feat: Raise when joining on the same keys twice

crates/polars-ops/src/frame/join/args.rs

@@ -185,7 +185,7 @@ impl JoinValidation {
         }
     }
 
-    pub(super) fn is_valid_join(&self, join_type: &JoinType) -> PolarsResult<()> {
+    pub fn is_valid_join(&self, join_type: &JoinType) -> PolarsResult<()> {
         if !self.needs_checks() {
             return Ok(());
         }

crates/polars-ops/src/frame/join/mod.rs

@@ -115,9 +115,8 @@ pub trait DataFrameJoinOps: IntoDf {
         _verbose: bool,
     ) -> PolarsResult<DataFrame> {
         let left_df = self.to_df();
-        args.validation.is_valid_join(&args.how)?;
-
         let should_coalesce = args.coalesce.coalesce(&args.how);
+        assert_eq!(selected_left.len(), selected_right.len());
 
         #[cfg(feature = "cross_join")]
         if let JoinType::Cross = args.how {
@@ -163,16 +162,6 @@ pub trait DataFrameJoinOps: IntoDf {
             }
         }
 
-        polars_ensure!(
-            selected_left.len() == selected_right.len(),
-            ComputeError:
-                format!(
-                    "the number of columns given as join key (left: {}, right:{}) should be equal",
-                    selected_left.len(),
-                    selected_right.len()
-                )
-        );
-
         if let Some((l, r)) = selected_left
             .iter()
             .zip(&selected_right)

crates/polars-plan/src/logical_plan/conversion/dsl_to_ir.rs

@@ -324,6 +324,23 @@ pub fn to_alp_impl(
                 }
             }
 
+            let mut joined_on = PlHashSet::new();
+            for (l, r) in left_on.iter().zip(right_on.iter()) {
+                polars_ensure!(joined_on.insert((l, r)), InvalidOperation: "joins on same keys twice; already joined on {} and {}", l, r)
+            }
+            drop(joined_on);
+            options.args.validation.is_valid_join(&options.args.how)?;
+
+            polars_ensure!(
+                left_on.len() == right_on.len(),
+                ComputeError:
+                    format!(
+                        "the number of columns given as join key (left: {}, right:{}) should be equal",
+                        left_on.len(),
+                        right_on.len()
+                    )
+            );
+
             let input_left = to_alp_impl(owned(input_left), expr_arena, lp_arena, convert)
                 .map_err(|e| e.context(failed_input!(join left)))?;
             let input_right = to_alp_impl(owned(input_right), expr_arena, lp_arena, convert)

crates/polars/tests/it/core/joins.rs

@@ -420,10 +420,6 @@ fn test_join_err() -> PolarsResult<()> {
     assert!(df1
         .join(&df2, vec!["a", "b"], vec!["a", "b"], JoinType::Left.into())
         .is_err());
-    // length of join keys don't match error
-    assert!(df1
-        .join(&df2, vec!["a"], vec!["a", "b"], JoinType::Left.into())
-        .is_err());
     Ok(())
 }
 

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

@@ -1002,3 +1002,10 @@ def test_join_empties(how: str) -> None:
 
     df = df1.join(df2, on="col2", how=how)
     assert df.height == 0
+
+
+def test_join_raise_on_redundant_keys() -> None:
+    left = pl.DataFrame({"a": [1, 2, 3], "b": [3, 4, 5], "c": [5, 6, 7]})
+    right = pl.DataFrame({"a": [2, 3, 4], "c": [4, 5, 6]})
+    with pytest.raises(pl.InvalidOperationError, match="already joined on"):
+        left.join(right, on=["a", "a"], how="outer_coalesce")