Refs #28333 -- Added partial support for filtering against window functions.

Adds support for joint predicates against window annotations through
subquery wrapping while maintaining errors for disjointed filter
attempts.

The "qualify" wording was used to refer to predicates against window
annotations as it's the name of a specialized Snowflake extension to
SQL that is to window functions what HAVING is to aggregates.

While not complete the implementation should cover most of the common
use cases for filtering against window functions without requiring
the complex subquery pushdown and predicate re-aliasing machinery to
deal with disjointed predicates against columns, aggregates, and window
functions.

A complete disjointed filtering implementation should likely be
deferred until proper QUALIFY support lands or the ORM gains a proper
subquery pushdown interface.

django/db/backends/mysql/compiler.py

@@ -28,10 +28,13 @@ class SQLDeleteCompiler(compiler.SQLDeleteCompiler, SQLCompiler):
         # the SQLDeleteCompiler's default implementation when multiple tables
         # are involved since MySQL/MariaDB will generate a more efficient query
         # plan than when using a subquery.
-        where, having = self.query.where.split_having()
+        where, having, qualify = self.query.where.split_having_qualify(
-        if self.single_alias or having:
+            must_group_by=self.query.group_by is not None
-            # DELETE FROM cannot be used when filtering against aggregates
+        )
-            # since it doesn't allow for GROUP BY and HAVING clauses.
+        if self.single_alias or having or qualify:
+            # DELETE FROM cannot be used when filtering against aggregates or
+            # window functions as it doesn't allow for GROUP BY/HAVING clauses
+            # and the subquery wrapping (necessary to emulate QUALIFY).
             return super().as_sql()
         result = [
             "DELETE %s FROM"

django/db/models/expressions.py

@@ -836,6 +836,7 @@ class ResolvedOuterRef(F):
     """
 
     contains_aggregate = False
+    contains_over_clause = False
 
     def as_sql(self, *args, **kwargs):
         raise ValueError(
@@ -1210,6 +1211,12 @@ class OrderByList(Func):
             return "", ()
         return super().as_sql(*args, **kwargs)
 
+    def get_group_by_cols(self):
+        group_by_cols = []
+        for order_by in self.get_source_expressions():
+            group_by_cols.extend(order_by.get_group_by_cols())
+        return group_by_cols
+
 
 @deconstructible(path="django.db.models.ExpressionWrapper")
 class ExpressionWrapper(SQLiteNumericMixin, Expression):
@@ -1631,7 +1638,6 @@ class Window(SQLiteNumericMixin, Expression):
     # be introduced in the query as a result is not desired.
     contains_aggregate = False
     contains_over_clause = True
-    filterable = False
 
     def __init__(
         self,
@@ -1733,7 +1739,12 @@ class Window(SQLiteNumericMixin, Expression):
         return "<%s: %s>" % (self.__class__.__name__, self)
 
     def get_group_by_cols(self, alias=None):
-        return []
+        group_by_cols = []
+        if self.partition_by:
+            group_by_cols.extend(self.partition_by.get_group_by_cols())
+        if self.order_by is not None:
+            group_by_cols.extend(self.order_by.get_group_by_cols())
+        return group_by_cols
 
 
 class WindowFrame(Expression):

django/db/models/fields/related_lookups.py

@@ -14,6 +14,7 @@ from django.utils.deprecation import RemovedInDjango50Warning
 
 class MultiColSource:
     contains_aggregate = False
+    contains_over_clause = False
 
     def __init__(self, alias, targets, sources, field):
         self.targets, self.sources, self.field, self.alias = (

django/db/models/sql/compiler.py

@@ -9,6 +9,7 @@ from django.db import DatabaseError, NotSupportedError
 from django.db.models.constants import LOOKUP_SEP
 from django.db.models.expressions import F, OrderBy, RawSQL, Ref, Value
 from django.db.models.functions import Cast, Random
+from django.db.models.lookups import Lookup
 from django.db.models.query_utils import select_related_descend
 from django.db.models.sql.constants import (
     CURSOR,
@@ -73,7 +74,9 @@ class SQLCompiler:
         """
         self.setup_query(with_col_aliases=with_col_aliases)
         order_by = self.get_order_by()
-        self.where, self.having = self.query.where.split_having()
+        self.where, self.having, self.qualify = self.query.where.split_having_qualify(
+            must_group_by=self.query.group_by is not None
+        )
         extra_select = self.get_extra_select(order_by, self.select)
         self.has_extra_select = bool(extra_select)
         group_by = self.get_group_by(self.select + extra_select, order_by)
@@ -584,6 +587,74 @@ class SQLCompiler:
             params.extend(part)
         return result, params
 
+    def get_qualify_sql(self):
+        where_parts = []
+        if self.where:
+            where_parts.append(self.where)
+        if self.having:
+            where_parts.append(self.having)
+        inner_query = self.query.clone()
+        inner_query.subquery = True
+        inner_query.where = inner_query.where.__class__(where_parts)
+        # Augment the inner query with any window function references that
+        # might have been masked via values() and alias(). If any masked
+        # aliases are added they'll be masked again to avoid fetching
+        # the data in the `if qual_aliases` branch below.
+        select = {
+            expr: alias for expr, _, alias in self.get_select(with_col_aliases=True)[0]
+        }
+        qual_aliases = set()
+        replacements = {}
+        expressions = list(self.qualify.leaves())
+        while expressions:
+            expr = expressions.pop()
+            if select_alias := (select.get(expr) or replacements.get(expr)):
+                replacements[expr] = select_alias
+            elif isinstance(expr, Lookup):
+                expressions.extend(expr.get_source_expressions())
+            else:
+                num_qual_alias = len(qual_aliases)
+                select_alias = f"qual{num_qual_alias}"
+                qual_aliases.add(select_alias)
+                inner_query.add_annotation(expr, select_alias)
+                replacements[expr] = select_alias
+        self.qualify = self.qualify.replace_expressions(
+            {expr: Ref(alias, expr) for expr, alias in replacements.items()}
+        )
+        inner_query_compiler = inner_query.get_compiler(
+            self.using, elide_empty=self.elide_empty
+        )
+        inner_sql, inner_params = inner_query_compiler.as_sql(
+            # The limits must be applied to the outer query to avoid pruning
+            # results too eagerly.
+            with_limits=False,
+            # Force unique aliasing of selected columns to avoid collisions
+            # and make rhs predicates referencing easier.
+            with_col_aliases=True,
+        )
+        qualify_sql, qualify_params = self.compile(self.qualify)
+        result = [
+            "SELECT * FROM (",
+            inner_sql,
+            ")",
+            self.connection.ops.quote_name("qualify"),
+            "WHERE",
+            qualify_sql,
+        ]
+        if qual_aliases:
+            # If some select aliases were unmasked for filtering purposes they
+            # must be masked back.
+            cols = [self.connection.ops.quote_name(alias) for alias in select.values()]
+            result = [
+                "SELECT",
+                ", ".join(cols),
+                "FROM (",
+                *result,
+                ")",
+                self.connection.ops.quote_name("qualify_mask"),
+            ]
+        return result, list(inner_params) + qualify_params
+
     def as_sql(self, with_limits=True, with_col_aliases=False):
         """
         Create the SQL for this query. Return the SQL string and list of
@@ -614,6 +685,9 @@ class SQLCompiler:
                 result, params = self.get_combinator_sql(
                     combinator, self.query.combinator_all
                 )
+            elif self.qualify:
+                result, params = self.get_qualify_sql()
+                order_by = None
             else:
                 distinct_fields, distinct_params = self.get_distinct()
                 # This must come after 'select', 'ordering', and 'distinct'

django/db/models/sql/where.py

@@ -35,48 +35,81 @@ class WhereNode(tree.Node):
     resolved = False
     conditional = True
 
-    def split_having(self, negated=False):
+    def split_having_qualify(self, negated=False, must_group_by=False):
         """
-        Return two possibly None nodes: one for those parts of self that
+        Return three possibly None nodes: one for those parts of self that
-        should be included in the WHERE clause and one for those parts of
+        should be included in the WHERE clause, one for those parts of self
-        self that must be included in the HAVING clause.
+        that must be included in the HAVING clause, and one for those parts
+        that refer to window functions.
         """
-        if not self.contains_aggregate:
+        if not self.contains_aggregate and not self.contains_over_clause:
-            return self, None
+            return self, None, None
         in_negated = negated ^ self.negated
-        # If the effective connector is OR or XOR and this node contains an
+        # Whether or not children must be connected in the same filtering
-        # aggregate, then we need to push the whole branch to HAVING clause.
+        # clause (WHERE > HAVING > QUALIFY) to maintain logical semantic.
-        may_need_split = (
+        must_remain_connected = (
             (in_negated and self.connector == AND)
             or (not in_negated and self.connector == OR)
             or self.connector == XOR
         )
-        if may_need_split and self.contains_aggregate:
+        if (
-            return None, self
+            must_remain_connected
+            and self.contains_aggregate
+            and not self.contains_over_clause
+        ):
+            # It's must cheaper to short-circuit and stash everything in the
+            # HAVING clause than split children if possible.
+            return None, self, None
         where_parts = []
         having_parts = []
+        qualify_parts = []
         for c in self.children:
-            if hasattr(c, "split_having"):
+            if hasattr(c, "split_having_qualify"):
-                where_part, having_part = c.split_having(in_negated)
+                where_part, having_part, qualify_part = c.split_having_qualify(
+                    in_negated, must_group_by
+                )
                 if where_part is not None:
                     where_parts.append(where_part)
                 if having_part is not None:
                     having_parts.append(having_part)
+                if qualify_part is not None:
+                    qualify_parts.append(qualify_part)
+            elif c.contains_over_clause:
+                qualify_parts.append(c)
             elif c.contains_aggregate:
                 having_parts.append(c)
             else:
                 where_parts.append(c)
-        having_node = (
+        if must_remain_connected and qualify_parts:
-            self.create(having_parts, self.connector, self.negated)
+            # Disjunctive heterogeneous predicates can be pushed down to
-            if having_parts
+            # qualify as long as no conditional aggregation is involved.
-            else None
+            if not where_parts or (where_parts and not must_group_by):
-        )
+                return None, None, self
+            elif where_parts:
+                # In theory this should only be enforced when dealing with
+                # where_parts containing predicates against multi-valued
+                # relationships that could affect aggregation results but this
+                # is complex to infer properly.
+                raise NotImplementedError(
+                    "Heterogeneous disjunctive predicates against window functions are "
+                    "not implemented when performing conditional aggregation."
+                )
         where_node = (
             self.create(where_parts, self.connector, self.negated)
             if where_parts
             else None
         )
-        return where_node, having_node
+        having_node = (
+            self.create(having_parts, self.connector, self.negated)
+            if having_parts
+            else None
+        )
+        qualify_node = (
+            self.create(qualify_parts, self.connector, self.negated)
+            if qualify_parts
+            else None
+        )
+        return where_node, having_node, qualify_node
 
     def as_sql(self, compiler, connection):
         """
@@ -183,6 +216,14 @@ class WhereNode(tree.Node):
         clone.relabel_aliases(change_map)
         return clone
 
+    def replace_expressions(self, replacements):
+        if replacement := replacements.get(self):
+            return replacement
+        clone = self.create(connector=self.connector, negated=self.negated)
+        for child in self.children:
+            clone.children.append(child.replace_expressions(replacements))
+        return clone
+
     @classmethod
     def _contains_aggregate(cls, obj):
         if isinstance(obj, tree.Node):
@@ -231,6 +272,10 @@ class WhereNode(tree.Node):
 
         return BooleanField()
 
+    @property
+    def _output_field_or_none(self):
+        return self.output_field
+
     def select_format(self, compiler, sql, params):
         # Wrap filters with a CASE WHEN expression if a database backend
         # (e.g. Oracle) doesn't support boolean expression in SELECT or GROUP
@@ -245,19 +290,28 @@ class WhereNode(tree.Node):
     def get_lookup(self, lookup):
         return self.output_field.get_lookup(lookup)
 
+    def leaves(self):
+        for child in self.children:
+            if isinstance(child, WhereNode):
+                yield from child.leaves()
+            else:
+                yield child
+
 
 class NothingNode:
     """A node that matches nothing."""
 
     contains_aggregate = False
+    contains_over_clause = False
 
     def as_sql(self, compiler=None, connection=None):
         raise EmptyResultSet
 
 
 class ExtraWhere:
-    # The contents are a black box - assume no aggregates are used.
+    # The contents are a black box - assume no aggregates or windows are used.
     contains_aggregate = False
+    contains_over_clause = False
 
     def __init__(self, sqls, params):
         self.sqls = sqls
@@ -269,9 +323,10 @@ class ExtraWhere:
 
 
 class SubqueryConstraint:
-    # Even if aggregates would be used in a subquery, the outer query isn't
+    # Even if aggregates or windows would be used in a subquery,
-    # interested about those.
+    # the outer query isn't interested about those.
     contains_aggregate = False
+    contains_over_clause = False
 
     def __init__(self, alias, columns, targets, query_object):
         self.alias = alias

docs/ref/models/expressions.txt

@@ -741,12 +741,6 @@ instead they are part of the selected columns.
 
 .. class:: Window(expression, partition_by=None, order_by=None, frame=None, output_field=None)
 
-    .. attribute:: filterable
-
-        Defaults to ``False``. The SQL standard disallows referencing window
-        functions in the ``WHERE`` clause and Django raises an exception when
-        constructing a ``QuerySet`` that would do that.
-
     .. attribute:: template
 
         Defaults to ``%(expression)s OVER (%(window)s)'``. If only the
@@ -819,6 +813,31 @@ to reduce repetition::
     >>>     ),
     >>> )
 
+Filtering against window functions is supported as long as lookups are not
+disjunctive (not using ``OR`` or ``XOR`` as a connector) and against a queryset
+performing aggregation.
+
+For example, a query that relies on aggregation and has an ``OR``-ed filter
+against a window function and a field is not supported. Applying combined
+predicates post-aggregation could cause rows that would normally be excluded
+from groups to be included::
+
+    >>> qs = Movie.objects.annotate(
+    >>>     category_rank=Window(
+    >>>         Rank(), partition_by='category', order_by='-rating'
+    >>>     ),
+    >>>     scenes_count=Count('actors'),
+    >>> ).filter(
+    >>>     Q(category_rank__lte=3) | Q(title__contains='Batman')
+    >>> )
+    >>> list(qs)
+    NotImplementedError: Heterogeneous disjunctive predicates against window functions
+    are not implemented when performing conditional aggregation.
+
+.. versionchanged:: 4.2
+
+    Support for filtering against window functions was added.
+
 Among Django's built-in database backends, MySQL 8.0.2+, PostgreSQL, and Oracle
 support window expressions. Support for different window expression features
 varies among the different databases. For example, the options in

docs/releases/4.2.txt

@@ -189,7 +189,9 @@ Migrations
 Models
 ~~~~~~
 
-* ...
+* ``QuerySet`` now extensively supports filtering against
+  :ref:`window-functions` with the exception of disjunctive filter lookups
+  against window functions when performing aggregation.
 
 Requests and Responses
 ~~~~~~~~~~~~~~~~~~~~~~

tests/expressions_window/models.py

@@ -17,6 +17,13 @@ class Employee(models.Model):
     bonus = models.DecimalField(decimal_places=2, max_digits=15, null=True)
 
 
+class PastEmployeeDepartment(models.Model):
+    employee = models.ForeignKey(
+        Employee, related_name="past_departments", on_delete=models.CASCADE
+    )
+    department = models.CharField(max_length=40, blank=False, null=False)
+
+
 class Detail(models.Model):
     value = models.JSONField()
 

tests/expressions_window/tests.py

@@ -6,10 +6,9 @@ from django.core.exceptions import FieldError
 from django.db import NotSupportedError, connection
 from django.db.models import (
     Avg,
-    BooleanField,
     Case,
+    Count,
     F,
-    Func,
     IntegerField,
     Max,
     Min,
@@ -41,15 +40,17 @@ from django.db.models.functions import (
     RowNumber,
     Upper,
 )
+from django.db.models.lookups import Exact
 from django.test import SimpleTestCase, TestCase, skipUnlessDBFeature
 
-from .models import Detail, Employee
+from .models import Classification, Detail, Employee, PastEmployeeDepartment
 
 
 @skipUnlessDBFeature("supports_over_clause")
 class WindowFunctionTests(TestCase):
     @classmethod
     def setUpTestData(cls):
+        classification = Classification.objects.create()
         Employee.objects.bulk_create(
             [
                 Employee(
@@ -59,6 +60,7 @@ class WindowFunctionTests(TestCase):
                     hire_date=e[3],
                     age=e[4],
                     bonus=Decimal(e[1]) / 400,
+                    classification=classification,
                 )
                 for e in [
                     ("Jones", 45000, "Accounting", datetime.datetime(2005, 11, 1), 20),
@@ -82,6 +84,13 @@ class WindowFunctionTests(TestCase):
                 ]
             ]
         )
+        employees = list(Employee.objects.order_by("pk"))
+        PastEmployeeDepartment.objects.bulk_create(
+            [
+                PastEmployeeDepartment(employee=employees[6], department="Sales"),
+                PastEmployeeDepartment(employee=employees[10], department="IT"),
+            ]
+        )
 
     def test_dense_rank(self):
         tests = [
@@ -902,6 +911,263 @@ class WindowFunctionTests(TestCase):
         )
         self.assertEqual(qs.count(), 12)
 
+    def test_filter(self):
+        qs = Employee.objects.annotate(
+            department_salary_rank=Window(
+                Rank(), partition_by="department", order_by="-salary"
+            ),
+            department_avg_age_diff=(
+                Window(Avg("age"), partition_by="department") - F("age")
+            ),
+        ).order_by("department", "name")
+        # Direct window reference.
+        self.assertQuerysetEqual(
+            qs.filter(department_salary_rank=1),
+            ["Adams", "Wilkinson", "Miller", "Johnson", "Smith"],
+            lambda employee: employee.name,
+        )
+        # Through a combined expression containing a window.
+        self.assertQuerysetEqual(
+            qs.filter(department_avg_age_diff__gt=0),
+            ["Jenson", "Jones", "Williams", "Miller", "Smith"],
+            lambda employee: employee.name,
+        )
+        # Intersection of multiple windows.
+        self.assertQuerysetEqual(
+            qs.filter(department_salary_rank=1, department_avg_age_diff__gt=0),
+            ["Miller"],
+            lambda employee: employee.name,
+        )
+        # Union of multiple windows.
+        self.assertQuerysetEqual(
+            qs.filter(Q(department_salary_rank=1) | Q(department_avg_age_diff__gt=0)),
+            [
+                "Adams",
+                "Jenson",
+                "Jones",
+                "Williams",
+                "Wilkinson",
+                "Miller",
+                "Johnson",
+                "Smith",
+                "Smith",
+            ],
+            lambda employee: employee.name,
+        )
+
+    def test_filter_conditional_annotation(self):
+        qs = (
+            Employee.objects.annotate(
+                rank=Window(Rank(), partition_by="department", order_by="-salary"),
+                case_first_rank=Case(
+                    When(rank=1, then=True),
+                    default=False,
+                ),
+                q_first_rank=Q(rank=1),
+            )
+            .order_by("name")
+            .values_list("name", flat=True)
+        )
+        for annotation in ["case_first_rank", "q_first_rank"]:
+            with self.subTest(annotation=annotation):
+                self.assertSequenceEqual(
+                    qs.filter(**{annotation: True}),
+                    ["Adams", "Johnson", "Miller", "Smith", "Wilkinson"],
+                )
+
+    def test_filter_conditional_expression(self):
+        qs = (
+            Employee.objects.filter(
+                Exact(Window(Rank(), partition_by="department", order_by="-salary"), 1)
+            )
+            .order_by("name")
+            .values_list("name", flat=True)
+        )
+        self.assertSequenceEqual(
+            qs, ["Adams", "Johnson", "Miller", "Smith", "Wilkinson"]
+        )
+
+    def test_filter_column_ref_rhs(self):
+        qs = (
+            Employee.objects.annotate(
+                max_dept_salary=Window(Max("salary"), partition_by="department")
+            )
+            .filter(max_dept_salary=F("salary"))
+            .order_by("name")
+            .values_list("name", flat=True)
+        )
+        self.assertSequenceEqual(
+            qs, ["Adams", "Johnson", "Miller", "Smith", "Wilkinson"]
+        )
+
+    def test_filter_values(self):
+        qs = (
+            Employee.objects.annotate(
+                department_salary_rank=Window(
+                    Rank(), partition_by="department", order_by="-salary"
+                ),
+            )
+            .order_by("department", "name")
+            .values_list(Upper("name"), flat=True)
+        )
+        self.assertSequenceEqual(
+            qs.filter(department_salary_rank=1),
+            ["ADAMS", "WILKINSON", "MILLER", "JOHNSON", "SMITH"],
+        )
+
+    def test_filter_alias(self):
+        qs = Employee.objects.alias(
+            department_avg_age_diff=(
+                Window(Avg("age"), partition_by="department") - F("age")
+            ),
+        ).order_by("department", "name")
+        self.assertQuerysetEqual(
+            qs.filter(department_avg_age_diff__gt=0),
+            ["Jenson", "Jones", "Williams", "Miller", "Smith"],
+            lambda employee: employee.name,
+        )
+
+    def test_filter_select_related(self):
+        qs = (
+            Employee.objects.alias(
+                department_avg_age_diff=(
+                    Window(Avg("age"), partition_by="department") - F("age")
+                ),
+            )
+            .select_related("classification")
+            .filter(department_avg_age_diff__gt=0)
+            .order_by("department", "name")
+        )
+        self.assertQuerysetEqual(
+            qs,
+            ["Jenson", "Jones", "Williams", "Miller", "Smith"],
+            lambda employee: employee.name,
+        )
+        with self.assertNumQueries(0):
+            qs[0].classification
+
+    def test_exclude(self):
+        qs = Employee.objects.annotate(
+            department_salary_rank=Window(
+                Rank(), partition_by="department", order_by="-salary"
+            ),
+            department_avg_age_diff=(
+                Window(Avg("age"), partition_by="department") - F("age")
+            ),
+        ).order_by("department", "name")
+        # Direct window reference.
+        self.assertQuerysetEqual(
+            qs.exclude(department_salary_rank__gt=1),
+            ["Adams", "Wilkinson", "Miller", "Johnson", "Smith"],
+            lambda employee: employee.name,
+        )
+        # Through a combined expression containing a window.
+        self.assertQuerysetEqual(
+            qs.exclude(department_avg_age_diff__lte=0),
+            ["Jenson", "Jones", "Williams", "Miller", "Smith"],
+            lambda employee: employee.name,
+        )
+        # Union of multiple windows.
+        self.assertQuerysetEqual(
+            qs.exclude(
+                Q(department_salary_rank__gt=1) | Q(department_avg_age_diff__lte=0)
+            ),
+            ["Miller"],
+            lambda employee: employee.name,
+        )
+        # Intersection of multiple windows.
+        self.assertQuerysetEqual(
+            qs.exclude(department_salary_rank__gt=1, department_avg_age_diff__lte=0),
+            [
+                "Adams",
+                "Jenson",
+                "Jones",
+                "Williams",
+                "Wilkinson",
+                "Miller",
+                "Johnson",
+                "Smith",
+                "Smith",
+            ],
+            lambda employee: employee.name,
+        )
+
+    def test_heterogeneous_filter(self):
+        qs = (
+            Employee.objects.annotate(
+                department_salary_rank=Window(
+                    Rank(), partition_by="department", order_by="-salary"
+                ),
+            )
+            .order_by("name")
+            .values_list("name", flat=True)
+        )
+        # Heterogeneous filter between window function and aggregates pushes
+        # the WHERE clause to the QUALIFY outer query.
+        self.assertSequenceEqual(
+            qs.filter(
+                department_salary_rank=1, department__in=["Accounting", "Management"]
+            ),
+            ["Adams", "Miller"],
+        )
+        self.assertSequenceEqual(
+            qs.filter(
+                Q(department_salary_rank=1)
+                | Q(department__in=["Accounting", "Management"])
+            ),
+            [
+                "Adams",
+                "Jenson",
+                "Johnson",
+                "Johnson",
+                "Jones",
+                "Miller",
+                "Smith",
+                "Wilkinson",
+                "Williams",
+            ],
+        )
+        # Heterogeneous filter between window function and aggregates pushes
+        # the HAVING clause to the QUALIFY outer query.
+        qs = qs.annotate(past_department_count=Count("past_departments"))
+        self.assertSequenceEqual(
+            qs.filter(department_salary_rank=1, past_department_count__gte=1),
+            ["Johnson", "Miller"],
+        )
+        self.assertSequenceEqual(
+            qs.filter(Q(department_salary_rank=1) | Q(past_department_count__gte=1)),
+            ["Adams", "Johnson", "Miller", "Smith", "Wilkinson"],
+        )
+
+    def test_limited_filter(self):
+        """
+        A query filtering against a window function have its limit applied
+        after window filtering takes place.
+        """
+        self.assertQuerysetEqual(
+            Employee.objects.annotate(
+                department_salary_rank=Window(
+                    Rank(), partition_by="department", order_by="-salary"
+                )
+            )
+            .filter(department_salary_rank=1)
+            .order_by("department")[0:3],
+            ["Adams", "Wilkinson", "Miller"],
+            lambda employee: employee.name,
+        )
+
+    def test_filter_count(self):
+        self.assertEqual(
+            Employee.objects.annotate(
+                department_salary_rank=Window(
+                    Rank(), partition_by="department", order_by="-salary"
+                )
+            )
+            .filter(department_salary_rank=1)
+            .count(),
+            5,
+        )
+
     @skipUnlessDBFeature("supports_frame_range_fixed_distance")
     def test_range_n_preceding_and_following(self):
         qs = Employee.objects.annotate(
@@ -1071,6 +1337,7 @@ class WindowFunctionTests(TestCase):
                 ),
                 year=ExtractYear("hire_date"),
             )
+            .filter(sum__gte=45000)
             .values("year", "sum")
             .distinct("year")
             .order_by("year")
@@ -1081,7 +1348,6 @@ class WindowFunctionTests(TestCase):
             {"year": 2008, "sum": 45000},
             {"year": 2009, "sum": 128000},
             {"year": 2011, "sum": 60000},
-            {"year": 2012, "sum": 40000},
             {"year": 2013, "sum": 84000},
         ]
         for idx, val in zip(range(len(results)), results):
@@ -1348,34 +1614,18 @@ class NonQueryWindowTests(SimpleTestCase):
             frame.window_frame_start_end(None, None, None)
 
     def test_invalid_filter(self):
-        msg = "Window is disallowed in the filter clause"
+        msg = (
-        qs = Employee.objects.annotate(dense_rank=Window(expression=DenseRank()))
+            "Heterogeneous disjunctive predicates against window functions are not "
-        with self.assertRaisesMessage(NotSupportedError, msg):
+            "implemented when performing conditional aggregation."
-            qs.filter(dense_rank__gte=1)
-        with self.assertRaisesMessage(NotSupportedError, msg):
-            qs.annotate(inc_rank=F("dense_rank") + Value(1)).filter(inc_rank__gte=1)
-        with self.assertRaisesMessage(NotSupportedError, msg):
-            qs.filter(id=F("dense_rank"))
-        with self.assertRaisesMessage(NotSupportedError, msg):
-            qs.filter(id=Func("dense_rank", 2, function="div"))
-        with self.assertRaisesMessage(NotSupportedError, msg):
-            qs.annotate(total=Sum("dense_rank", filter=Q(name="Jones"))).filter(total=1)
-
-    def test_conditional_annotation(self):
-        qs = Employee.objects.annotate(
-            dense_rank=Window(expression=DenseRank()),
-        ).annotate(
-            equal=Case(
-                When(id=F("dense_rank"), then=Value(True)),
-                default=Value(False),
-                output_field=BooleanField(),
-            ),
         )
-        # The SQL standard disallows referencing window functions in the WHERE
+        qs = Employee.objects.annotate(
-        # clause.
+            window=Window(Rank()),
-        msg = "Window is disallowed in the filter clause"
+            past_dept_cnt=Count("past_departments"),
-        with self.assertRaisesMessage(NotSupportedError, msg):
+        )
-            qs.filter(equal=True)
+        with self.assertRaisesMessage(NotImplementedError, msg):
+            list(qs.filter(Q(window=1) | Q(department="Accounting")))
+        with self.assertRaisesMessage(NotImplementedError, msg):
+            list(qs.exclude(window=1, department="Accounting"))
 
     def test_invalid_order_by(self):
         msg = (