other_attrs` before the fields are scanned. Both
-/// go on the struct, but only the `cfg`s are repeated on the `Default` impl. A `cfg` decides
-/// whether the type exists at all, so without it a configured-out struct leaves an impl behind
-/// referring to a type that isn't there.
+/// Attributes are split into `cfg_attrs` and `other_attrs` before the fields
+/// are scanned. Both go on the struct, but only the `cfg`s are repeated on the
+/// `Default` impl. A `cfg` decides whether the type exists at all, so without
+/// it a configured-out struct leaves an impl behind referring to a type that
+/// isn't there.
+///
+/// Out of `other_attrs`, we scan for `#[exhaustive]`. If found, we remove it
+/// but take into account that the record should be expanded _without_ an
+/// additional private field. The scan for both `cfg` attributes and the
+/// (non-existent) `exhaustive` attribute is done in one linear pass.


Looks like some autoformatting got you, seems like this was rewrapped from 100 to 80.

Nay, this one was fully intentional. I decided to go for this one
because I saw some other areas of libc that don't fully comply with the
80 columns for comments, and 100 columns for doc comments style guide.

What style guide is that? If it suggests doc comments are 100 then this moves away from that.

In any case, I'd prefer to leave as-is.

tgross35 · 2026-09-07T04:29:40Z

    pub struct timespec {
        pub tv_sec: time_t,
        pub tv_nsec: c_long,
    }


Could you mark this one #[exhaustive]? timespec shouldn't ever get new fields.

tgross35 · 2026-09-07T04:33:47Z

Same thing for the structs in this file that don't have private fields. We can probably remove them at some point, feel free to add a FIXME, but we should keep the breakage smaller.

Do this in the first commit (adding non-exhaustive/exhaustive) to avoid changing this in one patch then restoring behavior after.

tgross35 · 2026-09-07T04:55:23Z

            cfg_attrs: { }
            other_attrs: { }
            remaining_attrs: { $($attrs)* $(#$attr)* }
+            found_exhaustive: { false }


Nit: either call this found_exhaustive_attr, or needs_non_exhaustive_field and flip the logic

tgross35 · 2026-09-07T05:03:56Z

+    (
+        found_exhaustive: { false },
+        expansion: { decl },
+        body: { $(#[$attr:meta])* $vis:vis $name:ident { $($field:tt)* } }
+    ) => {
+        $(#[$attr])*
+        $vis struct $name { $($field)* __non_exhaustive: () }
+    };
+    (
+        found_exhaustive: { false },
+        expansion: { default_impl },
+        body: { $($field_default:tt)* }
+    ) => {
+        Self { $($field_default)* __non_exhaustive: () }
+    };
+
+    (
+        found_exhaustive: { true },
+        expansion: { decl },
+        body: { $(#[$attr:meta])* $vis:vis $name:ident { $($field:tt)* } }
+    ) => {
+        $(#[$attr])*
+        $vis struct $name { $($field)* }
+    };
+    (
+        found_exhaustive: { true },
+        expansion: { default_impl },
+        body: { $($field_default:tt)* }
+    ) => {
+        Self { $($field_default)* }
+    };


Nit: match simple statements without the braces so it doesn't look like something is getting grouped

found_exhaustive: true, expansion: default_impl,

tgross35 · 2026-09-07T05:07:28Z

+        finalize_exhaustiveness! {
+            found_exhaustive: { $found_exhaustive },
+            expansion: { decl },
+            body: {
+                $($other_attrs)*
+                $($cfg_attrs)*
+                $vis $name { $($processed_fields)* }
+            }
+        }

        $($cfg_attrs)*
-        // The impl names the type and its fields, which warns if either is deprecated.
+        // The impl names the type and its fields, which warns if either is
+        // deprecated.
        #[allow(deprecated)]
        impl ::core::default::Default for $name {
-            // Field attributes (`#[cfg]`, doc comments) get forwarded to the initializer too.
-            // Docs are harmless there but trip the lint, so silence it.
+            // Field attributes (`#[cfg]`, doc comments) get forwarded to the
+            // initializer too. Docs are harmless there but trip the lint, so
+            // silence it.
            #[allow(unused_doc_comments)]
            fn default() -> Self {
-                Self { $($processed_field_defaults)* }
+                finalize_exhaustiveness! {
+                    found_exhaustive: { $found_exhaustive },
+                    expansion: { default_impl },
+                    body: { $($processed_field_defaults)* }
+                }


Rather than use the same finalize_exhaustiveness macro for the expansion and default impl, split into to macros like emit_struct_definition and emit_struct_default_body to make things a bit easier to follow.

- Modify `struct_with_default` to automatically add a private field to all records such that they are always built field-by-field in downstream crates. - Add support for `exhaustive` custom attribute to opt out of having the record have a `__non_exhaustive` field added to it. - Annotate records without private fields under `src/new/helenos/time.rs` and `src/new/linux_uapi/linux/can.rs` with `exhaustive` attribute.

Add tests to ensure two things about the private field added to enforce non-exhaustiveness in records declared within `s_with_default` and `s_no_extra_traits_with_default`. - Ensure the field is added only if the `exhaustive` attribute is not found while munching the annotated attributes of the record item. - Ensure the `exhaustive` attribute also works when it appears between other sets of attributes.

- Rename `s_with_default`, `s_no_extra_traits_with_default` and `struct_with_default` to `s2`, `s_no_extra_traits2`, and `custom_struct`. These macros now both add a `Default` impl and a private field to enforce non-exhaustiveness. - Rewrap comments and doc comments to follow 80 and 100 columns guides.

rustbot · 2026-09-08T16:33:54Z

This PR was rebased onto a different main 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.

rustbot added the S-waiting-on-review label Aug 8, 2026

dybucc changed the title ~~macros: add support for exhaustive attr~~ macros: add support for exhaustive attribute Aug 8, 2026

dybucc mentioned this pull request Aug 8, 2026

Mark all structs non_exhaustive for 1.0 #4080

Open

dybucc force-pushed the non_exhaustive-macro-extension branch from 8c61548 to 5a738cb Compare August 8, 2026 08:06

rustbot added the stable-nominated This PR should be considered for cherry-pick to libc's stable release branch label Aug 8, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch from 5a738cb to f11b352 Compare August 8, 2026 08:13

tgross35 reviewed Aug 25, 2026

View reviewed changes

Comment thread src/macros.rs Outdated

Comment thread src/macros.rs Outdated

tgross35 reviewed Aug 25, 2026

View reviewed changes

Comment thread src/lib.rs Outdated

dybucc force-pushed the non_exhaustive-macro-extension branch from f11b352 to f38fc87 Compare August 29, 2026 10:24

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-review label Aug 29, 2026

rustbot added the S-waiting-on-author label Aug 29, 2026

rustbot added S-waiting-on-review and removed S-waiting-on-author labels Aug 29, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch from 72651eb to f596ba7 Compare August 30, 2026 13:35

This comment has been minimized.

Sign in to view

tgross35 requested changes Aug 31, 2026

View reviewed changes

rustbot added S-waiting-on-author and removed S-waiting-on-review labels Aug 31, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch from f596ba7 to 8a61e3e Compare September 1, 2026 11:53

rustbot added the O-linux label Sep 1, 2026

This comment has been minimized.

Sign in to view

rustbot added S-waiting-on-review and removed S-waiting-on-author labels Sep 1, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch 2 times, most recently from e2d5adb to a58d229 Compare September 1, 2026 12:44

rustbot added O-bsd O-freebsd labels Sep 1, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch from a58d229 to d492384 Compare September 1, 2026 12:45

tgross35 requested changes Sep 7, 2026

View reviewed changes

rustbot added S-waiting-on-author and removed S-waiting-on-review labels Sep 7, 2026

dybucc force-pushed the non_exhaustive-macro-extension branch from d492384 to ccf62b2 Compare September 8, 2026 13:01

dybucc added 3 commits September 8, 2026 18:29

dybucc force-pushed the non_exhaustive-macro-extension branch from ccf62b2 to dc74109 Compare September 8, 2026 16:33

Uh oh!

Conversation

dybucc commented Aug 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Uh oh!

tgross35 commented Aug 25, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

dybucc commented Aug 29, 2026

Uh oh!

rustbot commented Aug 29, 2026

Uh oh!

dybucc commented Aug 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Footnotes

Uh oh!

This comment has been minimized.

This comment has been minimized.

tgross35 left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

dybucc commented Sep 1, 2026

Uh oh!

tgross35 left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

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!

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!

Choose a reason for hiding this comment

Uh oh!

rustbot commented Sep 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

dybucc commented Aug 8, 2026 •

edited

Loading

dybucc commented Aug 29, 2026 •

edited

Loading

tgross35 left a comment •

edited by rustbot

Loading

tgross35 left a comment •

edited by rustbot

Loading