Macro cpython::py_class

source ·

macro_rules! py_class {
    (class $class:ident |$py: ident| { $( $body:tt )* }) => { ... };
    ($visibility:vis class $class:ident |$py: ident| { $( $body:tt )* }) => { ... };
}

Expand description

Defines new python extension class. A py_class! macro invocation generates code that declares a new Python class. Additionally, it generates a Rust struct of the same name, which allows accessing instances of that Python class from Rust.

§Syntax

py_class!(pub class MyType |py| { ... })

pub makes the generated Rust struct visible outside the current module. It has no effect on the visibility from Python. You may use any Rust visibility keyword. For example, pub(crate) would also be valid.
MyType is the name of the Python class.
py is an identifier that will be made available as a variable of type Python in all function bodies.
{ ... } is the class body, described in more detail below.

§Example

use cpython::{Python, PyResult, PyDict, py_class};

py_class!(class MyType |py| {
    data number: i32;
    def __new__(_cls, arg: i32) -> PyResult<MyType> {
        MyType::create_instance(py, arg)
    }
    def half(&self) -> PyResult<i32> {
        println!("half() was called with self={:?}", self.number(py));
        Ok(self.number(py) / 2)
    }
});

fn main() {
    let gil = Python::acquire_gil();
    let py = gil.python();
    let dict = PyDict::new(py);
    dict.set_item(py, "MyType", py.get_type::<MyType>()).unwrap();
    py.run("assert MyType(42).half() == 21", None, Some(&dict)).unwrap();
}

§Generated Rust type

The above example generates the following Rust type:

struct MyType { ... }

impl ToPythonObject for MyType { ... }
impl PythonObject for MyType { ... }
impl PythonObjectWithCheckedDowncast for MyType { ... }
impl PythonObjectWithTypeObject for MyType { ... }
impl PythonObjectFromPyClassMacro for MyType { ... }

impl MyType {
    fn create_instance(py: Python, number: i32) -> PyResult<MyType> { ... }

    // data accessors
    fn number<'a>(&'a self, py: Python<'a>) -> &'a i32 { ... }

    // functions callable from python
    pub fn __new__(_cls: &PyType, py: Python, arg: i32) -> PyResult<MyType> {
        MyType::create_instance(py, arg)
    }

    pub fn half(&self, py: Python) -> PyResult<i32> {
        println!("half() was called with self={:?}", self.number(py));
        Ok(self.number(py) / 2)
    }
}

The generated type implements a number of traits from the cpython crate.
The inherent create_instance method can create new Python objects given the values for the data fields.
- Note: Any visibility keyword on the class will also be used for this method.
Private accessors functions are created for the data fields.
All functions callable from Python are also exposed as public Rust functions.
To convert from MyType to PyObject, use as_object() or into_object() (from the PythonObject trait).
To convert PyObject to MyType, use obj.cast_as::<MyType>(py) or obj.cast_into::<MyType>(py).

§py_class body

The body of a py_class! supports the following definitions:

§Data declarations

data data_name: data_type;

Declares a data field within the Python class. Used to store Rust data directly in the Python object instance.

Because Python code can pass all Python objects to other threads, data_type must be Send + 'static.

Because Python object instances can be freely shared (Python has no concept of “ownership”), data fields cannot be declared as mut. If mutability is required, you have to use interior mutability (Cell or RefCell).

If data members are used to store references to other Python objects, make sure to read the section “Garbage Collector Integration”.

Data declarations are not accessible from Python. On the Rust side, data is accessed through the automatically generated accessor functions:

impl MyType {
    fn data_name<'a>(&'a self, py: Python<'a>) -> &'a data_type { ... }
}

§Shared data declarations

@shared data data_name: data_type;

Declares a “shareable” data field within the Python class. A data field of this type can be borrowed immutably by another Python object such as a Python iterator. See PySharedRefCell documentation for details.

On the Rust side, data is accessed through the automatically generated accessor functions:

impl MyType {
    fn data_name<'a>(&'a self, py: Python<'a>) -> PySharedRef<'a, $data_type> { ... }
}

§Instance methods

def method_name(&self, parameter-list) -> PyResult<...> { ... } pub(crate) def method_name(&self, parameter-list) -> PyResult<...> { ... }

Declares an instance method callable from Python.

Because Python objects are potentially shared, the self parameter must always be a shared reference (&self).
For details on parameter-list, see the documentation of py_argparse!().
The return type must be PyResult<T> for some T that implements ToPyObject.
Visibility of the method in Rust defaults to pub. You may specify a visibility keyword before the def to change the visibility, for example, to pub(crate). Changing visibility in Rust does not affect visibility in Python.

§Class methods

@classmethod def method_name(cls, parameter-list) -> PyResult<...> { ... } @classmethod pub(crate) def method_name(cls, parameter-list) -> PyResult<...> { ... }

Declares a class method callable from Python.

The first parameter is the type object of the class on which the method is called. This may be the type object of a derived class.
The first parameter implicitly has type &PyType. This type must not be explicitly specified.
For details on parameter-list, see the documentation of py_argparse!().
The return type must be PyResult<T> for some T that implements ToPyObject.
Visibility of the method in Rust defaults to pub. You may specify a visibility keyword before the def to change the visibility, for example, to pub(crate). Changing visibility in Rust does not affect visibility in Python.

§Static methods

@staticmethod def method_name(parameter-list) -> PyResult<...> { ... } @staticmethod pub(crate) def method_name(parameter-list) -> PyResult<...> { ... }

Declares a static method callable from Python.

For details on parameter-list, see the documentation of py_argparse!().
The return type must be PyResult<T> for some T that implements ToPyObject.
Visibility of the method in Rust defaults to pub. You may specify a visibility keyword before the def to change the visibility, for example, to pub(crate). Changing visibility in Rust does not affect visibility in Python.

§Properties

@property def property_name(&self) -> PyResult<...> { ... }

@property_name.setter def set_method_name(&self, value: Option<impl FromPyObject>) -> PyResult<()> { ... }

Declares a Python data attribute backed by Rust methods to get its value and, optionally, to set or delete it.

§Setter details

The setter is optional. If omitted, the attribute will be read-only and any setting or deleting attempt will raise AttributeError.
Unlike Python, the setter method name must be different from the property name. The setter method name is used to call the setter from Rust.
A None value represents that the property is being deleted, for instance with Python del. As with Python properties, what should happen is entirely up to the implementation.
The value type can be any type that implements FromPyObject, or a reference or optional reference to any type that implements RefFromPyObject. In the latter case, the type of the value is Option<Option<&impl RefFromPyObject>>, where None means the property is being deleted, Some(None) means the property is being set to Python None, and Some(Some(value)) means the property is being set to the given value.

§new

def __new__(cls, parameter-list) -> PyResult<...> { ... }

Declares a constructor method callable from Python.

If no __new__ method is declared, object instances can only be created from Rust (via MyType::create_instance), but not from Python.
The first parameter is the type object of the class to create. This may be the type object of a derived class declared in Python.
The first parameter implicitly has type &PyType. This type must not be explicitly specified.
For details on parameter-list, see the documentation of py_argparse!().
The return type must be PyResult<T> for some T that implements ToPyObject. Usually, T will be MyType.

§Garbage Collector Integration

If your type owns references to other python objects, you will need to integrate with Python’s garbage collector so that the GC is aware of those references. To do this, implement the special member functions __traverse__ and __clear__. These correspond to the slots tp_traverse and tp_clear in the Python C API.

__traverse__ must call visit.call() for each reference to another python object.

__clear__ must clear out any mutable references to other python objects (thus breaking reference cycles). Immutable references do not have to be cleared, as every cycle must contain at least one mutable reference.

Example:

use std::{mem, cell};
use cpython::{PyObject, PyDrop, py_class};

py_class!(class ClassWithGCSupport |py| {
    data obj: cell::RefCell<Option<PyObject>>;

    def __traverse__(&self, visit) {
        if let Some(ref obj) = *self.obj(py).borrow() {
            visit.call(obj)?
        }
        Ok(())
    }

    def __clear__(&self) {
        let old_obj = mem::replace(&mut *self.obj(py).borrow_mut(), None);
        // Release reference only after the mutable borrow has expired,
        // see Caution note below.
        old_obj.release_ref(py);
    }
});

Caution: __traverse__ may be called by the garbage collector:

during any python operation that takes a Python token as argument
indirectly from the PyObject (or derived type) Drop implementation
if your code releases the GIL, at any time by other threads.

If you are using RefCell<PyObject>, you must not perform any of the above operations while your code holds a mutable borrow, or you may cause the borrow in __traverse__ to panic.

This is why the example above uses the mem::replace/release_ref dance: release_ref (or the implicit Drop) can only be called safely in a separate statement, after the mutable borrow on the RefCell has expired.

Note that this restriction applies not only to __clear__, but to all methods that use RefCell::borrow_mut.

§Iterator Types

Iterators can be defined using the Python special methods __iter__ and __next__:

def __iter__(&self) -> PyResult<impl ToPyObject>
def __next__(&self) -> PyResult<Option<impl ToPyObject>>

Returning Ok(None) from __next__ indicates that that there are no further items.