polars-python 0.53.0

use arrow::array::{Array, PrimitiveArray};
use arrow::ffi;
use arrow::ffi::{ArrowArray, ArrowArrayStream, ArrowArrayStreamReader, ArrowSchema};
use polars::prelude::*;
use polars_ffi::version_0::SeriesExport;
use pyo3::exceptions::{PyTypeError, PyValueError};
use pyo3::prelude::*;
use pyo3::pybacked::PyBackedBytes;
use pyo3::types::{PyCapsule, PyTuple, PyType};

use super::PySeries;
use crate::error::PyPolarsErr;

/// Validate PyCapsule has provided name
fn validate_pycapsule_name(capsule: &Bound<PyCapsule>, expected_name: &str) -> PyResult<()> {
    let capsule_name = capsule.name()?;
    if let Some(capsule_name) = capsule_name {
        let capsule_name = unsafe { capsule_name.as_cstr() };
        if capsule_name.to_str() != Ok(expected_name) {
            return Err(PyValueError::new_err(format!(
                "Expected name '{expected_name}' in PyCapsule, instead got '{capsule_name:?}'"
            )));
        }
    } else {
        return Err(PyValueError::new_err(
            "Expected schema PyCapsule to have name set.",
        ));
    }

    Ok(())
}

/// Import `__arrow_c_array__` across Python boundary
pub(crate) fn call_arrow_c_array<'py>(
    ob: &Bound<'py, PyAny>,
) -> PyResult<(Bound<'py, PyCapsule>, Bound<'py, PyCapsule>)> {
    if !ob.hasattr("__arrow_c_array__")? {
        return Err(PyValueError::new_err(
            "Expected an object with dunder __arrow_c_array__",
        ));
    }

    let tuple = ob.getattr("__arrow_c_array__")?.call0()?;
    if !tuple.is_instance_of::<PyTuple>() {
        return Err(PyTypeError::new_err(
            "Expected __arrow_c_array__ to return a tuple.",
        ));
    }

    let schema_capsule = tuple.get_item(0)?.cast_into()?;
    let array_capsule = tuple.get_item(1)?.cast_into()?;
    Ok((schema_capsule, array_capsule))
}

pub(crate) fn import_array_pycapsules(
    schema_capsule: &Bound<PyCapsule>,
    array_capsule: &Bound<PyCapsule>,
) -> PyResult<(arrow::datatypes::Field, Box<dyn Array>)> {
    let field = import_schema_pycapsule(schema_capsule)?;

    validate_pycapsule_name(array_capsule, "arrow_array")?;

    // # Safety
    // array_capsule holds a valid C ArrowArray pointer, as defined by the Arrow PyCapsule
    // Interface
    unsafe {
        #[allow(deprecated)]
        let array_ptr = std::ptr::replace(array_capsule.pointer() as _, ArrowArray::empty());
        let array = ffi::import_array_from_c(array_ptr, field.dtype().clone()).unwrap();

        Ok((field, array))
    }
}

pub(crate) fn import_schema_pycapsule(
    schema_capsule: &Bound<PyCapsule>,
) -> PyResult<arrow::datatypes::Field> {
    validate_pycapsule_name(schema_capsule, "arrow_schema")?;

    // # Safety
    // schema_capsule holds a valid C ArrowSchema pointer, as defined by the Arrow PyCapsule
    // Interface
    unsafe {
        #[allow(deprecated)]
        let schema_ptr = schema_capsule.reference::<ArrowSchema>();
        let field = ffi::import_field_from_c(schema_ptr).unwrap();

        Ok(field)
    }
}

/// Import `__arrow_c_stream__` across Python boundary.
fn call_arrow_c_stream<'py>(ob: &Bound<'py, PyAny>) -> PyResult<Bound<'py, PyCapsule>> {
    if !ob.hasattr("__arrow_c_stream__")? {
        return Err(PyValueError::new_err(
            "Expected an object with dunder __arrow_c_stream__",
        ));
    }

    let capsule = ob.getattr("__arrow_c_stream__")?.call0()?.cast_into()?;
    Ok(capsule)
}

pub(crate) fn import_stream_pycapsule(capsule: &Bound<PyCapsule>) -> PyResult<PySeries> {
    validate_pycapsule_name(capsule, "arrow_array_stream")?;

    // # Safety
    // capsule holds a valid C ArrowArrayStream pointer, as defined by the Arrow PyCapsule
    // Interface
    let mut stream = unsafe {
        // Takes ownership of the pointed to ArrowArrayStream
        // This acts to move the data out of the capsule pointer, setting the release callback to NULL
        #[allow(deprecated)]
        let stream_ptr = Box::new(std::ptr::replace(
            capsule.pointer() as _,
            ArrowArrayStream::empty(),
        ));
        ArrowArrayStreamReader::try_new(stream_ptr)
            .map_err(|err| PyValueError::new_err(err.to_string()))?
    };

    let mut produced_arrays: Vec<Box<dyn Array>> = vec![];
    while let Some(array) = unsafe { stream.next() } {
        produced_arrays.push(array.map_err(PyPolarsErr::from)?);
    }

    // Series::try_from fails for an empty vec of chunks
    let s = if produced_arrays.is_empty() {
        let polars_dt = DataType::from_arrow_field(stream.field());
        Series::new_empty(stream.field().name.clone(), &polars_dt)
    } else {
        Series::try_from((stream.field(), produced_arrays)).map_err(PyPolarsErr::from)?
    };
    Ok(PySeries::new(s))
}
#[pymethods]
impl PySeries {
    #[classmethod]
    pub fn from_arrow_c_array(_cls: &Bound<PyType>, ob: &Bound<'_, PyAny>) -> PyResult<Self> {
        let (schema_capsule, array_capsule) = call_arrow_c_array(ob)?;
        let (field, array) = import_array_pycapsules(&schema_capsule, &array_capsule)?;
        let s = Series::try_from((&field, array)).unwrap();
        Ok(PySeries::new(s))
    }

    #[classmethod]
    pub fn from_arrow_c_stream(_cls: &Bound<PyType>, ob: &Bound<'_, PyAny>) -> PyResult<Self> {
        let capsule = call_arrow_c_stream(ob)?;
        import_stream_pycapsule(&capsule)
    }

    #[classmethod]
    /// Import a series via polars-ffi
    /// Takes ownership of the [`SeriesExport`] at [`location`]
    /// # Safety
    /// [`location`] should be the address of an allocated and initialized [`SeriesExport`]
    pub unsafe fn _import(_cls: &Bound<PyType>, location: usize) -> PyResult<Self> {
        let location = location as *mut SeriesExport;

        // # Safety
        // `location` should be valid for reading
        let series = unsafe {
            let export = location.read();
            polars_ffi::version_0::import_series(export).map_err(PyPolarsErr::from)?
        };
        Ok(PySeries::from(series))
    }

    #[staticmethod]
    pub fn _import_decimal_from_iceberg_binary_repr(
        bytes_list: &Bound<PyAny>, // list[bytes | None]
        precision: usize,
        scale: usize,
    ) -> PyResult<Self> {
        // From iceberg spec:
        // * Decimal(P, S): Stores unscaled value as two’s-complement
        //   big-endian binary, using the minimum number of bytes for the
        //   value.
        let max_abs_decimal_value = 10_i128.pow(u32::try_from(precision).unwrap()) - 1;

        let out: Vec<i128> = bytes_list
            .try_iter()?
            .map(|bytes| {
                let be_bytes: Option<PyBackedBytes> = bytes?.extract()?;

                let mut le_bytes: [u8; 16] = [0; _];

                if let Some(be_bytes) = be_bytes.as_deref() {
                    if be_bytes.len() > le_bytes.len() {
                        return Err(PyValueError::new_err(format!(
                            "iceberg binary data for decimal exceeded 16 bytes: {}",
                            be_bytes.len()
                        )));
                    }

                    for (i, byte) in be_bytes.iter().rev().enumerate() {
                        le_bytes[i] = *byte;
                    }
                }

                let value = i128::from_le_bytes(le_bytes);

                if value.abs() > max_abs_decimal_value {
                    return Err(PyValueError::new_err(format!(
                        "iceberg decoded value for decimal exceeded precision: \
                        value: {value}, precision: {precision}",
                    )));
                }

                Ok(value)
            })
            .collect::<PyResult<_>>()?;

        Ok(PySeries::from(unsafe {
            Series::from_chunks_and_dtype_unchecked(
                PlSmallStr::EMPTY,
                vec![PrimitiveArray::<i128>::from_vec(out).boxed()],
                &DataType::Decimal(precision, scale),
            )
        }))
    }
}