fix: Make sum on strings error in group_by context (#25456)

Co-authored-by: Claude <[email protected]>

Author: wtn

crates/polars-plan/src/plans/aexpr/schema.rs

@@ -148,6 +148,26 @@ impl AExpr {
                     Sum(expr) => {
                         let mut field = ctx.arena.get(*expr).to_field_impl(ctx)?;
                         let dt = match field.dtype() {
+                            String | Binary | BinaryOffset | List(_) => {
+                                polars_bail!(
+                                    InvalidOperation: "`sum` operation not supported for dtype `{}`",
+                                    field.dtype()
+                                )
+                            },
+                            #[cfg(feature = "dtype-array")]
+                            Array(_, _) => {
+                                polars_bail!(
+                                    InvalidOperation: "`sum` operation not supported for dtype `{}`",
+                                    field.dtype()
+                                )
+                            },
+                            #[cfg(feature = "dtype-struct")]
+                            Struct(_) => {
+                                polars_bail!(
+                                    InvalidOperation: "`sum` operation not supported for dtype `{}`",
+                                    field.dtype()
+                                )
+                            },
                             Boolean => Some(IDX_DTYPE),
                             UInt8 | Int8 | Int16 | UInt16 => Some(Int64),
                             _ => None,

py-polars/tests/unit/operations/namespaces/list/test_eval.py

@@ -209,6 +209,11 @@ def test_list_eval_in_group_by_schema(ldf: pl.LazyFrame, expr: pl.Expr) -> None:
         pl.col("a").first().list.len()
     ).collect().to_series()[0] == 0
 
+    # skip sum on struct types
+    dtype = ldf.collect_schema()["a"]
+    assert isinstance(dtype, pl.List)
+    skip = skip or ("sum" in str(expr) and isinstance(dtype.inner, pl.Struct))
+
     for q in [q_select, q_group_by, q_over]:
         if not skip:
             assert q.collect_schema() == q.collect().schema

py-polars/tests/unit/operations/test_group_by.py

@@ -13,7 +13,10 @@
 import polars as pl
 import polars.selectors as cs
 from polars import Expr
-from polars.exceptions import ColumnNotFoundError
+from polars.exceptions import (
+    ColumnNotFoundError,
+    InvalidOperationError,
+)
 from polars.meta import get_index_type
 from polars.testing import assert_frame_equal, assert_series_equal
 from polars.testing.parametric import column, dataframes, series
@@ -2909,3 +2912,11 @@ def test_agg_first_last_non_null_25405() -> None:
         }
     )
     assert_frame_equal(result.collect(), expected)
+
+
+def test_group_by_sum_on_strings_should_error_24659() -> None:
+    with pytest.raises(
+        InvalidOperationError,
+        match=r"`sum`.*operation not supported for dtype.*str",
+    ):
+        pl.DataFrame({"str": ["a", "b"]}).group_by(1).agg(pl.col.str.sum())

py-polars/tests/unit/streaming/test_streaming_group_by.py

@@ -117,7 +117,6 @@ def test_streaming_group_by_types() -> None:
                     pl.col("person_name").first().alias("str_first"),
                     pl.col("person_name").last().alias("str_last"),
                     pl.col("person_name").mean().alias("str_mean"),
-                    pl.col("person_name").sum().alias("str_sum"),
                     pl.col("bool").first().alias("bool_first"),
                     pl.col("bool").last().alias("bool_first"),
                 ]