This is a fork of rusty-cheddar until such time as the rusty-binder version is available.
moz-cheddar is a library for converting Rust source files into C header files.
A note on versioning:
While moz-cheddar is still pre-v1.0.0
it will likely go through
numerous breaking changes. We attempt to follow semver and bump
the minor version any time a new feature is added or output
behavior is changed.
moz-cheddar targets C99 or later (for single line comments, and
use of stdint.h
and stdbool.h
).
The most useful way to use moz-cheddar is in a build script.
To do this add the following build-dependencies
section to
your Cargo.toml
(to use it as a normal library simply replace
build-dependencies
with dependencies
):
# Cargo.toml
[build-dependencies]
moz-cheddar = "0.4.0"
Then create the following build.rs
:
// build.rs
extern crate cheddar;
fn main() {
cheddar::Cheddar::new().expect("could not read manifest")
.run_build("include/my_header.h");
}
This should work as is providing you've set up your project correctly.
Don't forget to add a build = ...
to your [package]
section,
see the cargo docs for more info.
moz-cheddar will then create a my_header.h
file in include/
.
Note that moz-cheddar emits very few warnings, it is up to the
programmer to write a library which can be correctly called from C.
You can also place your API in a module to help keep your source code neat. To do this you must supply the name of the module to Cheddar, then ensure that the items are available in the top-level scope:
// build.rs
extern crate cheddar;
fn main() {
cheddar::Cheddar::new().expect("could not read manifest")
.module("c_api").expect("malformed module path")
.run_build("target/include/rusty.h");
}
// src/lib.rs
pub use c_api::*;
mod c_api {
// api goes here ...
}
There are also .compile()
and .compile_code()
methods for finer control.
In the examples below, boilerplate has been omitted from the header.
moz-cheddar converts pub type A = B
into typedef B A;
.
Types containing generics are ignored.
Rust:
type UInt32 = u32;
pub type UInt64 = u64;
pub type MyOption<T> = Option<T>
Header:
// Some boilerplate omitted.
typedef uint64_t UInt64;
// Some more boilerplate omitted.
moz-cheddar will convert public enums which are marked #[repr(C)]
.
If the enum is generic or contains tuple or struct variants then
cheddar
will fail. moz-cheddar should correctly handle explicit
discriminants.
Rust:
#[repr(C)]
pub enum Colours {
Red = -6,
Blue,
Green = 7,
Yellow,
}
// This would fail if it was #[repr(C)].
pub enum Tastes<T> {
Savoury(T),
Sweet,
}
// This would fail if it was public.
#[repr(C)]
enum Units {
Kg(f64),
M(f64),
S(f64),
A(f64),
K(f64),
Mol(f64),
Cd(f64),
}
Header:
// Some boilerplate omitted.
typedef enum Colours {
Colours_Red = -6,
Colours_Blue,
Colours_Green = 7,
Colours_Yellow,
} Colours;
// Some more boilerplate omitted.
Structs are handled very similarly to enums, they must be public,
marked #[repr(C)]
, and they must not contain generics.
This currently only checked at the struct-level.
Generic fields are not checked.
Rust:
#[repr(C)]
pub struct Person {
age: i32,
height: f64,
weight: f64,
}
Header:
// Some boilerplate omitted.
typedef struct Person {
int32_t age;
double height;
double weight;
} Person;
// Some more boilerplate omitted.
One common C idiom is to hide the implementation of a struct using an opaque struct, which can only be used behind a pointer. This is especially useful in Rust-C interfaces as it allows you to use any arbitrary Rust struct in C.
To define an opaque struct you must define a public newtype which
is marked as #[repr(C)]
.
Rust:
struct Foo<T> {
bar: i32,
baz: Option<T>,
}
#[repr(C)]
pub struct MyCrate_Foo(Foo<PathBuf>);
Header:
// Some boilerplate omitted.
typedef struct MyCrate_Foo MyCrate_Foo;
// Some boilerplate omitted.
Note that the newtype must not be generic but the type that it wraps can be arbitrary.
For moz-cheddar to pick up on a function declaration it must be public,
marked #[no_mangle]
and have one of the following ABIs:
- C
- Cdecl
- Stdcall
- Fastcall
- System
If you believe one of these has been included in error, or if one has been omitted, then please open an issue at the repo.
moz-cheddar will fail on functions which are marked as diverging (-> !
).
Rust:
use std::ops::Add;
#[no_mangle]
pub extern fn hello() {
println!("Hello!");
}
fn add<O, R, L: Add<R, Output=O>>(l: L, r: R) -> O {
l + r
}
#[no_mangle]
#[allow(non_snake_case)]
pub extern fn MyAdd_add_u8(l: u8, r: u8) -> u8 {
add(l, r)
}
#[no_mangle]
#[allow(non_snake_case)]
pub extern fn MyAdd_add_u16(l: u16, r: u16) -> u16 {
add(l, r)
}
Header:
// Some boilerplate omitted.
void hello();
uint8_t MyAdd_add_u8(uint8_t l, uint8_t r);
uint16_t MyAdd_add_u16(uint16_t l, uint16_t r);
// Some more boilerplate omitted.
You must not put types defined in other modules in an exported type signature without hiding it behind an opaque struct. This is because the C compiler must know the layout of the type and moz-cheddar can not yet search other modules.
The very important exception to this rule are the C ABI types defined in
the libc
crate and std::os::raw
. Types from these two modules must
be fully qualified (e.g. libc::c_void
or std::os::raw::c_longlong
)
so that they can be converted properly. Importing them with a use
statement will not work.
Contributions to moz-cheddar are more than welcome.
If you find a bug or have a feature request please open an issue.
If you find the source code unclear in any way then I consider that a bug. I try to make my source code as clear as possible, so any help in that regard is appreciated.
Pull requests are also welcome, of course.
The tests require you to have a version (> v2.7.2
) of CppHeaderParser
installed for the version of Python which is installed as python
(usually Python 2). Furthermore, due to the fact that the tests are a
massive pile of wanky hacks, you must be in the same directory as
moz-cheddar's Cargo.toml
to successfully run them.
If you don't have this, try
pip install CppHeaderParser