Implement Support For repr(int) non-C-like enums

[Gankra]

It turns out repr(u32, i8, etc...)on non-C-like enums has had an unofficially specified and C(++)-compatible layout for a good while. I have filed an RFC to make it officially specified, but in the mean time we can set up cbindgen to generate this layout, which will let us use native Rust tagged unions in our bindings, and with today's stable Rust!

You can check out the RFC for details (it's not quite the obvious layout), but here's a test Rust program to prove that it works:

works_for_me.rs

use std::time::Duration;
use std::mem;

#[repr(u8)]
#[derive(Copy, Clone, Eq, PartialEq, Debug)]
enum MyEnum {
    A(u32),
    B { x: u8, y: i16 },
    C,
    D(Option<u32>),
    E(Duration),
}

#[allow(non_snake_case)]
#[repr(C)]
union MyEnumRepr {
    A: MyEnumVariantA,
    B: MyEnumVariantB,
    C: MyEnumVariantC,
    D: MyEnumVariantD,
    E: MyEnumVariantE,
}

#[repr(u8)] #[derive(Copy, Clone)] enum MyEnumTag { A, B, C, D, E }
#[repr(C)] #[derive(Copy, Clone)] struct MyEnumVariantA(MyEnumTag, u32);
#[repr(C)] #[derive(Copy, Clone)] struct MyEnumVariantB { tag: MyEnumTag, x: u8, y: i16 }
#[repr(C)] #[derive(Copy, Clone)] struct MyEnumVariantC(MyEnumTag);
#[repr(C)] #[derive(Copy, Clone)] struct MyEnumVariantD(MyEnumTag, Option<u32>);
#[repr(C)] #[derive(Copy, Clone)] struct MyEnumVariantE(MyEnumTag, Duration);

fn main() {
    let result: Vec<Result<MyEnum, ()>> = vec![
        Ok(MyEnum::A(17)),
        Ok(MyEnum::B { x: 206, y: 1145 }),
        Ok(MyEnum::C),
        Err(()),
        Ok(MyEnum::D(Some(407))),
        Ok(MyEnum::D(None)),
        Ok(MyEnum::E(Duration::from_secs(100))),
        Err(()),
    ];
        
    let input: Vec<u8> = vec![
        0,  17, 0, 0, 0,
        1,  206,  121, 4,
        2,
        8,
        3,  0,  151, 1, 0, 0,
        3,  1,
        4,  100, 0, 0, 0,  0, 0, 0, 0,  0, 0, 0, 0,
        0,
    ];
    
    let mut output = vec![];
    let mut buf = &input[..];
    
    unsafe {
        let mut dest: MyEnum = mem::uninitialized();
        while buf.len() > 0 {
            match parse_my_enum(&mut dest, &mut buf) {
                Ok(()) => output.push(Ok(dest)),
                Err(()) => output.push(Err(())),
            }
        }
    }
    
    assert_eq!(output, result);
}

#[repr(u8)]
#[derive(Copy, Clone, Eq, PartialEq, Debug)]
enum MyEnum {
    A(u32),
    B { x: u8, y: i16 },
    C,
    D(Option<u32>),
    E(Duration),
}

fn parse_my_enum<'a>(dest: &'a mut MyEnum, buf: &mut &[u8]) -> Result<(), ()> {
    unsafe {
        let dest: &'a mut MyEnumRepr = mem::transmute(dest);
        let tag = read_u8(buf)?;
        
        dest.A.0 = match tag {
            0 => MyEnumTag::A,
            1 => MyEnumTag::B,
            2 => MyEnumTag::C,
            3 => MyEnumTag::D,
            4 => MyEnumTag::E,
            _ => return Err(()),
        };
                
        match dest.B.tag {
            MyEnumTag::A => {
                dest.A.1 = read_u32_le(buf)?;
            }
            MyEnumTag::B => {
                dest.B.x = read_u8(buf)?;
                dest.B.y = read_u16_le(buf)? as i16;
            }
            MyEnumTag::C => {
                /* do nothing */
            }
            MyEnumTag::D => {
                let is_some = read_u8(buf)? == 0;
                if is_some {
                    dest.D.1 = Some(read_u32_le(buf)?);
                } else {
                    dest.D.1 = None;
                }
            }
            MyEnumTag::E => {
                let secs = read_u64_le(buf)?;
                let nanos = read_u32_le(buf)?;
                dest.E.1 = Duration::new(secs, nanos);
            }
        }
        Ok(())
    }
}

fn read_u64_le(buf: &mut &[u8]) -> Result<u64, ()> {
    if buf.len() < 8 { return Err(()) }
    let val = (buf[0] as u64) << 0 
            | (buf[1] as u64) << 8
            | (buf[2] as u64) << 16
            | (buf[3] as u64) << 24
            | (buf[4] as u64) << 32
            | (buf[5] as u64) << 40
            | (buf[6] as u64) << 48
            | (buf[7] as u64) << 56;
    *buf = &buf[8..];
    Ok(val)
}

fn read_u32_le(buf: &mut &[u8]) -> Result<u32, ()> {
    if buf.len() < 4 { return Err(()) }
    let val = (buf[0] as u32) << 0 
            | (buf[1] as u32) << 8
            | (buf[2] as u32) << 16
            | (buf[3] as u32) << 24;
    *buf = &buf[4..];
    Ok(val)
}

fn read_u16_le(buf: &mut &[u8]) -> Result<u16, ()> {
    if buf.len() < 2 { return Err(()) }
    let val = (buf[0] as u16) << 0 
            | (buf[1] as u16) << 8;
    *buf = &buf[2..];
    Ok(val)
}

fn read_u8(buf: &mut &[u8]) -> Result<u8, ()> {
    if buf.len() < 1 { return Err(()) }
    let val = buf[0];
    *buf = &buf[1..];
    Ok(val)
}

[Gankra]

CC @jrmuizel @mstange

[eqrion]

Cool, this sounds really useful. The cbindgen work shouldn't be too difficult, just adding a new item and sharing some code with structs.

[RReverser]

Just so that we don't duplicate the work - I'm trying to implement this now.

[Gankra]

Relevant: rust-lang/rust#46123

[Gankra]

Above PR was merged, so now repr(C, int) and repr(C) should work as spec'd in the next nightly.

[RReverser]

I was stalled on refactoring generics in #85 (which affects tagged enums too), but happy to PR current state of things soon.