fix: keep insertion order when MetaFieldGroupingRanker sort values have mixed types (#11942)

Co-authored-by: bogdankostic <bogdankostic@web.de>

Author: otiscuilei

haystack/components/rankers/meta_field_grouping_ranker.py

@@ -4,9 +4,11 @@
 
 from collections import defaultdict
 
-from haystack import Document, component
+from haystack import Document, component, logging
 from haystack.utils.misc import _deduplicate_documents
 
+logger = logging.getLogger(__name__)
+
 
 @component
 class MetaFieldGroupingRanker:
@@ -117,9 +119,21 @@ def run(self, documents: list[Document]) -> dict[str, list[Document]]:
             for docs in subgroups.values():
                 if sort_field:
                     # Sort by the field value, placing documents with a missing value last.
-                    # The (is_missing, value) tuple ensures that only actual field values are
-                    # compared, making the sort work for numbers, strings, and other types.
-                    docs.sort(key=lambda d: (d.meta.get(sort_field) is None, d.meta.get(sort_field)))
+                    # The (is_missing, value) tuple keeps documents with a missing value out of the
+                    # value comparison, but two present values of mutually non-comparable types
+                    # (e.g. an int and a str) would still raise a TypeError. In that case we keep the
+                    # group's insertion order instead of crashing, mirroring MetaFieldRanker.
+                    try:
+                        docs.sort(key=lambda d: (d.meta.get(sort_field) is None, d.meta.get(sort_field)))
+                    except TypeError as error:
+                        logger.warning(
+                            "Tried to sort Documents with IDs {document_ids}, but got TypeError with the "
+                            "message: {error}\nKeeping the original order of the Documents in this group "
+                            "since sorting by '{sort_field}' is not possible.",
+                            document_ids=",".join([doc.id for doc in docs]),
+                            error=error,
+                            sort_field=sort_field,
+                        )
                 ordered_docs.extend(docs)
 
         ordered_docs.extend(no_group_docs)

releasenotes/notes/meta-field-grouping-ranker-mixed-type-sort-161c30a3ec07c5ac.yaml

@@ -0,0 +1,7 @@
+---
+fixes:
+  - |
+    Fixed ``MetaFieldGroupingRanker`` crashing with an uncaught ``TypeError`` when a group contained
+    ``sort_docs_by`` values of mutually non-comparable types (for example an ``int`` and a ``str``). The
+    sort is now wrapped in a ``try``/``except``, so in that case the group's original insertion order is
+    kept and a warning is logged, mirroring the behavior of ``MetaFieldRanker``.

test/components/rankers/test_meta_field_grouping_ranker.py

@@ -185,6 +185,23 @@ def test_run_metadata_with_different_data_types(self) -> None:
         assert result["documents"][1].meta["group"] == 42
         assert result["documents"][2].meta["group"] is True
 
+    def test_run_sort_docs_by_mixed_uncomparable_types(self) -> None:
+        """
+        Test that the ranker does not crash when a group's sort_docs_by values have mutually
+        non-comparable present types (e.g. int and str), keeping the group's insertion order instead.
+        """
+        docs_with_mixed_sort_values = [
+            Document(content="int value", meta={"group": "g1", "split_id": 3}),
+            Document(content="str value", meta={"group": "g1", "split_id": "10"}),
+        ]
+        sample_ranker = MetaFieldGroupingRanker(group_by="group", sort_docs_by="split_id")
+        result = sample_ranker.run(documents=docs_with_mixed_sort_values)
+        assert "documents" in result
+        assert len(result["documents"]) == 2
+        # Insertion order is preserved because the values cannot be compared.
+        assert result["documents"][0].content == "int value"
+        assert result["documents"][1].content == "str value"
+
     def test_run_deduplicates_documents(self) -> None:
         """
         Test that duplicate documents are removed before grouping.