Skip to main content

polars_python/
utils.rs

1use std::panic::AssertUnwindSafe;
2
3use polars::frame::DataFrame;
4use polars::series::IntoSeries;
5use polars_error::PolarsResult;
6use polars_error::signals::{KeyboardInterrupt, catch_keyboard_interrupt};
7use pyo3::exceptions::PyKeyboardInterrupt;
8use pyo3::marker::Ungil;
9use pyo3::{PyErr, PyResult, Python};
10
11use crate::dataframe::PyDataFrame;
12use crate::error::PyPolarsErr;
13use crate::series::PySeries;
14use crate::timeout::{cancel_polars_timeout, schedule_polars_timeout};
15
16/// Calls method on downcasted ChunkedArray for all possible publicly exposed Polars dtypes.
17#[macro_export]
18macro_rules! apply_all_polars_dtypes {
19    ($self:expr, $method:ident, $($args:expr),*) => {
20        match $self.dtype() {
21            DataType::Boolean => $self.bool().unwrap().$method($($args),*),
22            DataType::UInt8 => $self.u8().unwrap().$method($($args),*),
23            DataType::UInt16 => $self.u16().unwrap().$method($($args),*),
24            DataType::UInt32 => $self.u32().unwrap().$method($($args),*),
25            DataType::UInt64 => $self.u64().unwrap().$method($($args),*),
26            DataType::UInt128 => $self.u128().unwrap().$method($($args),*),
27            DataType::Int8 => $self.i8().unwrap().$method($($args),*),
28            DataType::Int16 => $self.i16().unwrap().$method($($args),*),
29            DataType::Int32 => $self.i32().unwrap().$method($($args),*),
30            DataType::Int64 => $self.i64().unwrap().$method($($args),*),
31            DataType::Int128 => $self.i128().unwrap().$method($($args),*),
32            DataType::Float16 => $self.f16().unwrap().$method($($args),*),
33            DataType::Float32 => $self.f32().unwrap().$method($($args),*),
34            DataType::Float64 => $self.f64().unwrap().$method($($args),*),
35            DataType::String => $self.str().unwrap().$method($($args),*),
36            DataType::Binary => $self.binary().unwrap().$method($($args),*),
37            DataType::Decimal(_, _) => $self.decimal().unwrap().$method($($args),*),
38
39            DataType::Date => $self.date().unwrap().$method($($args),*),
40            DataType::Datetime(_, _) => $self.datetime().unwrap().$method($($args),*),
41            DataType::Duration(_) => $self.duration().unwrap().$method($($args),*),
42            DataType::Time => $self.time().unwrap().$method($($args),*),
43
44            DataType::List(_) => $self.list().unwrap().$method($($args),*),
45            DataType::Struct(_) => $self.struct_().unwrap().$method($($args),*),
46            DataType::Array(_, _) => $self.array().unwrap().$method($($args),*),
47
48            dt @ (DataType::Categorical(_, _) | DataType::Enum(_, _)) => match dt.cat_physical().unwrap() {
49                CategoricalPhysical::U8 => $self.cat8().unwrap().$method($($args),*),
50                CategoricalPhysical::U16 => $self.cat16().unwrap().$method($($args),*),
51                CategoricalPhysical::U32 => $self.cat32().unwrap().$method($($args),*),
52            },
53
54            #[cfg(feature = "object")]
55            DataType::Object(_) => {
56                $self
57                .as_any()
58                .downcast_ref::<ObjectChunked<ObjectValue>>()
59                .unwrap()
60                .$method($($args),*)
61            },
62            DataType::Extension(_, _) => $self.ext().unwrap().$method($($args),*),
63
64            DataType::Null => $self.null().unwrap().$method($($args),*),
65
66            dt @ (DataType::BinaryOffset | DataType::Unknown(_)) => panic!("dtype {:?} not supported", dt)
67        }
68    }
69}
70
71/// Boilerplate for `|e| PyPolarsErr::from(e).into()`
72#[allow(unused)]
73pub(crate) fn to_py_err<E: Into<PyPolarsErr>>(e: E) -> PyErr {
74    e.into().into()
75}
76
77pub trait EnterPolarsExt {
78    /// Whenever you have a block of code in the public Python API that
79    /// (potentially) takes a long time, wrap it in enter_polars. This will
80    /// ensure we release the GIL and catch KeyboardInterrupts.
81    ///
82    /// This not only can increase performance and usability, it can avoid
83    /// deadlocks on the GIL for Python UDFs.
84    fn enter_polars<T, E, F>(self, f: F) -> PyResult<T>
85    where
86        F: Ungil + Send + FnOnce() -> Result<T, E>,
87        T: Ungil + Send,
88        E: Ungil + Send + Into<PyPolarsErr>;
89
90    /// Same as enter_polars, but wraps the result in PyResult::Ok, useful
91    /// shorthand for infallible functions.
92    #[inline(always)]
93    fn enter_polars_ok<T, F>(self, f: F) -> PyResult<T>
94    where
95        Self: Sized,
96        F: Ungil + Send + FnOnce() -> T,
97        T: Ungil + Send,
98    {
99        self.enter_polars(move || PyResult::Ok(f()))
100    }
101
102    /// Same as enter_polars, but expects a PolarsResult<DataFrame> as return
103    /// which is converted to a PyDataFrame.
104    #[inline(always)]
105    fn enter_polars_df<F>(self, f: F) -> PyResult<PyDataFrame>
106    where
107        Self: Sized,
108        F: Ungil + Send + FnOnce() -> PolarsResult<DataFrame>,
109    {
110        self.enter_polars(f).map(PyDataFrame::new)
111    }
112
113    /// Same as enter_polars, but expects a PolarsResult<S> as return which
114    /// is converted to a PySeries through S: IntoSeries.
115    #[inline(always)]
116    fn enter_polars_series<T, F>(self, f: F) -> PyResult<PySeries>
117    where
118        Self: Sized,
119        T: Ungil + Send + IntoSeries,
120        F: Ungil + Send + FnOnce() -> PolarsResult<T>,
121    {
122        self.enter_polars(f).map(|s| PySeries::new(s.into_series()))
123    }
124}
125
126impl EnterPolarsExt for Python<'_> {
127    fn enter_polars<T, E, F>(self, f: F) -> PyResult<T>
128    where
129        F: Ungil + Send + FnOnce() -> Result<T, E>,
130        T: Ungil + Send,
131        E: Ungil + Send + Into<PyPolarsErr>,
132    {
133        let timeout = schedule_polars_timeout();
134        let ret = self.detach(|| catch_keyboard_interrupt(AssertUnwindSafe(f)));
135        cancel_polars_timeout(timeout);
136        match ret {
137            Ok(Ok(ret)) => Ok(ret),
138            Ok(Err(err)) => Err(PyErr::from(err.into())),
139            Err(KeyboardInterrupt) => Err(PyKeyboardInterrupt::new_err("")),
140        }
141    }
142}