Unpacking methods on sequences
Python supports unpacking all objects which support the Sequence protocol. I believe it does so by simply calling __getitem__ for consecutive integer indices.
Currently in PyO3 only tuple objects can be extracted as Rust tuples. I think it'd be nice to have an unpack method which would emulate Python's unpacking behavior and support all objects which implement the Sequence interface.
I'd implement it as a utility trait:
trait Unpackable<'py>: Sized {
fn unpack(obj: Borrowed<'_, 'py, PyAny>) -> PyResult<Self>;
}
Which can be implemented for tuples and arrays up to a certain size.
impl<'py, T0, T1, T2> Unpackable<'py> for (T0, T1, T2)
where
T0: for<'a> FromPyObject<'a, 'py>,
T1: for<'a> FromPyObject<'a, 'py>,
T2: for<'a> FromPyObject<'a, 'py>,
{
fn unpack(obj: Borrowed<'_, 'py, PyAny>) -> PyResult<Self> {
let t0 = match obj.get_item(0)?.extract::<T0>() {
Ok(v) => v,
Err(e) => return Err(e.into()),
};
let t1 = match obj.get_item(1)?.extract::<T1>() {
Ok(v) => v,
Err(e) => return Err(e.into()),
};
let t2 = match obj.get_item(2)?.extract::<T2>() {
Ok(v) => v,
Err(e) => return Err(e.into()),
};
Ok((t0, t1, t2))
}
}
I used match here because I couldn't get the Into<PyErr> try operator inference to work on extract in this example.
The trait does not support depending on the 'a lifetime of the object because get_item creates new objects which are tied to the scope of the unpack function.
Additionally, this function can use __len__ for better error messages. But that would technically deviate from the official implementation: Python allows returning nonsensical lengths in __len__, unpacking will work correctly regardless
Thanks for the suggestion, it's an interesting idea. You present the trait, I guess users could call as
// subject to type inference
let (x, y, z) = Unpackable::unpack(obj)?;
I wonder if we can make it more obvious that this is mirroring Python sequence unpacking...
Type inference might be tricky here, to the same degree that extract has to deal with. It'll probably require annotations somewhere. I personally prefer the let form:
use pyo3::Unpackable;
let (a, c, g, t): (f64, f64, f64, f64) = obj.unpack()?;
Another alternative would be to add a standalone function with a generic:
let (a, c, g, t) = unpack::<(f64, f64, f64, f64)>(obj)?;
Personally, I find Rust destructuring similar to Python's unpacking. I a bunch of snippets like those ones my mixed Python/Rust codebase:
let (left, right) = tree.children_of(&internal);
left, right = tree.children_of(node)