perf: Speed up list operations that use amortized_iter()

crates/polars-core/src/chunked_array/list/iterator.rs

@@ -2,6 +2,7 @@ use std::marker::PhantomData;
 use std::ptr::NonNull;
 use std::rc::Rc;
 
+use crate::chunked_array::flags::StatisticsFlags;
 use crate::prelude::*;
 use crate::series::amortized_iter::{unstable_series_container_and_ptr, AmortSeries, ArrayBox};
 
@@ -81,10 +82,13 @@ impl<I: Iterator<Item = Option<ArrayBox>>> Iterator for AmortizedListIter<'_, I>
                     // update the inner state
                     unsafe { *self.inner.as_mut() = array_ref };
 
+                    // As an optimization, we try to minimize how many calls to
+                    // _get_inner_mut() we do.
+                    let series_mut_inner = series_mut._get_inner_mut();
                     // last iteration could have set the sorted flag (e.g. in compute_len)
-                    series_mut.clear_flags();
+                    series_mut_inner._set_flags(StatisticsFlags::empty());
                     // make sure that the length is correct
-                    series_mut._get_inner_mut().compute_len();
+                    series_mut_inner.compute_len();
                 }
 
                 // SAFETY:

crates/polars-core/src/lib.rs

@@ -2,6 +2,7 @@
 #![cfg_attr(feature = "simd", feature(portable_simd))]
 #![allow(ambiguous_glob_reexports)]
 #![cfg_attr(feature = "nightly", allow(clippy::non_canonical_partial_ord_impl))] // remove once stable
+
 extern crate core;
 
 #[macro_use]

py-polars/tests/benchmark/test_list_op.py

@@ -0,0 +1,23 @@
+from typing import Callable
+
+import pytest
+
+import polars as pl
+
+pytestmark = pytest.mark.benchmark()
+
+
+@pytest.fixture(scope="module")
+def lists_and_values() -> pl.DataFrame:
+    return pl.DataFrame(
+        {"lists": [[0, 2, 1, 3, 5]] * 10_000, "values": [3, 1, 4, 5, 0] * 2000}
+    )
+
+
+def test_list_contains(
+    benchmark: Callable[[Callable[[], None]], object], lists_and_values: pl.DataFrame
+) -> None:
+    def go() -> None:
+        lists_and_values.select(pl.col("lists").list.contains(pl.col("values")))
+
+    benchmark(go)