fix(rust, python): disallow alias in inline join expressions

polars/polars-lazy/polars-plan/src/logical_plan/builder.rs

@@ -602,6 +602,19 @@ impl LogicalPlanBuilder {
         right_on: Vec<Expr>,
         options: JoinOptions,
     ) -> Self {
+        for e in left_on.iter().chain(right_on.iter()) {
+            if has_expr(e, |e| matches!(e, Expr::Alias(_, _))) {
+                return LogicalPlan::Error {
+                    input: Box::new(self.0),
+                    err: PolarsError::ComputeError(
+                        "'alias' is not allowed in a join key. Use 'with_columns' first.".into(),
+                    )
+                    .into(),
+                }
+                .into();
+            }
+        }
+
         let schema_left = try_delayed!(self.0.schema(), &self.0, into);
         let schema_right = try_delayed!(other.schema(), &self.0, into);
 

polars/tests/it/io/json.rs

@@ -129,6 +129,7 @@ fn read_ndjson_with_trailing_newline() {
     assert!(expected.frame_equal(&df));
 }
 #[test]
+#[cfg(feature = "dtype-struct")]
 fn test_read_ndjson_iss_5875() {
     let jsonlines = r#"
     {"struct": {"int_inner": [1, 2, 3], "float_inner": 5.0, "str_inner": ["a", "b", "c"]}}
@@ -158,6 +159,7 @@ fn test_read_ndjson_iss_5875() {
 }
 
 #[test]
+#[cfg(feature = "dtype-struct")]
 fn test_read_ndjson_iss_5875_part2() {
     let jsonlines = r#"
     {"struct": {"int_list_inner": [4, 5, 6]}}
@@ -188,6 +190,7 @@ fn test_read_ndjson_iss_5875_part2() {
     assert_eq!(schema, df.unwrap().schema());
 }
 #[test]
+#[cfg(feature = "dtype-struct")]
 fn test_read_ndjson_iss_5875_part3() {
     let jsonlines = r#"
     {"key1":"value1", "key2": "value2", "key3": {"k1": 2, "k3": "value5", "k10": 5}}

polars/tests/it/joins.rs

@@ -12,17 +12,17 @@ fn join_nans_outer() -> PolarsResult<()> {
         .lazy();
     let a1 = df1
         .clone()
-        .groupby(vec![col("w").alias("w"), col("t").alias("t")])
+        .groupby(vec![col("w").alias("w"), col("t")])
         .agg(vec![col("c").sum().alias("c_sum")]);
     let a2 = df1
-        .groupby(vec![col("w").alias("w"), col("t").alias("t")])
+        .groupby(vec![col("w").alias("w"), col("t")])
         .agg(vec![col("c").max().alias("c_max")]);
 
     let res = a1
         .join_builder()
         .with(a2)
-        .left_on(vec![col("w").alias("w"), col("t").alias("t")])
-        .right_on(vec![col("w").alias("w"), col("t").alias("t")])
+        .left_on(vec![col("w"), col("t")])
+        .right_on(vec![col("w"), col("t")])
         .how(JoinType::Outer)
         .finish()
         .collect()?;

polars/tests/it/lazy/projection_queries.rs

@@ -153,6 +153,7 @@ fn test_projection_5086() -> PolarsResult<()> {
 }
 
 #[test]
+#[cfg(feature = "dtype-struct")]
 fn test_unnest_pushdown() -> PolarsResult<()> {
     let df = df![
         "collection" => Series::full_null("", 1, &DataType::Int32),

py-polars/tests/unit/test_errors.py

@@ -334,3 +334,12 @@ def test_arr_eval_named_cols() -> None:
         pl.ComputeError,
     ):
         df.select(pl.col("B").arr.eval(pl.element().append(pl.col("A"))))
+
+
+def test_alias_in_join_keys() -> None:
+    df = pl.DataFrame({"A": ["a", "b"], "B": [["a", "b"], ["c", "d"]]})
+    with pytest.raises(
+        pl.ComputeError,
+        match=r"'alias' is not allowed in a join key. Use 'with_columns' first",
+    ):
+        df.join(df, on=pl.col("A").alias("foo"))

py-polars/tests/unit/test_joins.py

@@ -475,64 +475,6 @@ def test_join_chunks_alignment_4720() -> None:
     }
 
 
-def test_join_inline_alias_4694() -> None:
-    df = pl.DataFrame(
-        [
-            {"ts": datetime(2021, 2, 1, 9, 20), "a1": 1.04, "a2": 0.9},
-            {"ts": datetime(2021, 2, 1, 9, 50), "a1": 1.04, "a2": 0.9},
-            {"ts": datetime(2021, 2, 2, 10, 20), "a1": 1.04, "a2": 0.9},
-            {"ts": datetime(2021, 2, 2, 11, 20), "a1": 1.08, "a2": 0.9},
-            {"ts": datetime(2021, 2, 3, 11, 50), "a1": 1.08, "a2": 0.9},
-            {"ts": datetime(2021, 2, 3, 13, 20), "a1": 1.16, "a2": 0.8},
-            {"ts": datetime(2021, 2, 4, 13, 50), "a1": 1.18, "a2": 0.8},
-        ]
-    ).lazy()
-
-    join_against = pl.DataFrame(
-        [
-            {"d": datetime(2021, 2, 3, 0, 0), "ets": datetime(2021, 2, 4, 0, 0)},
-            {"d": datetime(2021, 2, 3, 0, 0), "ets": datetime(2021, 2, 5, 0, 0)},
-            {"d": datetime(2021, 2, 3, 0, 0), "ets": datetime(2021, 2, 6, 0, 0)},
-        ]
-    ).lazy()
-
-    # this adds "dd" column to the lhs followed by a projection
-    # the projection optimizer must realize that this column is added inline and ensure
-    # it is not dropped.
-    assert df.join(
-        join_against,
-        left_on=pl.col("ts").dt.truncate("1d").alias("dd"),
-        right_on=pl.col("d"),
-    ).select(pl.all()).collect().to_dict(False) == {
-        "ts": [
-            datetime(2021, 2, 3, 11, 50),
-            datetime(2021, 2, 3, 11, 50),
-            datetime(2021, 2, 3, 11, 50),
-            datetime(2021, 2, 3, 13, 20),
-            datetime(2021, 2, 3, 13, 20),
-            datetime(2021, 2, 3, 13, 20),
-        ],
-        "a1": [1.08, 1.08, 1.08, 1.16, 1.16, 1.16],
-        "a2": [0.9, 0.9, 0.9, 0.8, 0.8, 0.8],
-        "dd": [
-            datetime(2021, 2, 3, 0, 0),
-            datetime(2021, 2, 3, 0, 0),
-            datetime(2021, 2, 3, 0, 0),
-            datetime(2021, 2, 3, 0, 0),
-            datetime(2021, 2, 3, 0, 0),
-            datetime(2021, 2, 3, 0, 0),
-        ],
-        "ets": [
-            datetime(2021, 2, 4, 0, 0),
-            datetime(2021, 2, 5, 0, 0),
-            datetime(2021, 2, 6, 0, 0),
-            datetime(2021, 2, 4, 0, 0),
-            datetime(2021, 2, 5, 0, 0),
-            datetime(2021, 2, 6, 0, 0),
-        ],
-    }
-
-
 def test_sorted_flag_after_joins() -> None:
     np.random.seed(1)
     dfa = pl.DataFrame(