Replaces annotations don't work with base types

[agrover]

I use bindgen on the data structure shown, and then I'm left with signed char arrays that require a transmute to treat as bytes. for example:

fn hdr_set_name(hdr: &mut dmi::dm_ioctl, name: &str) -> () {
    let name_dest: &mut [u8; DM_NAME_LEN] = unsafe { transmute(&mut hdr.name) };
    let bytes = name.as_bytes();
    name_dest[..bytes.len()].clone_from_slice(bytes);
}

It would be great if bindgen could optionally translate these as unsigned chars instead of signed chars (the default). Or, maybe I'm going about this all wrong??? 😃

Input C/C++ Header

#define DM_NAME_LEN 128
#define DM_UUID_LEN 129
struct dm_ioctl {
	char name[DM_NAME_LEN];	/* device name */
	char uuid[DM_UUID_LEN];	/* unique identifier for
				 * the block device */
	char data[7];		/* padding or data */
};

Bindgen Invocation

    let bindings = bindgen::Builder::default()
        .header("wrapper.h")
        .derive_default(true)
        .impl_debug(true)
        .generate()
        .expect("Unable to generate bindings");

Actual Results

# [ repr ( C ) ]
# [ derive ( Copy , Clone ) ]
pub struct dm_ioctl {
    pub name: [::std::os::raw::c_char; 128usize],
    pub uuid: [::std::os::raw::c_char; 129usize],
    pub data: [::std::os::raw::c_char; 7usize],
}

Desired results

# [ repr ( C ) ]
# [ derive ( Copy , Clone ) ]
pub struct dm_ioctl {
    pub name: [::std::os::raw::c_uchar; 128usize],
    pub uuid: [::std::os::raw::c_uchar; 129usize],
    pub data: [::std::os::raw::c_uchar; 7usize],
}

[fitzgen]

Hi @agrover! Thanks for filing an issue.

Right now,

• If bindgen sees char then it translates into c_char.

• If bindgen sees signed char then it translates into c_schar.

• If bindgen sees unsigned char then it translates into c_uchar.

Note that c_char is a type alias of either c_schar or c_uchar depending on the target platform and whether char is signed or not on it.

So

struct chars {
    char c;
    unsigned char uc;
    signed char sc;
};

generates

#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct chars {
    pub c: ::std::os::raw::c_char,
    pub uc: ::std::os::raw::c_uchar,
    pub sc: ::std::os::raw::c_schar,
}

Given all that, I think bindgen exhibits desired behavior here. I recognize that this means that using the bindings might often end up with more explicit casts and transmutes than you might do in C, but I think that is as much a reflection of Rust as it is bindgen.

But perhaps I am overlooking or missing something?

[agrover]

I'm in agreement that this is correct behavior. Opening this issue was really to see if there might be an existing option in bindgen that could help tweak this, or to see if it was a common enough issue to warrant development ("custom type substitution"?) If not, we can keep the explicit casts, and we can close this issue. Thanks.

[fitzgen]

We do have <div rustbindgen replaces="..." /> annotations, but they apparently don't work properly with scalars:

struct chars {
    char c;
    unsigned char uc;
    signed char sc;
};

/// <div rustbindgen replaces="char" />
using replaces_char = unsigned char;

generates

/* automatically generated by rust-bindgen */

#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct chars {
    pub c: ::std::os::raw::c_char,
    pub uc: ::std::os::raw::c_uchar,
    pub sc: ::std::os::raw::c_schar,
}
/// <div rustbindgen replaces="char" />
pub type char = ::std::os::raw::c_uchar;

but it should have pub c: char, instead of the pub c: ::std::os:raw::c_char.

With this bug fixed, then we could force all char to be unsigned char.

[jhod0]

@fitzgen

Is this still something which should be implemented, and if so is this the main code which would need changing to implement this?

[fitzgen]

Is this still something which should be implemented

Yes

and if so is this the main code which would need changing to implement this?

No, we would want to do the replacement at the IR level, rather than during codegen.