Do not insert dummy fields into bitfields in cases where this would cause misalignment.

[flowerhack]

Suppose you have the following C++ code:

// reproduce.h
struct __attribute__((aligned(8))) StructWithBitfieldAndDouble {
  unsigned int bitfield: 32;
  double standard_field;
};

// reproduce.cpp
 #include "reproduce.h"
 #include <iostream>
int main() {
  std::cout << "cpp_size=" << sizeof(SomeStruct) << std::endl;
  std::cout << "cpp_align=" << alignof(SomeStruct) << std::endl;
  return 0;
}

As of right now, Bindgen will produce the following Rust code:

 #[repr(C)]
 #[repr(align(8))]
 #[derive(Debug, Default, Copy, Clone)]
pub struct StructWithBitfieldAndDouble {
  pub _bindgen_align: [u64; 0],
  pub _bitfield_1: __BindgenBitfieldUnit<[u8, 4usize]>,
  pub standard_field: f64,
}

If Bindgen compiles this targeting a 32-bit architecture, the C++ struct will be 16 bytes in size (as one would expect, based on the aligned(8)), but the Rust struct will be only 12 bytes in size (because the alignment of u64 is 4, so _bindgen_align, as a zero-length struct, ends up occupying 0 bytes of space, and the struct's overall size will thus be 12).

This occurs because of the way dummy fields are added to bitfields as per #3247.

We should check that the target primitive's alignment is greater-than-or-equal-to the explicitly-defined alignment in these cases, and fall back to simply using the explicit alignment in those cases.

[flowerhack]

r? @pvdrz

Hi, I'm new! but I read the CONTRIBUTING.md and believe I'm following all the guidelines there; let me know if I missed anything.

I didn't see any kind of LLM policy, but, in case there is one and I missed it:

• I used Gemini to help determine the root cause of the compilation failure we saw in Chromium, which was caused by this particular bit of Bindgen code.
• I verified Gemini's reasoning myself (the code in Chromium that caused the failure was code I wrote myself, so I was familiar with the area and confident I could evaluate it).
• I used Gemini to minimize the test case we had for demonstrating this failure.
• The fix, the commit message and all comments were written by myself.
• I used Bindgen's standard workflow for generating expectation files.