fix(queryset): make distinct().count() ignore join-duplicated rows

CHANGELOG.rst

@@ -26,6 +26,7 @@ Fixed
 - ``QuerySet.count()`` now matches the limited query result for the LIMIT/OFFSET edge cases: it returns ``0`` (instead of a negative number) when ``offset()`` exceeds the total row count, and ``0`` (instead of the total) for ``limit(0)``. (#2208)
 - Field declarations on models now resolve to their concrete type (e.g. ``CharField[str]``) in Pyright/Pylance instead of ``Field[Unknown]``; the ``Field.__new__`` type-check stub now returns ``Self``. (#2216)
 - ``TransactionContext`` now returns a ``TransactionalDBClient`` instead of a raw database connection. This change gives the correct inferred type for the transaction context. (#2232)
+- ``QuerySet.distinct().count()`` no longer counts rows duplicated by a join (e.g. filtering on a m2m relation); it now counts distinct primary keys and matches the number of rows the query returns. (#2178)
 
 
 1.1.7

tests/test_queryset.py

@@ -9,6 +9,7 @@
     MinRelation,
     Node,
     Reporter,
+    Team,
     Tournament,
     Tree,
 )
@@ -227,6 +228,18 @@ async def test_distinct(db, intfields_data):
     ]
 
 
+@pytest.mark.asyncio
+async def test_distinct_count_with_m2m_join(db):
+    tournament = await Tournament.create(name="Tournament")
+    event = await Event.create(name="Event", tournament=tournament)
+    for name in ("Team A", "Team B", "Team C"):
+        await event.participants.add(await Team.create(name=name))
+
+    queryset = Event.filter(participants__name__in=["Team A", "Team B", "Team C"]).distinct()
+    assert len(await queryset) == 1
+    assert await queryset.count() == 1
+
+
 @pytest.mark.asyncio
 async def test_limit_offset_values_list(db, intfields_data):
     # Test limit/offset/ordering values_list

tortoise/queryset.py

@@ -9,8 +9,9 @@
 from pypika_tortoise import JoinType, Order, Table
 from pypika_tortoise.analytics import Count
 from pypika_tortoise.functions import Cast
+from pypika_tortoise.functions import Count as DistinctCount
 from pypika_tortoise.queries import QueryBuilder, _SetOperation
-from pypika_tortoise.terms import Case, Field, Star, Term, ValueWrapper
+from pypika_tortoise.terms import Case, Field, Function, Star, Term, ValueWrapper
 
 from tortoise.backends.base.client import BaseDBAsyncClient, Capabilities
 from tortoise.exceptions import (
@@ -806,6 +807,7 @@ def count(self) -> CountQuery:
             offset=self._offset,
             force_indexes=self._force_indexes,
             use_indexes=self._use_indexes,
+            distinct=self._distinct,
         )
 
     def exists(self) -> ExistsQuery:
@@ -1520,6 +1522,7 @@ class CountQuery(AwaitableQuery):
         "_offset",
         "_force_indexes",
         "_use_indexes",
+        "_distinct",
     )
 
     def __init__(
@@ -1533,6 +1536,7 @@ def __init__(
         offset: int | None,
         force_indexes: set[str],
         use_indexes: set[str],
+        distinct: bool = False,
     ) -> None:
         super().__init__(model)
         self._q_objects = q_objects
@@ -1543,13 +1547,23 @@ def __init__(
         self._db = db
         self._force_indexes = force_indexes
         self._use_indexes = use_indexes
+        self._distinct = distinct
 
     def _make_query(self) -> None:
         self.query = copy(self.model._meta.basequery)
         self.resolve_filters()
-        count_term = Count(Star())
-        if self.query._groupbys:
-            count_term = count_term.over()
+        count_term: Function
+        if self._distinct and not self.query._groupbys:
+            # ``DISTINCT`` on the outer SELECT does not apply to ``COUNT(*)``,
+            # so rows duplicated by a join would still be counted. Count the
+            # distinct primary keys instead.
+            count_term = DistinctCount(
+                self.model._meta.basetable[self.model._meta.db_pk_column]
+            ).distinct()
+        else:
+            count_term = Count(Star())
+            if self.query._groupbys:
+                count_term = count_term.over()
 
         # remove annotations
         self.query._selects = []