polars_python/lazyframe/visitor/

nodes.rs

#[cfg(feature = "iejoin")]
use polars::prelude::JoinTypeOptionsIR;
use polars::prelude::deletion::DeletionFilesList;
use polars::prelude::python_dsl::PythonScanSource;
use polars::prelude::{ColumnMapping, PredicateFileSkip};
use polars_core::prelude::IdxSize;
use polars_io::cloud::CloudOptions;
use polars_ops::prelude::JoinType;
use polars_plan::plans::{HintIR, IR};
use polars_plan::prelude::{FileScanIR, FunctionIR, PythonPredicate, UnifiedScanArgs};
use pyo3::IntoPyObjectExt;
use pyo3::exceptions::{PyNotImplementedError, PyValueError};
use pyo3::prelude::*;
use pyo3::types::{PyDict, PyList, PyString};

use super::expr_nodes::PyGroupbyOptions;
use crate::PyDataFrame;
use crate::lazyframe::visit::PyExprIR;

fn scan_type_to_pyobject(
    py: Python<'_>,
    scan_type: &FileScanIR,
    cloud_options: &Option<CloudOptions>,
) -> PyResult<Py<PyAny>> {
    match scan_type {
        #[cfg(feature = "csv")]
        FileScanIR::Csv { options } => {
            let options = serde_json::to_string(options)
                .map_err(|err| PyValueError::new_err(format!("{err:?}")))?;
            let cloud_options = serde_json::to_string(cloud_options)
                .map_err(|err| PyValueError::new_err(format!("{err:?}")))?;
            Ok(("csv", options, cloud_options).into_py_any(py)?)
        },
        #[cfg(feature = "parquet")]
        FileScanIR::Parquet { options, .. } => {
            let options = serde_json::to_string(options)
                .map_err(|err| PyValueError::new_err(format!("{err:?}")))?;
            let cloud_options = serde_json::to_string(cloud_options)
                .map_err(|err| PyValueError::new_err(format!("{err:?}")))?;
            Ok(("parquet", options, cloud_options).into_py_any(py)?)
        },
        #[cfg(feature = "ipc")]
        FileScanIR::Ipc { .. } => Err(PyNotImplementedError::new_err("ipc scan")),
        #[cfg(feature = "json")]
        FileScanIR::NDJson { options, .. } => {
            let options = serde_json::to_string(options)
                .map_err(|err| PyValueError::new_err(format!("{err:?}")))?;
            Ok(("ndjson", options).into_py_any(py)?)
        },
        #[cfg(feature = "scan_lines")]
        FileScanIR::Lines { name } => Ok(("lines", name.as_str()).into_py_any(py)?),
        FileScanIR::ExpandedPaths { name } => {
            Ok(("expanded-paths", name.as_str()).into_py_any(py)?)
        },
        FileScanIR::PythonDataset { .. } => {
            Err(PyNotImplementedError::new_err("python dataset scan"))
        },
        FileScanIR::Anonymous { .. } => Err(PyNotImplementedError::new_err("anonymous scan")),
    }
}

#[pyclass(frozen)]
/// Scan a table with an optional predicate from a python function
pub struct PythonScan {
    #[pyo3(get)]
    options: Py<PyAny>,
}

#[pyclass(frozen)]
/// Slice the table
pub struct Slice {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    offset: i64,
    #[pyo3(get)]
    len: IdxSize,
}

#[pyclass(frozen)]
/// Filter the table with a boolean expression
pub struct Filter {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    predicate: PyExprIR,
}

#[pyclass(frozen, skip_from_py_object)]
#[derive(Clone)]
pub struct PyFileOptions {
    inner: UnifiedScanArgs,
}

#[pymethods]
impl PyFileOptions {
    #[getter]
    fn n_rows(&self) -> Option<(i64, IdxSize)> {
        self.inner
            .pre_slice
            .clone()
            .map(|slice| slice.to_signed_offset_len())
    }
    #[getter]
    fn with_columns(&self) -> Option<Vec<&str>> {
        self.inner
            .projection
            .as_ref()?
            .iter()
            .map(|x| x.as_str())
            .collect::<Vec<_>>()
            .into()
    }
    #[getter]
    fn cache(&self, _py: Python<'_>) -> bool {
        self.inner.cache
    }
    #[getter]
    fn row_index(&self) -> Option<(&str, IdxSize)> {
        self.inner
            .row_index
            .as_ref()
            .map(|n| (n.name.as_str(), n.offset))
    }
    #[getter]
    fn rechunk(&self, _py: Python<'_>) -> bool {
        self.inner.rechunk
    }
    #[getter]
    fn hive_options(&self, _py: Python<'_>) -> PyResult<Py<PyAny>> {
        Err(PyNotImplementedError::new_err("hive options"))
    }
    #[getter]
    fn include_file_paths(&self, _py: Python<'_>) -> Option<&str> {
        self.inner.include_file_paths.as_deref()
    }

    /// One of:
    /// * None
    /// * ("iceberg-position-delete", dict[int, list[str]])
    #[getter]
    fn deletion_files(&self, py: Python<'_>) -> PyResult<Py<PyAny>> {
        Ok(match &self.inner.deletion_files {
            None => py.None().into_any(),
            Some(DeletionFilesList::IcebergPositionDelete(paths)) => {
                let out = PyDict::new(py);
                for (k, v) in paths.iter() {
                    out.set_item(*k, v.as_ref())?;
                }
                ("iceberg-position-delete", out)
                    .into_pyobject(py)?
                    .into_any()
                    .unbind()
            },
            Some(DeletionFilesList::Delta(provider)) => {
                ("delta-deletion-vector", provider.callback().0.clone_ref(py))
                    .into_pyobject(py)?
                    .into_any()
                    .unbind()
            },
        })
    }

    /// One of:
    /// * None
    /// * ("iceberg-column-mapping", <unimplemented>)
    #[getter]
    fn column_mapping(&self, py: Python<'_>) -> PyResult<Py<PyAny>> {
        Ok(match &self.inner.column_mapping {
            None => py.None().into_any(),

            Some(ColumnMapping::Iceberg { .. }) => unimplemented!(),
        })
    }
}

#[pyclass(frozen)]
/// Scan a table from file
pub struct Scan {
    #[pyo3(get)]
    paths: Py<PyAny>,
    #[pyo3(get)]
    file_info: Py<PyAny>,
    #[pyo3(get)]
    predicate: Option<PyExprIR>,
    #[pyo3(get)]
    file_options: PyFileOptions,
    #[pyo3(get)]
    scan_type: Py<PyAny>,
}

#[pyclass(frozen)]
/// Scan a table from an existing dataframe
pub struct DataFrameScan {
    #[pyo3(get)]
    df: PyDataFrame,
    #[pyo3(get)]
    projection: Py<PyAny>,
    #[pyo3(get)]
    selection: Option<PyExprIR>,
}

#[pyclass(frozen)]
/// Project out columns from a table
pub struct SimpleProjection {
    #[pyo3(get)]
    input: usize,
}

#[pyclass(frozen)]
/// Column selection
pub struct Select {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    expr: Vec<PyExprIR>,
    #[pyo3(get)]
    should_broadcast: bool,
}

#[pyclass(frozen)]
/// Sort the table
pub struct Sort {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    by_column: Vec<PyExprIR>,
    #[pyo3(get)]
    sort_options: (bool, Vec<bool>, Vec<bool>),
    #[pyo3(get)]
    slice: Option<(i64, usize)>,
}

#[pyclass(frozen)]
/// Cache the input at this point in the LP
pub struct Cache {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    id_: u128,
}

#[pyclass(frozen)]
/// Groupby aggregation
pub struct GroupBy {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    keys: Vec<PyExprIR>,
    #[pyo3(get)]
    aggs: Vec<PyExprIR>,
    #[pyo3(get)]
    apply: (),
    #[pyo3(get)]
    maintain_order: bool,
    #[pyo3(get)]
    options: Py<PyAny>,
}

#[pyclass(frozen)]
/// Join operation
pub struct Join {
    #[pyo3(get)]
    input_left: usize,
    #[pyo3(get)]
    input_right: usize,
    #[pyo3(get)]
    left_on: Vec<PyExprIR>,
    #[pyo3(get)]
    right_on: Vec<PyExprIR>,
    #[pyo3(get)]
    options: Py<PyAny>,
}

#[pyclass(frozen)]
/// Join operation
pub struct Gather {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    idxs: usize,
    #[pyo3(get)]
    null_on_oob: bool,
}

#[pyclass(frozen)]
/// Merge sorted operation
pub struct MergeSorted {
    #[pyo3(get)]
    input_left: usize,
    #[pyo3(get)]
    input_right: usize,
    #[pyo3(get)]
    key: String,
    #[pyo3(get)]
    maintain_order: bool,
}

#[pyclass(frozen)]
/// Adding columns to the table without a Join
pub struct HStack {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    exprs: Vec<PyExprIR>,
    #[pyo3(get)]
    should_broadcast: bool,
}

#[pyclass(frozen)]
/// Like Select, but all operations produce a single row.
pub struct Reduce {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    exprs: Vec<PyExprIR>,
}

#[pyclass(frozen)]
/// Remove duplicates from the table
pub struct Distinct {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    options: Py<PyAny>,
}
#[pyclass(frozen)]
/// A (User Defined) Function
pub struct MapFunction {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    function: Py<PyAny>,
}
#[pyclass(frozen)]
pub struct Union {
    #[pyo3(get)]
    inputs: Vec<usize>,
    #[pyo3(get)]
    slice: Option<(i64, usize)>,
    #[pyo3(get)]
    rows: (Option<usize>, usize),
    #[pyo3(get)]
    maintain_order: bool,
}
#[pyclass(frozen)]
/// Horizontal concatenation of multiple plans
pub struct HConcat {
    #[pyo3(get)]
    inputs: Vec<usize>,
    #[pyo3(get)]
    options: Py<PyAny>,
}
#[pyclass(frozen)]
/// This allows expressions to access other tables
pub struct ExtContext {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    contexts: Vec<usize>,
}

#[pyclass(frozen)]
pub struct Sink {
    #[pyo3(get)]
    input: usize,
    #[pyo3(get)]
    payload: Py<PyAny>,
}

pub(crate) fn into_py(py: Python<'_>, plan: &IR) -> PyResult<Py<PyAny>> {
    match plan {
        IR::PythonScan { options } => {
            let python_src = match options.python_source {
                PythonScanSource::Pyarrow => "pyarrow",
                PythonScanSource::Cuda => "cuda",
                PythonScanSource::IOPlugin => "io_plugin",
            };

            PythonScan {
                options: (
                    options
                        .scan_fn
                        .as_ref()
                        .map_or_else(|| py.None(), |s| s.0.clone_ref(py)),
                    options.with_columns.as_ref().map_or_else(
                        || Ok(py.None()),
                        |cols| {
                            cols.iter()
                                .map(|x| x.as_str())
                                .collect::<Vec<_>>()
                                .into_py_any(py)
                        },
                    )?,
                    python_src,
                    match &options.predicate {
                        PythonPredicate::None => py.None(),
                        PythonPredicate::PyArrow {
                            predicate,
                            has_residual,
                        } => {
                            ("pyarrow", predicate, "has_residual", has_residual).into_py_any(py)?
                        },
                        PythonPredicate::Polars(e) => ("polars", e.node().0).into_py_any(py)?,
                    },
                    options
                        .n_rows
                        .map_or_else(|| Ok(py.None()), |s| s.into_py_any(py))?,
                )
                    .into_py_any(py)?,
            }
            .into_py_any(py)
        },
        IR::Slice { input, offset, len } => Slice {
            input: input.0,
            offset: *offset,
            len: *len,
        }
        .into_py_any(py),
        IR::Filter { input, predicate } => Filter {
            input: input.0,
            predicate: predicate.into(),
        }
        .into_py_any(py),
        IR::Scan {
            hive_parts: Some(_),
            ..
        } => Err(PyNotImplementedError::new_err(
            "scan with hive partitioning",
        )),
        IR::Scan {
            sources,
            file_info: _,
            hive_parts: _,
            predicate,
            predicate_file_skip_applied,
            output_schema: _,
            scan_type,
            unified_scan_args,
        } => {
            Scan {
                paths: {
                    let paths = sources
                        .into_paths()
                        .ok_or_else(|| PyNotImplementedError::new_err("scan with BytesIO"))?;

                    let out = PyList::new(py, [] as [(); 0])?;

                    // Manual conversion to preserve `uri://...` - converting Rust `Path` to `PosixPath`
                    // will corrupt to `uri:/...`
                    for path in paths.iter() {
                        out.append(path.as_str())?;
                    }

                    out.into_py_any(py)?
                },
                // TODO: file info
                file_info: py.None(),
                predicate: predicate
                    .as_ref()
                    .filter(|_| {
                        !matches!(
                            predicate_file_skip_applied,
                            Some(PredicateFileSkip {
                                no_residual_predicate: true,
                                original_len: _,
                            })
                        )
                    })
                    .map(|e| e.into()),
                file_options: PyFileOptions {
                    inner: (**unified_scan_args).clone(),
                },
                scan_type: scan_type_to_pyobject(py, scan_type, &unified_scan_args.cloud_options)?,
            }
        }
        .into_py_any(py),
        IR::DataFrameScan {
            df,
            schema: _,
            output_schema,
        } => DataFrameScan {
            df: PyDataFrame::new((**df).clone()),
            projection: output_schema.as_ref().map_or_else(
                || Ok(py.None()),
                |s| {
                    s.iter_names()
                        .map(|s| s.as_str())
                        .collect::<Vec<_>>()
                        .into_py_any(py)
                },
            )?,
            selection: None,
        }
        .into_py_any(py),
        IR::SimpleProjection { input, columns: _ } => {
            SimpleProjection { input: input.0 }.into_py_any(py)
        },
        IR::Select {
            input,
            expr,
            schema: _,
            options,
        } => Select {
            expr: expr.iter().map(|e| e.into()).collect(),
            input: input.0,
            should_broadcast: options.should_broadcast,
        }
        .into_py_any(py),
        IR::Sort {
            input,
            by_column,
            slice,
            sort_options,
        } => Sort {
            input: input.0,
            by_column: by_column.iter().map(|e| e.into()).collect(),
            sort_options: (
                sort_options.maintain_order,
                sort_options.nulls_last.clone(),
                sort_options.descending.clone(),
            ),
            slice: slice.as_ref().map(|t| (t.0, t.1)),
        }
        .into_py_any(py),
        IR::Cache { input, id } => Cache {
            input: input.0,
            id_: id.as_u128(),
        }
        .into_py_any(py),
        IR::GroupBy {
            input,
            keys,
            aggs,
            schema: _,
            apply,
            maintain_order,
            options,
        } => GroupBy {
            input: input.0,
            keys: keys.iter().map(|e| e.into()).collect(),
            aggs: aggs.iter().map(|e| e.into()).collect(),
            apply: apply.as_ref().map_or(Ok(()), |_| {
                Err(PyNotImplementedError::new_err(format!(
                    "apply inside GroupBy {plan:?}"
                )))
            })?,
            maintain_order: *maintain_order,
            options: PyGroupbyOptions::new(options.as_ref().clone()).into_py_any(py)?,
        }
        .into_py_any(py),
        IR::Join {
            input_left,
            input_right,
            schema: _,
            left_on,
            right_on,
            options,
        } => Join {
            input_left: input_left.0,
            input_right: input_right.0,
            left_on: left_on.iter().map(|e| e.into()).collect(),
            right_on: right_on.iter().map(|e| e.into()).collect(),
            options: {
                let how = &options.args.how;
                let name = Into::<&str>::into(how).into_pyobject(py)?;
                (
                    match how {
                        #[cfg(feature = "asof_join")]
                        JoinType::AsOf(_) => {
                            return Err(PyNotImplementedError::new_err("asof join"));
                        },
                        #[cfg(feature = "iejoin")]
                        JoinType::IEJoin => {
                            let Some(JoinTypeOptionsIR::IEJoin(ie_options)) = &options.options
                            else {
                                unreachable!()
                            };
                            (
                                name,
                                crate::Wrap(ie_options.operator1).into_py_any(py)?,
                                ie_options.operator2.as_ref().map_or_else(
                                    || Ok(py.None()),
                                    |op| crate::Wrap(*op).into_py_any(py),
                                )?,
                            )
                                .into_py_any(py)?
                        },
                        // This is a cross join fused with a predicate. Shown in the IR::explain as
                        // NESTED LOOP JOIN
                        JoinType::Cross if options.options.is_some() => {
                            return Err(PyNotImplementedError::new_err("nested loop join"));
                        },
                        _ => name.into_any().unbind(),
                    },
                    options.args.nulls_equal,
                    options.args.slice,
                    options.args.suffix().as_str(),
                    options.args.coalesce.coalesce(how),
                    Into::<&str>::into(options.args.maintain_order),
                )
                    .into_py_any(py)?
            },
        }
        .into_py_any(py),
        IR::Gather {
            input,
            idxs,
            null_on_oob,
        } => Gather {
            input: input.0,
            idxs: idxs.0,
            null_on_oob: *null_on_oob,
        }
        .into_py_any(py),
        IR::HStack {
            input,
            exprs,
            schema: _,
            options,
        } => HStack {
            input: input.0,
            exprs: exprs.iter().map(|e| e.into()).collect(),
            should_broadcast: options.should_broadcast,
        }
        .into_py_any(py),
        IR::Distinct { input, options } => Distinct {
            input: input.0,
            options: (
                Into::<&str>::into(options.keep_strategy),
                options.subset.as_ref().map_or_else(
                    || Ok(py.None()),
                    |f| {
                        f.iter()
                            .map(|s| s.as_ref())
                            .collect::<Vec<&str>>()
                            .into_py_any(py)
                    },
                )?,
                options.maintain_order,
                options.slice,
            )
                .into_py_any(py)?,
        }
        .into_py_any(py),
        IR::MapFunction { input, function } => MapFunction {
            input: input.0,
            function: match function {
                FunctionIR::OpaquePython(_) => {
                    return Err(PyNotImplementedError::new_err("opaque python mapfunction"));
                },
                FunctionIR::Opaque {
                    function: _,
                    schema: _,
                    predicate_pd: _,
                    projection_pd: _,
                    streamable: _,
                    fmt_str: _,
                } => return Err(PyNotImplementedError::new_err("opaque rust mapfunction")),
                FunctionIR::Unnest { columns, separator } => (
                    "unnest",
                    columns.iter().map(|s| s.to_string()).collect::<Vec<_>>(),
                    separator.as_ref().map(|s| s.to_string()),
                )
                    .into_py_any(py)?,
                FunctionIR::Rechunk => ("rechunk",).into_py_any(py)?,
                FunctionIR::Explode {
                    columns,
                    options,
                    schema: _,
                } => (
                    "explode",
                    columns.iter().map(|s| s.to_string()).collect::<Vec<_>>(),
                    options.empty_as_null,
                    options.keep_nulls,
                )
                    .into_py_any(py)?,
                #[cfg(feature = "pivot")]
                FunctionIR::Unpivot { args, schema: _ } => (
                    "unpivot",
                    args.index.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
                    args.on.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
                    args.variable_name.as_str().into_py_any(py)?,
                    args.value_name.as_str().into_py_any(py)?,
                )
                    .into_py_any(py)?,
                FunctionIR::RowIndex {
                    name,
                    schema: _,
                    offset,
                } => ("row_index", name.to_string(), offset.unwrap_or(0)).into_py_any(py)?,
                FunctionIR::FastCount {
                    sources,
                    scan_type,
                    alias,
                    ..
                } => {
                    let sources = sources
                        .into_paths()
                        .ok_or_else(|| {
                            PyNotImplementedError::new_err("FastCount with BytesIO sources")
                        })?
                        .iter()
                        .map(|p| p.as_str())
                        .collect::<Vec<_>>()
                        .into_py_any(py)?;

                    // FastCount does not support cloud options.
                    let cloud_options = None;

                    let scan_type = scan_type_to_pyobject(py, scan_type, &cloud_options)?;

                    let alias = alias
                        .as_ref()
                        .map(|a| a.as_str())
                        .map_or_else(|| Ok(py.None()), |s| s.into_py_any(py))?;

                    ("fast_count", sources, scan_type, alias).into_py_any(py)?
                },
                FunctionIR::Hint(hint) => match hint {
                    HintIR::Sorted(sorted_vec) => {
                        let sorted_info: Vec<_> = sorted_vec
                            .iter()
                            .map(|s| (s.column.as_str(), s.descending, s.nulls_last))
                            .collect();
                        ("hint_sorted", sorted_info).into_py_any(py)?
                    },
                },
            },
        }
        .into_py_any(py),
        IR::Union { inputs, options } => Union {
            inputs: inputs.iter().map(|n| n.0).collect(),
            // Remaining options are implementation detail, not logical
            slice: options.slice,
            rows: options.rows,
            maintain_order: options.maintain_order,
        }
        .into_py_any(py),
        IR::HConcat {
            inputs,
            schema: _,
            options,
        } => HConcat {
            inputs: inputs.iter().map(|n| n.0).collect(),
            options: (
                options.parallel,
                options.strict,
                options.broadcast_unit_length,
            )
                .into_py_any(py)?,
        }
        .into_py_any(py),
        IR::ExtContext {
            input,
            contexts,
            schema: _,
        } => ExtContext {
            input: input.0,
            contexts: contexts.iter().map(|n| n.0).collect(),
        }
        .into_py_any(py),
        IR::Sink { input, payload } => Sink {
            input: input.0,
            payload: PyString::new(
                py,
                &serde_json::to_string(payload)
                    .map_err(|err| PyValueError::new_err(format!("{err:?}")))?,
            )
            .into(),
        }
        .into_py_any(py),
        IR::SinkMultiple { .. } => Err(PyNotImplementedError::new_err(
            "Not expecting to see a SinkMultiple node",
        )),
        #[cfg(feature = "merge_sorted")]
        IR::MergeSorted {
            input_left,
            input_right,
            key,
            maintain_order,
        } => MergeSorted {
            input_left: input_left.0,
            input_right: input_right.0,
            key: key.to_string(),
            maintain_order: *maintain_order,
        }
        .into_py_any(py),
        IR::UnoptimizedDispatch { .. } => Err(PyNotImplementedError::new_err(
            "Not expecting to see a UnoptimizedDispatch node",
        )),
        IR::Invalid => Err(PyNotImplementedError::new_err("Invalid")),
    }
}