(Apologies if this already reported in some form: GitHub search shows a lot of semi-related open issues but none seem to define exactly this scenario - even if they may share the same problem in the end)
Consider the following bindings.hpp sample:
#pragma pack(4)
class Data {
void *test;
};
class Foo {
virtual void foo();
};
Running bindgen bindings.hpp (with a 64-bit target) prints:
/* automatically generated by rust-bindgen 0.71.1 */
#[repr(C, packed(4))]
#[derive(Debug, Copy, Clone)]
pub struct Data {
pub test: *mut ::std::os::raw::c_void,
}
#[allow(clippy::unnecessary_operation, clippy::identity_op)]
const _: () = {
["Size of Data"][::std::mem::size_of::<Data>() - 8usize];
["Alignment of Data"][::std::mem::align_of::<Data>() - 4usize];
["Offset of field: Data::test"][::std::mem::offset_of!(Data, test) - 0usize];
};
#[repr(C)]
pub struct Foo__bindgen_vtable(::std::os::raw::c_void);
#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct Foo {
pub vtable_: *const Foo__bindgen_vtable,
}
#[allow(clippy::unnecessary_operation, clippy::identity_op)]
const _: () = {
["Size of Foo"][::std::mem::size_of::<Foo>() - 8usize];
["Alignment of Foo"][::std::mem::align_of::<Foo>() - 4usize];
};
It seems that the layout tests picked up the #pragma pack(4) and expect an alignment of 4 for both Data and Foo (both contain pointer fields that are otherwise aligned to 8 bytes). However, packed(4) is only emitted for the "POD" Data structure but not for the Foo virtual class, meaning that these layout tests will now fail:
❯ rustc bindings.rs
...
error[E0080]: evaluation of constant value failed
--> bindings.rs:24:5
|
24 | ["Alignment of Foo"][::std::mem::align_of::<Foo>() - 4usize];
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ index out of bounds: the length is 1 but the index is 4
error: aborting due to 2 previous errors
Some errors have detailed explanations: E0080, E0601.
For more information about an error, try `rustc --explain E0080`.
The repr(C, packed(4)) should be emitted on struct Foo, and the alignment test is correct. After all, printing alignof(Foo) for the above code is 4 in cpp too.
Details
#include <iostream>
#pragma pack(4)
class Foo {
virtual void foo();
};
int main() {
std::cout << alignof(Foo) << std::endl;
}
$ clang -lstdc++ bindings.cpp && ./a.out
4
(Apologies if this already reported in some form: GitHub search shows a lot of semi-related open issues but none seem to define exactly this scenario - even if they may share the same problem in the end)
Consider the following
bindings.hppsample:Running
bindgen bindings.hpp(with a 64-bit target) prints:It seems that the layout tests picked up the
#pragma pack(4)and expect an alignment of4for bothDataandFoo(both contain pointer fields that are otherwise aligned to8bytes). However,packed(4)is only emitted for the "POD"Datastructure but not for theFoovirtual class, meaning that these layout tests will now fail:The
repr(C, packed(4))should be emitted onstruct Foo, and the alignment test is correct. After all, printingalignof(Foo)for the above code is4incpptoo.Details