Remove pattern matching support on non-string fields to improve performance of other queries

Author: timgraham

django_mongodb_backend/features.py

@@ -68,6 +68,10 @@ class DatabaseFeatures(GISFeatures, BaseDatabaseFeatures):
         "model_fields.test_jsonfield.TestQuerying.test_icontains",
         # Unexpected alias_refcount in alias_map.
         "queries.tests.Queries1Tests.test_order_by_tables",
+        # Regex lookup doesn't work on non-string fields.
+        "lookup.tests.LookupTests.test_regex_non_string",
+        # Querying ObjectID with string doesn't work.
+        "lookup.tests.LookupTests.test_lookup_int_as_str",
         # The $sum aggregation returns 0 instead of None for null.
         "aggregation.test_filter_argument.FilteredAggregateTests.test_plain_annotate",
         "aggregation.tests.AggregateTestCase.test_aggregation_default_passed_another_aggregate",

django_mongodb_backend/query_utils.py

@@ -50,4 +50,4 @@ def process_rhs(node, compiler, connection):
 def regex_match(field, regex_vals, insensitive=False):
     regex = {"$concat": regex_vals} if isinstance(regex_vals, tuple) else regex_vals
     options = "i" if insensitive else ""
-    return {"$regexMatch": {"input": {"$toString": field}, "regex": regex, "options": options}}
+    return {"$regexMatch": {"input": field, "regex": regex, "options": options}}

docs/source/releases/5.2.x.rst

@@ -2,6 +2,20 @@
 Django MongoDB Backend 5.2.x
 ============================
 
+5.2.0 beta 3
+============
+
+*Unreleased*
+
+Backwards incompatible changes
+------------------------------
+
+- Pattern matching lookups (``iexact``, ``startswith``, ``istartswith``,
+  ``endswith``, ``iendswith``, ``contains``, ``icontains``, ``regex``,
+  and ``iregex``) no longer support non-string fields. These lookups previously
+  cast their input using ``$toString`` but this caused some queries to perform
+  poorly because MongoDB couldn't use indexes when running the query.
+
 5.2.0 beta 2
 ============
 

docs/source/topics/known-issues.rst

@@ -50,6 +50,11 @@ Querying
   - You can study the skipped tests in ``DatabaseFeatures.django_test_skips``
     for more details on known issues.
 
+- Pattern matching lookups (:lookup:`iexact`, :lookup:`startswith`,
+  :lookup:`istartswith`, :lookup:`endswith`, :lookup:`iendswith`,
+  :lookup:`contains`, :lookup:`icontains`, :lookup:`regex`,
+  and :lookup:`iregex`) don't support non-string fields.
+
 Database functions
 ==================
 

tests/lookup_/models.py

@@ -1,6 +1,13 @@
 from django.db import models
 
 
+class Book(models.Model):
+    title = models.CharField(max_length=10)
+
+    def __str__(self):
+        return self.title
+
+
 class Number(models.Model):
     num = models.IntegerField(blank=True, null=True)
 

tests/lookup_/tests.py

@@ -1,6 +1,6 @@
 from django.test import TestCase
 
-from .models import Number
+from .models import Book, Number
 
 
 class NumericLookupTests(TestCase):
@@ -15,3 +15,18 @@ def test_lt(self):
 
     def test_lte(self):
         self.assertQuerySetEqual(Number.objects.filter(num__lte=3), self.objs[:4])
+
+
+class RegexTests(TestCase):
+    def test_mql(self):
+        # $regexMatch must not cast the input to string, otherwise MongoDB
+        # can't use the field's indexes.
+        with self.assertNumQueries(1) as ctx:
+            list(Book.objects.filter(title__regex="Moby Dick"))
+        query = ctx.captured_queries[0]["sql"]
+        self.assertEqual(
+            query,
+            "db.lookup__book.aggregate(["
+            "{'$match': {'$expr': {'$regexMatch': {'input': '$title', "
+            "'regex': 'Moby Dick', 'options': ''}}}}])",
+        )