pyforge-ffi 0.3.0

CPython 3.11+ C-API bindings for the PyForge ecosystem
Documentation

pyforge-ffi

This crate provides Rust FFI declarations for Python 3. It supports both the stable and the unstable component of the ABI through the use of cfg flags. Python 3.11+ is supported (CPython only). It is meant for advanced users only - regular PyForge users shouldn't need to interact with this crate at all.

The contents of this crate are not documented here, as it would entail basically copying the documentation from CPython. Consult the Python/C API Reference Manual for up-to-date documentation.

Minimum supported Rust and Python versions

Requires Rust 1.83 or greater.

pyforge-ffi supports CPython 3.11 or greater.

Example: Building Python Native modules

PyForge can be used to generate a native Python module. The easiest way to try this out for the first time is to use maturin. maturin is a tool for building and publishing Rust-based Python packages with minimal configuration. The following steps set up some files for an example Python module, install maturin, and then show how to build and import the Python module.

First, create a new folder (let's call it string_sum) containing the following two files:

Cargo.toml

[lib]
name = "string_sum"
# "cdylib" is necessary to produce a shared library for Python to import from.
#
# Downstream Rust code (including code in `bin/`, `examples/`, and `tests/`) will not be able
# to `use string_sum;` unless the "rlib" or "lib" crate type is also included, e.g.:
# crate-type = ["cdylib", "rlib"]
crate-type = ["cdylib"]

[dependencies]
pyforge-ffi = "0.1.0"

[build-dependencies]
# This is only necessary if you need to configure your build based on
# the Python version or the compile-time configuration for the interpreter.
pyforge_build_config = "0.1.0"

If you need to use conditional compilation based on Python version or how Python was compiled, you need to add pyforge-build-config as a build-dependency in your Cargo.toml as in the example above and either create a new build.rs file or modify an existing one so that pyforge_build_config::use_pyforge_cfgs() gets called at build time:

build.rs

fn main() {
    pyforge_build_config::use_pyforge_cfgs()
}

src/lib.rs

#[cfg(Py_3_15)]
use std::ffi::c_void;
use std::ffi::{c_char, c_long};
use std::ptr;

use pyforge_ffi::*;

#[cfg(not(Py_3_15))]
static mut MODULE_DEF: PyModuleDef = PyModuleDef {
    m_base: PyModuleDef_HEAD_INIT,
    m_name: c"string_sum".as_ptr(),
    m_doc: c"A Python module written in Rust.".as_ptr(),
    m_size: 0,
    m_methods: (&raw mut METHODS).cast(),
    m_slots: (&raw mut SLOTS).cast(),
    m_traverse: None,
    m_clear: None,
    m_free: None,
};

static mut METHODS: [PyMethodDef; 2] = [
    PyMethodDef {
        ml_name: c"sum_as_string".as_ptr(),
        ml_meth: PyMethodDefPointer {
            PyCFunctionFast: sum_as_string,
        },
        ml_flags: METH_FASTCALL,
        ml_doc: c"returns the sum of two integers as a string".as_ptr(),
    },
    // A zeroed PyMethodDef to mark the end of the array.
    PyMethodDef::zeroed(),
];

#[cfg(Py_3_15)]
PyABIInfo_VAR!(ABI_INFO);

const SLOTS_LEN: usize =
    1 + cfg!(Py_3_12) as usize + cfg!(Py_GIL_DISABLED) as usize + 4 * (cfg!(Py_3_15) as usize);
static mut SLOTS: [PyModuleDef_Slot; SLOTS_LEN] = [
    #[cfg(Py_3_15)]
    PyModuleDef_Slot {
        slot: Py_mod_abi,
        value: (&raw mut ABI_INFO).cast(),
    },
    #[cfg(Py_3_15)]
    PyModuleDef_Slot {
        slot: Py_mod_name,
        // safety: Python does not write to this field
        value: c"string_sum".as_ptr() as *mut c_void,
    },
    #[cfg(Py_3_15)]
    PyModuleDef_Slot {
        slot: Py_mod_doc,
        // safety: Python does not write to this field
        value: c"A Python module written in Rust.".as_ptr() as *mut c_void,
    },
    #[cfg(Py_3_15)]
    PyModuleDef_Slot {
        slot: Py_mod_methods,
        value: (&raw mut METHODS).cast(),
    },
    #[cfg(Py_3_12)]
    PyModuleDef_Slot {
        slot: Py_mod_multiple_interpreters,
        value: Py_MOD_PER_INTERPRETER_GIL_SUPPORTED,
    },
    #[cfg(Py_GIL_DISABLED)]
    PyModuleDef_Slot {
        slot: Py_mod_gil,
        value: Py_MOD_GIL_NOT_USED,
    },
    PyModuleDef_Slot {
        slot: 0,
        value: ptr::null_mut(),
    },
];

// The module initialization function
#[cfg(not(Py_3_15))]
#[allow(non_snake_case, reason = "must be named `PyInit_<your_module>`")]
#[no_mangle]
pub unsafe extern "C" fn PyInit_string_sum() -> *mut PyObject {
    PyModuleDef_Init(&raw mut MODULE_DEF)
}

#[cfg(Py_3_15)]
#[allow(non_snake_case, reason = "must be named `PyModExport_<your_module>`")]
#[no_mangle]
pub unsafe extern "C" fn PyModExport_string_sum() -> *mut PyModuleDef_Slot {
    (&raw mut SLOTS).cast()
}

/// A helper to parse function arguments
/// If we used PyForge's proc macros they'd handle all of this boilerplate for us :)
unsafe fn parse_arg_as_i32(obj: *mut PyObject, n_arg: usize) -> Option<i32> {
    if PyLong_Check(obj) == 0 {
        let msg = format!(
            "sum_as_string expected an int for positional argument {}\0",
            n_arg
        );
        PyErr_SetString(PyExc_TypeError, msg.as_ptr().cast::<c_char>());
        return None;
    }

    // Let's keep the behaviour consistent on platforms where `c_long` is bigger than 32 bits.
    // In particular, it is an i32 on Windows but i64 on most Linux systems
    let mut overflow = 0;
    let i_long: c_long = PyLong_AsLongAndOverflow(obj, &mut overflow);

    #[allow(
        irrefutable_let_patterns,
        reason = "some platforms have c_long equal to i32"
    )]
    if overflow != 0 {
        raise_overflowerror(obj);
        None
    } else if let Ok(i) = i_long.try_into() {
        Some(i)
    } else {
        raise_overflowerror(obj);
        None
    }
}

unsafe fn raise_overflowerror(obj: *mut PyObject) {
    let obj_repr = PyObject_Str(obj);
    if !obj_repr.is_null() {
        let mut size = 0;
        let p = PyUnicode_AsUTF8AndSize(obj_repr, &mut size);
        if !p.is_null() {
            let s = std::str::from_utf8_unchecked(std::slice::from_raw_parts(
                p.cast::<u8>(),
                size as usize,
            ));
            let msg = format!("cannot fit {} in 32 bits\0", s);

            PyErr_SetString(PyExc_OverflowError, msg.as_ptr().cast::<c_char>());
        }
        Py_DECREF(obj_repr);
    }
}

pub unsafe extern "C" fn sum_as_string(
    _self: *mut PyObject,
    args: *mut *mut PyObject,
    nargs: Py_ssize_t,
) -> *mut PyObject {
    if nargs != 2 {
        PyErr_SetString(
            PyExc_TypeError,
            c"sum_as_string expected 2 positional arguments".as_ptr(),
        );
        return std::ptr::null_mut();
    }

    let (first, second) = (*args, *args.add(1));

    let first = match parse_arg_as_i32(first, 1) {
        Some(x) => x,
        None => return std::ptr::null_mut(),
    };
    let second = match parse_arg_as_i32(second, 2) {
        Some(x) => x,
        None => return std::ptr::null_mut(),
    };

    match first.checked_add(second) {
        Some(sum) => {
            let string = sum.to_string();
            PyUnicode_FromStringAndSize(string.as_ptr().cast::<c_char>(), string.len() as isize)
        }
        None => {
            PyErr_SetString(PyExc_OverflowError, c"arguments too large to add".as_ptr());
            std::ptr::null_mut()
        }
    }
}

With those two files in place, now maturin needs to be installed. This can be done using Python's package manager pip. First, load up a new Python virtualenv, and install maturin into it:

$ cd string_sum
$ python -m venv .env
$ source .env/bin/activate
$ pip install maturin

Now build and execute the module:

$ maturin develop
# lots of progress output as maturin runs the compilation...
$ python
>>> import string_sum
>>> string_sum.sum_as_string(5, 20)
'25'

As well as with maturin, it is possible to build using setuptools-rust or manually. Both offer more flexibility than maturin but require further configuration.

While most projects use the safe wrapper provided by PyForge, you can take a look at the orjson library as an example on how to use pyforge-ffi directly. For those well versed in C and Rust the tutorials from the CPython documentation can be easily converted to rust as well.