remove unneeded expensive assert (#3069)

polars/polars-core/src/series/unstable.rs

@@ -33,10 +33,9 @@ impl<'a> UnstableSeries<'a> {
     }
 
     /// Creates a new `[UnsafeSeries]`
-    /// # Panics
-    /// panics if `inner_chunk` is not from `Series`.
-    pub fn new_with_chunk(series: &'a Series, inner_chunk: &ArrayRef) -> Self {
-        assert_eq!(series.chunks()[0].as_ref(), inner_chunk.as_ref());
+    /// # Safety
+    /// Inner chunks must be from `Series` otherwise the dtype may be incorrect and lead to UB.
+    pub(crate) unsafe fn new_with_chunk(series: &'a Series, inner_chunk: &ArrayRef) -> Self {
         UnstableSeries {
             container: series,
             inner: NonNull::new(inner_chunk as *const ArrayRef as *mut ArrayRef).unwrap(),

py-polars/src/apply/series.rs

@@ -5,7 +5,7 @@ use crate::Wrap;
 use polars::chunked_array::builder::get_list_builder;
 use polars::prelude::*;
 use pyo3::prelude::*;
-use pyo3::types::{PyBool, PyCFunction, PyDict, PyFloat, PyInt, PyList, PyString, PyTuple};
+use pyo3::types::{PyBool, PyCFunction, PyDict, PyFloat, PyList, PyString, PyTuple};
 
 /// Find the output type and dispatch to that implementation.
 fn infer_and_finish<'a, A: ApplyLambda<'a>>(

py-polars/tests/test_errors.py

@@ -1,3 +1,5 @@
+import io
+
 import numpy as np
 import pytest
 
@@ -35,3 +37,10 @@ def test_error_on_invalid_by_in_asof_join() -> None:
     df2 = df1.with_column(pl.col("a").cast(pl.Categorical))
     with pytest.raises(pl.ComputeError):
         df1.join_asof(df2, on="b", by=["a", "c"])
+
+
+def test_not_found_error() -> None:
+    csv = "a,b,c\n2,1,1"
+    df = pl.read_csv(io.StringIO(csv))
+    with pytest.raises(pl.NotFoundError):
+        df.select("d")

py-polars/tests/test_window.py

@@ -106,3 +106,21 @@ def test_arange_no_rows() -> None:
     df = pl.DataFrame(dict(x=[]))
     out = df.with_column(pl.arange(0, pl.count()).over("x"))  # type: ignore
     assert out.frame_equal(pl.DataFrame({"x": [], "literal": []}))
+
+
+def test_no_panic_on_nan_3067() -> None:
+    df = pl.DataFrame(
+        {
+            "group": ["a", "a", "a", "b", "b", "b"],
+            "total": [1.0, 2, 3, 4, 5, np.NaN],
+        }
+    )
+
+    df.select([pl.col("total").shift().over("group")])["total"].to_list() == [
+        None,
+        1.0,
+        2.0,
+        None,
+        4.0,
+        5.0,
+    ]