ctest: add suport for C enum

[mbyx]

Description

Adds the ability to specify whether a type alias and a set of constants that use that type alias maps to a C enum, as well as the ability to skip them.

Sources

Checklist

• Relevant tests in libc-test/semver have been updated
• No placeholder or unstable values like *LAST or *MAX are
  included (see #3131)
• Tested locally (cd libc-test && cargo test --target mytarget);
  especially relevant for platforms that may not be checked in CI

[mbyx]

The signededness of a C enum is implementation defined, so should that test be skipped for all C enums?

[tgross35]

Could this take &Type rather than &str? To match rename_alias.

Maybe the name alias_is_c_enum would be a bit more descriptive, there are a handful of ways to represent C enums https://mdaverde.com/posts/rust-bindgen-enum/

[mbyx]

It can't, because it's also used to filter out constants with the same type. The only way to do it would be to create a Type struct in that method, but then the public and ty fields would be incorrect.

[tgross35]

The signededness of a C enum is implementation defined, so should that test be skipped for all C enums?

repr(C) should generally match. I know there might be some problems with that, but have you run into any here?

Could you add a test in ctest-test to ensure this catches incorrect item values and things like the wrong repr type?

[mbyx]

If I use an unsigned type in simple.h for the enum, it fails on the signededness test on windows. If I use a signed type it fails on linux. This isn't really a problem in the actual ports because the windows port doesn't use enums, but these integration tests are run on all platforms.

[tgross35]

I think that's actually an important catch; the fallback here

libc/src/macros.rs

Line 279 in 2e670a8

(@ty) => { $crate::c_uint };

should probably be updated so the type is signed on Windows.

[mbyx]

The signededness of a C enum is implementation defined, so should that test be skipped for all C enums?

repr(C) should generally match. I know there might be some problems with that, but have you run into any here?

Could you add a test in ctest-test to ensure this catches incorrect item values and things like the wrong repr type?

Since it piggybacks on tests for constants, incorrect item values would be caught with that. As for the repr type I'm not sure how to catch that. If it uses too small a repr size the constant test would overflow and catch it. For now ctest-test is only run on one platform, but on others a different type would have to be used.

[mbyx]

FreeBSD is strange. A lot of the times it segfaults on macros.rs, but it only seems to segfault when rustc is called.

[tgross35]

The signededness of a C enum is implementation defined, so should that test be skipped for all C enums?

repr(C) should generally match. I know there might be some problems with that, but have you run into any here?
Could you add a test in ctest-test to ensure this catches incorrect item values and things like the wrong repr type?

Since it piggybacks on tests for constants, incorrect item values would be caught with that. As for the repr type I'm not sure how to catch that. If it uses too small a repr size the constant test would overflow and catch it. For now ctest-test is only run on one platform, but on others a different type would have to be used.

I think this could look something like this:

pub type enum_repr_too_small = u16;
pub const ENUM_REPR_TOO_SMALL_A: enum_repr_too_small = 0;

#[cfg(windows)]
pub type enum_wrong_signedness = i32;
#[cfg(not(windows))]
pub const ENUM_WRONG_SIGNEDNESS_A: enum_wrong_signedness = 0;

mapping to C:

enum enum_repr_too_small {
    ENUM_REPR_TOO_SMALL_A;
}

enum enum_wrong_signedness {
    ENUM_WRONG_SIGNEDNESS_A;
}

And then set these as alias_is_c_enum. Then the test at

libc/ctest-test/tests/all.rs

Line 39 in 10d68a4

fn t2() {

should see the right outputs

[tgross35]

Instead of using different macro arms, add a pub(crate) type CEnumRepr to src/types.rs and use it here.

This will also simplify the tests, you can do e.g. 0 as CEnumRepr rather than splitting on Windows