Fix test case from #2246

[youknowone]

close #2246 fix align codegen of #2240 fix align codegen of #2159 fix align codegen of #1538

[youknowone]

mm.. test passes only because test doesn't include case about:

typedef unsigned char uint8_t;
typedef unsigned short uint16_t;

struct A {
  uint8_t a;
  uint16_t b;
  uint8_t c;
} __attribute__((aligned(2)));

struct B {
  uint8_t a;
  uint16_t b;
  uint8_t c;
} __attribute__((aligned(2), packed));


#include <assert.h>
#include <stdio.h>
int main() {
  printf("%lu %lu\n", sizeof(struct A), sizeof(struct B));
  assert(sizeof(struct A) == sizeof(struct B));
}

[youknowone]

is this the answer?

#[repr(C, align(2))]
#[derive(Debug, Default, Copy, Clone)]
pub struct B(B__packed);

#[repr(C, packed)]
#[derive(Debug, Default, Copy, Clone)]
pub struct B__packed {
    pub a: u8,
    pub b: u16,
    pub c: u8,
}

#[test]
fn bindgen_test_layout_B() {
    const UNINIT: ::std::mem::MaybeUninit<B> =
        ::std::mem::MaybeUninit::uninit();
    let ptr = UNINIT.as_ptr();
    assert_eq!(
        ::std::mem::size_of::<B>(),
        4usize,
        concat!("Size of: ", stringify!(B))
    );
    assert_eq!(
        ::std::mem::align_of::<B>(),
        2usize,
        concat!("Alignment of ", stringify!(B))
    );
    assert_eq!(
        unsafe { ::std::ptr::addr_of!((*ptr).0.a) as usize - ptr as usize },
        0usize,
        concat!("Offset of field: ", stringify!(B), "::", stringify!(a))
    );
    assert_eq!(
        unsafe { ::std::ptr::addr_of!((*ptr).0.b) as usize - ptr as usize },
        1usize,
        concat!("Offset of field: ", stringify!(B), "::", stringify!(b))
    );
    assert_eq!(
        unsafe { ::std::ptr::addr_of!((*ptr).0.c) as usize - ptr as usize },
        3usize,
        concat!("Offset of field: ", stringify!(B), "::", stringify!(c))
    );
}

[youknowone]

I added the last commit to check the CI result. I don't believe that's the correct way. Please suppose I am not understanding the remaining problems and how bindgen handle packed and aligned.

[youknowone]

@emilio do you have advices?

[youknowone]

@emilio can I get advice for remaining issues?

[emilio]

Sorry for the lag, it's on my list of things to do. I plan to dig a bit into how C++ compilers treat repr+aligned hints because I'm not super-convinced there's a way to emulate it in Rust, but if you've done that research and can point to it it'd be extremely helpful.

[youknowone]

I think emulating in rust is done here. I forgot the datails now but If you need explanation, I can review it again.

If bindgen cannot distinguish explicitly aligned and implicitly aligned struct, could you guide me to how to get clang-calcuated size and bindgen-calculated size for the struct? Then I will try to detect which alignment used by them.

[bors-servo]

:umbrella: The latest upstream changes (presumably #2284) made this pull request unmergeable. Please resolve the merge conflicts.

[CGMossa]

Can I kindly ask that this gets looked at please?

[pvdrz]

I've seen several issues and PRs discussing problems with packed and align representations. I'll gather some info and have a proper discussion around it with other contributors.