feat: raise informative error message when grouping by expressions (#1677)

* raise informative error message when grouping by expressions

* add hint

Author: MarcoGorelli

narwhals/dataframe.py

@@ -2209,9 +2209,18 @@ def group_by(
             │ c   ┆ 3   ┆ 1   │
             └─────┴─────┴─────┘
         """
+        from narwhals.expr import Expr
         from narwhals.group_by import GroupBy
+        from narwhals.series import Series
 
-        return GroupBy(self, *flatten(keys), drop_null_keys=drop_null_keys)
+        flat_keys = flatten(keys)
+        if any(isinstance(x, (Expr, Series)) for x in flat_keys):
+            msg = (
+                "`group_by` with expression or Series keys is not (yet?) supported.\n\n"
+                "Hint: instead of `df.group_by(nw.col('a'))`, use `df.group_by('a')`."
+            )
+            raise NotImplementedError(msg)
+        return GroupBy(self, *flat_keys, drop_null_keys=drop_null_keys)
 
     def sort(
         self,
@@ -4444,9 +4453,18 @@ def group_by(
             │ c   ┆ 3   ┆ 1   │
             └─────┴─────┴─────┘
         """
+        from narwhals.expr import Expr
         from narwhals.group_by import LazyGroupBy
+        from narwhals.series import Series
 
-        return LazyGroupBy(self, *flatten(keys), drop_null_keys=drop_null_keys)
+        flat_keys = flatten(keys)
+        if any(isinstance(x, (Expr, Series)) for x in flat_keys):
+            msg = (
+                "`group_by` with expression or Series keys is not (yet?) supported.\n\n"
+                "Hint: instead of `df.group_by(nw.col('a'))`, use `df.group_by('a')`."
+            )
+            raise NotImplementedError(msg)
+        return LazyGroupBy(self, *flat_keys, drop_null_keys=drop_null_keys)
 
     def sort(
         self,

tests/group_by_test.py

@@ -453,3 +453,9 @@ def test_all_kind_of_aggs(
         "i": [3, 2],
     }
     assert_equal_data(result, expected)
+
+
+def test_group_by_expr(constructor: Constructor) -> None:
+    df = nw.from_native(constructor({"a": [1, 1, 3], "b": [4, 5, 6]}))
+    with pytest.raises(NotImplementedError, match=r"not \(yet\?\) supported"):
+        df.group_by(nw.col("a")).agg(nw.col("b").mean())  # type: ignore[arg-type]