fix(resolver): do not copy sort when flattening a join or append tran…

…sform

closes #4063 #4947

Signed-off-by: Luka Peschke <mail@lukapeschke.com>

prqlc/prqlc/src/semantic/resolver/flatten.rs

@@ -131,12 +131,31 @@ impl PlFold for Flattener {
                     kind => (self.fold_expr(*t.input)?, fold_transform_kind(self, kind)?),
                 };
 
+                // In case we're appending or joining another pipeline, we do not want to apply the
+                // sub-pipeline's sort, as it may result in column lookup errors. Without this, we
+                // would try to join on `album_id` in the outer pipeline of the following query, but
+                // the column does not exist
+                //
+                // from artists
+                // join side:left (
+                //   from albums
+                //   sort {`album_id`}
+                //   derive {`album_name` = `name`}
+                //   select {`artist_id`, `album_name`}
+                // ) (this.id == that.artist_id)
+                let sort = if matches!(kind, TransformKind::Join { .. } | TransformKind::Append(_))
+                {
+                    vec![]
+                } else {
+                    self.sort.clone()
+                };
+
                 ExprKind::TransformCall(TransformCall {
                     input: Box::new(input),
                     kind: Box::new(kind),
                     partition: self.partition.clone(),
                     frame: self.window.clone(),
-                    sort: self.sort.clone(),
+                    sort,
                 })
             }
             kind => self.fold_expr_kind(kind)?,

prqlc/prqlc/tests/integration/sql.rs

@@ -885,6 +885,89 @@ fn test_intersect_07() {
     );
 }
 
+#[test]
+fn test_sort_in_nested_join() {
+    assert_snapshot!(compile(r#"
+    from albums
+    join side:left (
+        from artists
+        sort {-`artist-id`}
+        take 10
+    ) (this.artist_id == that.artist_id) | take 10
+    "#).unwrap(),
+        @r#"
+    WITH table_0 AS (
+      SELECT
+        *
+      FROM
+        artists
+      ORDER BY
+        "artist-id" DESC
+      LIMIT
+        10
+    )
+    SELECT
+      albums.*,
+      table_0.*
+    FROM
+      albums
+      LEFT JOIN table_0 ON albums.artist_id = table_0.artist_id
+    LIMIT
+      10
+    "#
+    );
+}
+
+#[test]
+fn test_sort_in_nested_append() {
+    assert_snapshot!(compile(r#"
+    from `albums`
+    select { `album_id`, `title` }
+    sort {+`album_id`}
+    take 2
+    append (
+        from `albums`
+        select { `album_id`, `title` }
+        sort {-`album_id`}
+        take 2
+    )
+    "#).unwrap(),
+        @r#"
+        WITH table_0 AS (
+          SELECT
+            album_id,
+            title
+          FROM
+            albums
+          ORDER BY
+            album_id DESC
+          LIMIT
+            2
+        )
+        SELECT
+          *
+        FROM
+          (
+            SELECT
+              album_id,
+              title
+            FROM
+              albums
+            ORDER BY
+              album_id
+            LIMIT
+              2
+          ) AS table_1
+        UNION
+        ALL
+        SELECT
+          *
+        FROM
+          table_0
+    "#
+    );
+}
+
 #[test]
 fn test_rn_ids_are_unique() {
     // this is wrong, output will have duplicate y_id and x_id