[Mcsimps2] search first in pipeline

docs/changelog.rst

@@ -23,6 +23,7 @@ Changes in 1.0.0
 - Bug Fix - Fix Document.compare_indexes() not working correctly for text indexes on multiple fields #2612
 - BREAKING CHANGE: wrap _document_registry (normally not used by end users) with _DocumentRegistry which acts as a singleton to access the registry
 - Log a warning in case users creates multiple Document classes with the same name as it can lead to unexpected behavior #1778
+- Fix use of $search, $searchMeta, or $vectorSearch in aggregate #2878
 - BugFix - Fix use of $geoNear or $collStats in aggregate #2493
 - BREAKING CHANGE: Further to the deprecation warning, remove ability to use an unpacked list to `Queryset.aggregate(*pipeline)`, a plain list must be provided instead `Queryset.aggregate(pipeline)`, as it's closer to pymongo interface
 - BREAKING CHANGE: Further to the deprecation warning, remove `full_response` from `QuerySet.modify` as it wasn't supported with Pymongo 3+

mongoengine/queryset/base.py

@@ -1382,13 +1382,20 @@ def aggregate(self, pipeline, **kwargs):
         if self._skip is not None:
             initial_pipeline.append({"$skip": self._skip})
 
-        # geoNear and collStats must be the first stages in the pipeline if present
+        # Some aggregation stages must precede MongoEngine's implicit stages.
         first_step = []
         new_user_pipeline = []
         for step_step in pipeline:
-            if "$geoNear" in step_step:
-                first_step.append(step_step)
-            elif "$collStats" in step_step:
+            if any(
+                operator in step_step
+                for operator in (
+                    "$geoNear",
+                    "$collStats",
+                    "$search",
+                    "$searchMeta",
+                    "$vectorSearch",
+                )
+            ):
                 first_step.append(step_step)
             else:
                 new_user_pipeline.append(step_step)

tests/queryset/test_queryset_aggregation.py

@@ -1,3 +1,5 @@
+from unittest.mock import patch
+
 import pytest
 from pymongo.read_preferences import ReadPreference
 
@@ -373,3 +375,21 @@ class SomeDoc(Document):
         res = list(SomeDoc.objects.aggregate(pipeline))
         assert len(res) == 1
         assert res[0]["count"] == 2
+
+    def test_aggregate__search_stage__precedes_implicit_match(self):
+        class SearchableDoc(Document):
+            last_name = StringField()
+
+        for search_stage in ("$search", "$searchMeta", "$vectorSearch"):
+            with self.subTest(search_stage=search_stage):
+                search_step = {search_stage: {}}
+
+                # End-to-end execution requires MongoDB Search (mongot), so mock
+                # the collection and verify the pipeline passed to PyMongo.
+                with patch.object(SearchableDoc, "_collection") as collection:
+                    SearchableDoc.objects(last_name="bar").aggregate([search_step])
+
+                assert collection.aggregate.call_args.args[0] == [
+                    search_step,
+                    {"$match": {"last_name": "bar"}},
+                ]