Fixed #26608 -- Added support for window expressions (OVER clause).

Thanks Josh Smeaton, Mariusz Felisiak, Sergey Fedoseev, Simon Charettes,
Adam Chainz/Johnson and Tim Graham for comments and reviews and Jamie
Cockburn for initial patch.

django/db/backends/base/features.py

@@ -236,6 +236,9 @@ class BaseDatabaseFeatures:
     # Does the backend support indexing a TextField?
     supports_index_on_text_field = True
 
+    # Does the backed support window expressions (expression OVER (...))?
+    supports_over_clause = False
+
     # Does the backend support CAST with precision?
     supports_cast_with_precision = True
 

django/db/backends/base/operations.py

@@ -4,7 +4,7 @@ from importlib import import_module
 
 from django.conf import settings
 from django.core.exceptions import ImproperlyConfigured
-from django.db import transaction
+from django.db import NotSupportedError, transaction
 from django.db.backends import utils
 from django.utils import timezone
 from django.utils.dateparse import parse_duration
@@ -39,6 +39,13 @@ class BaseDatabaseOperations:
     # CharField data type if the max_length argument isn't provided.
     cast_char_field_without_max_length = None
 
+    # Start and end points for window expressions.
+    PRECEDING = 'PRECEDING'
+    FOLLOWING = 'FOLLOWING'
+    UNBOUNDED_PRECEDING = 'UNBOUNDED ' + PRECEDING
+    UNBOUNDED_FOLLOWING = 'UNBOUNDED ' + FOLLOWING
+    CURRENT_ROW = 'CURRENT ROW'
+
     def __init__(self, connection):
         self.connection = connection
         self._cache = None
@@ -598,3 +605,34 @@ class BaseDatabaseOperations:
             rhs_sql, rhs_params = rhs
             return "(%s - %s)" % (lhs_sql, rhs_sql), lhs_params + rhs_params
         raise NotImplementedError("This backend does not support %s subtraction." % internal_type)
+
+    def window_frame_start(self, start):
+        if isinstance(start, int):
+            if start < 0:
+                return '%d %s' % (abs(start), self.PRECEDING)
+            elif start == 0:
+                return self.CURRENT_ROW
+        elif start is None:
+            return self.UNBOUNDED_PRECEDING
+        raise ValueError("start argument must be a negative integer, zero, or None, but got '%s'." % start)
+
+    def window_frame_end(self, end):
+        if isinstance(end, int):
+            if end == 0:
+                return self.CURRENT_ROW
+            elif end > 0:
+                return '%d %s' % (end, self.FOLLOWING)
+        elif end is None:
+            return self.UNBOUNDED_FOLLOWING
+        raise ValueError("end argument must be a positive integer, zero, or None, but got '%s'." % end)
+
+    def window_frame_rows_start_end(self, start=None, end=None):
+        """
+        Return SQL for start and end points in an OVER clause window frame.
+        """
+        if not self.connection.features.supports_over_clause:
+            raise NotSupportedError('This backend does not support window expressions.')
+        return self.window_frame_start(start), self.window_frame_end(end)
+
+    def window_frame_range_start_end(self, start=None, end=None):
+        return self.window_frame_rows_start_end(start, end)

django/db/backends/mysql/features.py

@@ -81,6 +81,10 @@ class DatabaseFeatures(BaseDatabaseFeatures):
             result = cursor.fetchone()
             return result and result[0] == 1
 
+    @cached_property
+    def supports_over_clause(self):
+        return self.connection.mysql_version >= (8, 0, 2)
+
     @cached_property
     def supports_transactions(self):
         """

django/db/backends/oracle/features.py

@@ -54,3 +54,4 @@ class DatabaseFeatures(BaseDatabaseFeatures):
         END;
     """
     supports_callproc_kwargs = True
+    supports_over_clause = True

django/db/backends/postgresql/features.py

@@ -48,6 +48,7 @@ class DatabaseFeatures(BaseDatabaseFeatures):
             V_I := P_I;
         END;
     $$ LANGUAGE plpgsql;"""
+    supports_over_clause = True
 
     @cached_property
     def supports_aggregate_filter_clause(self):

django/db/backends/postgresql/operations.py

@@ -1,6 +1,7 @@
 from psycopg2.extras import Inet
 
 from django.conf import settings
+from django.db import NotSupportedError
 from django.db.backends.base.operations import BaseDatabaseOperations
 
 
@@ -247,3 +248,12 @@ class DatabaseOperations(BaseDatabaseOperations):
             rhs_sql, rhs_params = rhs
             return "(interval '1 day' * (%s - %s))" % (lhs_sql, rhs_sql), lhs_params + rhs_params
         return super().subtract_temporals(internal_type, lhs, rhs)
+
+    def window_frame_range_start_end(self, start=None, end=None):
+        start_, end_ = super().window_frame_range_start_end(start, end)
+        if (start and start < 0) or (end and end > 0):
+            raise NotSupportedError(
+                'PostgreSQL only supports UNBOUNDED together with PRECEDING '
+                'and FOLLOWING.'
+            )
+        return start_, end_

django/db/models/__init__.py

@@ -6,8 +6,8 @@ from django.db.models.deletion import (
     CASCADE, DO_NOTHING, PROTECT, SET, SET_DEFAULT, SET_NULL, ProtectedError,
 )
 from django.db.models.expressions import (
-    Case, Exists, Expression, ExpressionWrapper, F, Func, OuterRef, Subquery,
-    Value, When,
+    Case, Exists, Expression, ExpressionList, ExpressionWrapper, F, Func,
+    OuterRef, RowRange, Subquery, Value, ValueRange, When, Window, WindowFrame,
 )
 from django.db.models.fields import *  # NOQA
 from django.db.models.fields import __all__ as fields_all
@@ -64,8 +64,9 @@ __all__ += [
     'ObjectDoesNotExist', 'signals',
     'CASCADE', 'DO_NOTHING', 'PROTECT', 'SET', 'SET_DEFAULT', 'SET_NULL',
     'ProtectedError',
-    'Case', 'Exists', 'Expression', 'ExpressionWrapper', 'F', 'Func',
-    'OuterRef', 'Subquery', 'Value', 'When',
+    'Case', 'Exists', 'Expression', 'ExpressionList', 'ExpressionWrapper', 'F',
+    'Func', 'OuterRef', 'RowRange', 'Subquery', 'Value', 'ValueRange', 'When',
+    'Window', 'WindowFrame',
     'FileField', 'ImageField', 'OrderWrt', 'Lookup', 'Transform', 'Manager',
     'Prefetch', 'Q', 'QuerySet', 'prefetch_related_objects', 'DEFERRED', 'Model',
     'ForeignKey', 'ForeignObject', 'OneToOneField', 'ManyToManyField',

django/db/models/aggregates.py

@@ -15,6 +15,7 @@ class Aggregate(Func):
     contains_aggregate = True
     name = None
     filter_template = '%s FILTER (WHERE %%(filter)s)'
+    window_compatible = True
 
     def __init__(self, *args, filter=None, **kwargs):
         self.filter = filter

django/db/models/expressions.py

@@ -3,6 +3,7 @@ import datetime
 from decimal import Decimal
 
 from django.core.exceptions import EmptyResultSet, FieldError
+from django.db import connection
 from django.db.models import fields
 from django.db.models.query_utils import Q
 from django.utils.deconstruct import deconstructible
@@ -140,6 +141,10 @@ class BaseExpression:
     # aggregate specific fields
     is_summary = False
     _output_field_resolved_to_none = False
+    # Can the expression be used in a WHERE clause?
+    filterable = True
+    # Can the expression can be used as a source expression in Window?
+    window_compatible = False
 
     def __init__(self, output_field=None):
         if output_field is not None:
@@ -206,6 +211,13 @@ class BaseExpression:
                 return True
         return False
 
+    @cached_property
+    def contains_over_clause(self):
+        for expr in self.get_source_expressions():
+            if expr and expr.contains_over_clause:
+                return True
+        return False
+
     @cached_property
     def contains_column_references(self):
         for expr in self.get_source_expressions():
@@ -232,6 +244,7 @@ class BaseExpression:
         c.is_summary = summarize
         c.set_source_expressions([
             expr.resolve_expression(query, allow_joins, reuse, summarize)
+            if expr else None
             for expr in c.get_source_expressions()
         ])
         return c
@@ -482,6 +495,9 @@ class TemporalSubtraction(CombinedExpression):
 @deconstructible
 class F(Combinable):
     """An object capable of resolving references to existing query objects."""
+    # Can the expression be used in a WHERE clause?
+    filterable = True
+
     def __init__(self, name):
         """
         Arguments:
@@ -767,6 +783,23 @@ class Ref(Expression):
         return [self]
 
 
+class ExpressionList(Func):
+    """
+    An expression containing multiple expressions. Can be used to provide a
+    list of expressions as an argument to another expression, like an
+    ordering clause.
+    """
+    template = '%(expressions)s'
+
+    def __init__(self, *expressions, **extra):
+        if len(expressions) == 0:
+            raise ValueError('%s requires at least one expression.' % self.__class__.__name__)
+        super().__init__(*expressions, **extra)
+
+    def __str__(self):
+        return self.arg_joiner.join(str(arg) for arg in self.source_expressions)
+
+
 class ExpressionWrapper(Expression):
     """
     An expression that can wrap another expression so that it can provide
@@ -1118,3 +1151,168 @@ class OrderBy(BaseExpression):
 
     def desc(self):
         self.descending = True
+
+
+class Window(Expression):
+    template = '%(expression)s OVER (%(window)s)'
+    # Although the main expression may either be an aggregate or an
+    # expression with an aggregate function, the GROUP BY that will
+    # be introduced in the query as a result is not desired.
+    contains_aggregate = False
+    contains_over_clause = True
+    filterable = False
+
+    def __init__(self, expression, partition_by=None, order_by=None, frame=None, output_field=None):
+        self.partition_by = partition_by
+        self.order_by = order_by
+        self.frame = frame
+
+        if not getattr(expression, 'window_compatible', False):
+            raise ValueError(
+                "Expression '%s' isn't compatible with OVER clauses." %
+                expression.__class__.__name__
+            )
+
+        if self.partition_by is not None:
+            if not isinstance(self.partition_by, (tuple, list)):
+                self.partition_by = (self.partition_by,)
+            self.partition_by = ExpressionList(*self.partition_by)
+
+        if self.order_by is not None:
+            if isinstance(self.order_by, (list, tuple)):
+                self.order_by = ExpressionList(*self.order_by)
+            elif not isinstance(self.order_by, BaseExpression):
+                raise ValueError(
+                    'order_by must be either an Expression or a sequence of '
+                    'expressions.'
+                )
+        super().__init__(output_field=output_field)
+        self.source_expression = self._parse_expressions(expression)[0]
+
+    def _resolve_output_field(self):
+        return self.source_expression.output_field
+
+    def get_source_expressions(self):
+        return [self.source_expression, self.partition_by, self.order_by, self.frame]
+
+    def set_source_expressions(self, exprs):
+        self.source_expression, self.partition_by, self.order_by, self.frame = exprs
+
+    def as_sql(self, compiler, connection, function=None, template=None):
+        connection.ops.check_expression_support(self)
+        expr_sql, params = compiler.compile(self.source_expression)
+        window_sql, window_params = [], []
+
+        if self.partition_by is not None:
+            sql_expr, sql_params = self.partition_by.as_sql(
+                compiler=compiler, connection=connection,
+                template='PARTITION BY %(expressions)s',
+            )
+            window_sql.extend(sql_expr)
+            window_params.extend(sql_params)
+
+        if self.order_by is not None:
+            window_sql.append(' ORDER BY ')
+            order_sql, order_params = compiler.compile(self.order_by)
+            window_sql.extend(''.join(order_sql))
+            window_params.extend(order_params)
+
+        if self.frame:
+            frame_sql, frame_params = compiler.compile(self.frame)
+            window_sql.extend(' ' + frame_sql)
+            window_params.extend(frame_params)
+
+        params.extend(window_params)
+        template = template or self.template
+
+        return template % {
+            'expression': expr_sql,
+            'window': ''.join(window_sql).strip()
+        }, params
+
+    def __str__(self):
+        return '{} OVER ({}{}{})'.format(
+            str(self.source_expression),
+            'PARTITION BY ' + str(self.partition_by) if self.partition_by else '',
+            'ORDER BY ' + str(self.order_by) if self.order_by else '',
+            str(self.frame or ''),
+        )
+
+    def __repr__(self):
+        return '<%s: %s>' % (self.__class__.__name__, self)
+
+    def get_group_by_cols(self):
+        return []
+
+
+class WindowFrame(Expression):
+    """
+    Model the frame clause in window expressions. There are two types of frame
+    clauses which are subclasses, however, all processing and validation (by no
+    means intended to be complete) is done here. Thus, providing an end for a
+    frame is optional (the default is UNBOUNDED FOLLOWING, which is the last
+    row in the frame).
+    """
+    template = '%(frame_type)s BETWEEN %(start)s AND %(end)s'
+
+    def __init__(self, start=None, end=None):
+        self.start = start
+        self.end = end
+
+    def set_source_expressions(self, exprs):
+        self.start, self.end = exprs
+
+    def get_source_expressions(self):
+        return [Value(self.start), Value(self.end)]
+
+    def as_sql(self, compiler, connection):
+        connection.ops.check_expression_support(self)
+        start, end = self.window_frame_start_end(connection, self.start.value, self.end.value)
+        return self.template % {
+            'frame_type': self.frame_type,
+            'start': start,
+            'end': end,
+        }, []
+
+    def __repr__(self):
+        return '<%s: %s>' % (self.__class__.__name__, self)
+
+    def get_group_by_cols(self):
+        return []
+
+    def __str__(self):
+        if self.start is not None and self.start < 0:
+            start = '%d %s' % (abs(self.start), connection.ops.PRECEDING)
+        elif self.start is not None and self.start == 0:
+            start = connection.ops.CURRENT_ROW
+        else:
+            start = connection.ops.UNBOUNDED_PRECEDING
+
+        if self.end is not None and self.end > 0:
+            end = '%d %s' % (self.end, connection.ops.FOLLOWING)
+        elif self.end is not None and self.end == 0:
+            end = connection.ops.CURRENT_ROW
+        else:
+            end = connection.ops.UNBOUNDED_FOLLOWING
+        return self.template % {
+            'frame_type': self.frame_type,
+            'start': start,
+            'end': end,
+        }
+
+    def window_frame_start_end(self, connection, start, end):
+        raise NotImplementedError('Subclasses must implement window_frame_start_end().')
+
+
+class RowRange(WindowFrame):
+    frame_type = 'ROWS'
+
+    def window_frame_start_end(self, connection, start, end):
+        return connection.ops.window_frame_rows_start_end(start, end)
+
+
+class ValueRange(WindowFrame):
+    frame_type = 'RANGE'
+
+    def window_frame_start_end(self, connection, start, end):
+        return connection.ops.window_frame_range_start_end(start, end)

django/db/models/functions/__init__.py

@@ -8,6 +8,10 @@ from .datetime import (
     Trunc, TruncDate, TruncDay, TruncHour, TruncMinute, TruncMonth,
     TruncQuarter, TruncSecond, TruncTime, TruncYear,
 )
+from .window import (
+    CumeDist, DenseRank, FirstValue, Lag, LastValue, Lead, NthValue, Ntile,
+    PercentRank, Rank, RowNumber,
+)
 
 __all__ = [
     # base
@@ -18,4 +22,7 @@ __all__ = [
     'ExtractQuarter', 'ExtractSecond', 'ExtractWeek', 'ExtractWeekDay',
     'ExtractYear', 'Trunc', 'TruncDate', 'TruncDay', 'TruncHour', 'TruncMinute',
     'TruncMonth', 'TruncQuarter', 'TruncSecond', 'TruncTime', 'TruncYear',
+    # window
+    'CumeDist', 'DenseRank', 'FirstValue', 'Lag', 'LastValue', 'Lead',
+    'NthValue', 'Ntile', 'PercentRank', 'Rank', 'RowNumber',
 ]

django/db/models/functions/window.py

@@ -0,0 +1,118 @@
+from django.db.models import FloatField, Func, IntegerField
+
+__all__ = [
+    'CumeDist', 'DenseRank', 'FirstValue', 'Lag', 'LastValue', 'Lead',
+    'NthValue', 'Ntile', 'PercentRank', 'Rank', 'RowNumber',
+]
+
+
+class CumeDist(Func):
+    function = 'CUME_DIST'
+    name = 'CumeDist'
+    output_field = FloatField()
+    window_compatible = True
+
+
+class DenseRank(Func):
+    function = 'DENSE_RANK'
+    name = 'DenseRank'
+    output_field = IntegerField()
+    window_compatible = True
+
+
+class FirstValue(Func):
+    arity = 1
+    function = 'FIRST_VALUE'
+    name = 'FirstValue'
+    window_compatible = True
+
+
+class LagLeadFunction(Func):
+    window_compatible = True
+
+    def __init__(self, expression, offset=1, default=None, **extra):
+        if expression is None:
+            raise ValueError(
+                '%s requires a non-null source expression.' %
+                self.__class__.__name__
+            )
+        if offset is None or offset <= 0:
+            raise ValueError(
+                '%s requires a positive integer for the offset.' %
+                self.__class__.__name__
+            )
+        args = (expression, offset)
+        if default is not None:
+            args += (default,)
+        super().__init__(*args, **extra)
+
+    def _resolve_output_field(self):
+        sources = self.get_source_expressions()
+        return sources[0].output_field
+
+
+class Lag(LagLeadFunction):
+    function = 'LAG'
+    name = 'Lag'
+
+
+class LastValue(Func):
+    arity = 1
+    function = 'LAST_VALUE'
+    name = 'LastValue'
+    window_compatible = True
+
+
+class Lead(LagLeadFunction):
+    function = 'LEAD'
+    name = 'Lead'
+
+
+class NthValue(Func):
+    function = 'NTH_VALUE'
+    name = 'NthValue'
+    window_compatible = True
+
+    def __init__(self, expression, nth=1, **extra):
+        if expression is None:
+            raise ValueError('%s requires a non-null source expression.' % self.__class__.__name__)
+        if nth is None or nth <= 0:
+            raise ValueError('%s requires a positive integer as for nth.' % self.__class__.__name__)
+        super().__init__(expression, nth, **extra)
+
+    def _resolve_output_field(self):
+        sources = self.get_source_expressions()
+        return sources[0].output_field
+
+
+class Ntile(Func):
+    function = 'NTILE'
+    name = 'Ntile'
+    output_field = IntegerField()
+    window_compatible = True
+
+    def __init__(self, num_buckets=1, **extra):
+        if num_buckets <= 0:
+            raise ValueError('num_buckets must be greater than 0.')
+        super().__init__(num_buckets, **extra)
+
+
+class PercentRank(Func):
+    function = 'PERCENT_RANK'
+    name = 'PercentRank'
+    output_field = FloatField()
+    window_compatible = True
+
+
+class Rank(Func):
+    function = 'RANK'
+    name = 'Rank'
+    output_field = IntegerField()
+    window_compatible = True
+
+
+class RowNumber(Func):
+    function = 'ROW_NUMBER'
+    name = 'RowNumber'
+    output_field = IntegerField()
+    window_compatible = True

django/db/models/lookups.py

@@ -115,6 +115,10 @@ class Lookup:
     def contains_aggregate(self):
         return self.lhs.contains_aggregate or getattr(self.rhs, 'contains_aggregate', False)
 
+    @cached_property
+    def contains_over_clause(self):
+        return self.lhs.contains_over_clause or getattr(self.rhs, 'contains_over_clause', False)
+
     @property
     def is_summary(self):
         return self.lhs.is_summary or getattr(self.rhs, 'is_summary', False)

django/db/models/sql/compiler.py

@@ -1107,6 +1107,8 @@ class SQLInsertCompiler(SQLCompiler):
                 )
             if value.contains_aggregate:
                 raise FieldError("Aggregate functions are not allowed in this query")
+            if value.contains_over_clause:
+                raise FieldError('Window expressions are not allowed in this query.')
         else:
             value = field.get_db_prep_save(value, connection=self.connection)
         return value
@@ -1262,6 +1264,8 @@ class SQLUpdateCompiler(SQLCompiler):
                 val = val.resolve_expression(self.query, allow_joins=False, for_save=True)
                 if val.contains_aggregate:
                     raise FieldError("Aggregate functions are not allowed in this query")
+                if val.contains_over_clause:
+                    raise FieldError('Window expressions are not allowed in this query.')
             elif hasattr(val, 'prepare_database_save'):
                 if field.remote_field:
                     val = field.get_db_prep_save(

django/db/models/sql/query.py

@@ -13,7 +13,7 @@ from string import ascii_uppercase
 from django.core.exceptions import (
     EmptyResultSet, FieldDoesNotExist, FieldError,
 )
-from django.db import DEFAULT_DB_ALIAS, connections
+from django.db import DEFAULT_DB_ALIAS, NotSupportedError, connections
 from django.db.models.aggregates import Count
 from django.db.models.constants import LOOKUP_SEP
 from django.db.models.expressions import Col, Ref
@@ -1125,6 +1125,13 @@ class Query:
         if not arg:
             raise FieldError("Cannot parse keyword query %r" % arg)
         lookups, parts, reffed_expression = self.solve_lookup_type(arg)
+
+        if not getattr(reffed_expression, 'filterable', True):
+            raise NotSupportedError(
+                reffed_expression.__class__.__name__ + ' is disallowed in '
+                'the filter clause.'
+            )
+
         if not allow_joins and len(parts) > 1:
             raise FieldError("Joined field references are not permitted in this query")
 

django/db/models/sql/where.py

@@ -167,6 +167,16 @@ class WhereNode(tree.Node):
     def contains_aggregate(self):
         return self._contains_aggregate(self)
 
+    @classmethod
+    def _contains_over_clause(cls, obj):
+        if isinstance(obj, tree.Node):
+            return any(cls._contains_over_clause(c) for c in obj.children)
+        return obj.contains_over_clause
+
+    @cached_property
+    def contains_over_clause(self):
+        return self._contains_over_clause(self)
+
     @property
     def is_summary(self):
         return any(child.is_summary for child in self.children)

docs/ref/models/database-functions.txt

@@ -819,3 +819,132 @@ Usage example::
      'minute': 'minute': datetime.datetime(2014, 6, 15, 14, 30, tzinfo=<UTC>),
      'second': datetime.datetime(2014, 6, 15, 14, 30, 50, tzinfo=<UTC>)
     }
+
+.. _window-functions:
+
+Window functions
+================
+
+.. versionadded:: 2.0
+
+There are a number of functions to use in a
+:class:`~django.db.models.expressions.Window` expression for computing the rank
+of elements or the :class:`Ntile` of some rows.
+
+``CumeDist``
+------------
+
+.. class:: CumeDist(*expressions, **extra)
+
+Calculates the cumulative distribution of a value within a window or partition.
+The cumulative distribution is defined as the number of rows preceding or
+peered with the current row divided by the total number of rows in the frame.
+
+``DenseRank``
+-------------
+
+.. class:: DenseRank(*expressions, **extra)
+
+Equivalent to :class:`Rank` but does not have gaps.
+
+``FirstValue``
+--------------
+
+.. class:: FirstValue(expression, **extra)
+
+Returns the value evaluated at the row that's the first row of the window
+frame, or ``None`` if no such value exists.
+
+``Lag``
+-------
+
+.. class:: Lag(expression, offset=1, default=None, **extra)
+
+Calculates the value offset by ``offset``, and if no row exists there, returns
+``default``.
+
+``default`` must have the same type as the ``expression``, however, this is
+only validated by the database and not in Python.
+
+``LastValue``
+-------------
+
+.. class:: LastValue(expression, **extra)
+
+Comparable to :class:`FirstValue`, it calculates the last value in a given
+frame clause.
+
+``Lead``
+--------
+
+.. class:: Lead(expression, offset=1, default=None, **extra)
+
+Calculates the leading value in a given :ref:`frame <window-frames>`. Both
+``offset`` and ``default`` are evaluated with respect to the current row.
+
+``default`` must have the same type as the ``expression``, however, this is
+only validated by the database and not in Python.
+
+``NthValue``
+------------
+
+.. class:: NthValue(expression, nth=1, **extra)
+
+Computes the row relative to the offset ``nth`` (must be a positive value)
+within the window. Returns ``None`` if no row exists.
+
+Some databases may handle a nonexistent nth-value differently. For example,
+Oracle returns an empty string rather than ``None`` for character-based
+expressions. Django doesn't do any conversions in these cases.
+
+``Ntile``
+---------
+
+.. class:: Ntile(num_buckets=1, **extra)
+
+Calculates a partition for each of the rows in the frame clause, distributing
+numbers as evenly as possible between 1 and ``num_buckets``. If the rows don't
+divide evenly into a number of buckets, one or more buckets will be represented
+more frequently.
+
+``PercentRank``
+---------------
+
+.. class:: PercentRank(*expressions, **extra)
+
+Computes the percentile rank of the rows in the frame clause. This
+computation is equivalent to evaluating::
+
+    (rank - 1) / (total rows - 1)
+
+The following table explains the calculation for the percentile rank of a row:
+
+===== ===== ==== ============ ============
+Row # Value Rank Calculation  Percent Rank
+===== ===== ==== ============ ============
+1     15    1    (1-1)/(7-1)  0.0000
+2     20    2    (2-1)/(7-1)  0.1666
+3     20    2    (2-1)/(7-1)  0.1666
+4     20    2    (2-1)/(7-1)  0.1666
+5     30    5    (5-1)/(7-1)  0.6666
+6     30    5    (5-1)/(7-1)  0.6666
+7     40    7    (7-1)/(7-1)  1.0000
+===== ===== ==== ============ ============
+
+``Rank``
+--------
+
+.. class:: Rank(*expressions, **extra)
+
+Comparable to ``RowNumber``, this function ranks rows in the window. The
+computed rank contains gaps. Use :class:`DenseRank` to compute rank without
+gaps.
+
+``RowNumber``
+-------------
+
+.. class:: RowNumber(*expressions, **extra)
+
+Computes the row number according to the ordering of either the frame clause
+or the ordering of the whole query if there is no partitioning of the
+:ref:`window frame <window-frames>`.

docs/ref/models/expressions.txt

@@ -353,6 +353,13 @@ The ``Aggregate`` API is as follows:
         generated. Specifically, the ``function`` will be interpolated as the
         ``function`` placeholder within :attr:`template`. Defaults to ``None``.
 
+    .. attribute:: window_compatible
+
+        .. versionadded:: 2.0
+
+        Defaults to ``True`` since most aggregate functions can be used as the
+        source expression in :class:`~django.db.models.expressions.Window`.
+
 The ``expression`` argument can be the name of a field on the model, or another
 expression. It will be converted to a string and used as the ``expressions``
 placeholder within the ``template``.
@@ -649,6 +656,184 @@ should avoid them if possible.
     force you to acknowledge that you're not interpolating your SQL with user
     provided data.
 
+Window functions
+----------------
+
+.. versionadded:: 2.0
+
+Window functions provide a way to apply functions on partitions. Unlike a
+normal aggregation function which computes a final result for each set defined
+by the group by, window functions operate on :ref:`frames <window-frames>` and
+partitions, and compute the result for each row.
+
+You can specify multiple windows in the same query which in Django ORM would be
+equivalent to including multiple expressions in a :doc:`QuerySet.annotate()
+</topics/db/aggregation>` call. The ORM doesn't make use of named windows,
+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
+        ``expression`` argument is provided, the window clause will be blank.
+
+The ``Window`` class is the main expression for an ``OVER`` clause.
+
+The ``expression`` argument is either a :ref:`window function
+<window-functions>`, an :ref:`aggregate function <aggregation-functions>`, or
+an expression that's compatible in a window clause.
+
+The ``partition_by`` argument is a list of expressions (column names should be
+wrapped in an ``F``-object) that control the partitioning of the rows.
+Partitioning narrows which rows are used to compute the result set.
+
+The ``output_field`` is specified either as an argument or by the expression.
+
+The ``order_by`` argument accepts a sequence of expressions on which you can
+call :meth:`~django.db.models.Expression.asc` and
+:meth:`~django.db.models.Expression.desc`. The ordering controls the order in
+which the expression is applied. For example, if you sum over the rows in a
+partition, the first result is just the value of the first row, the second is
+the sum of first and second row.
+
+The ``frame`` parameter specifies which other rows that should be used in the
+computation. See :ref:`window-frames` for details.
+
+For example, to annotate each movie with the average rating for the movies by
+the same studio in the same genre and release year::
+
+    >>> from django.db.models import Avg, ExtractYear, F, Window
+    >>> Movie.objects.annotate(
+    >>>     avg_rating=Window(
+    >>>         expression=Avg('rating'),
+    >>>         partition_by=[F('studio'), F('genre')],
+    >>>         order_by=ExtractYear('released').asc(),
+    >>>     ),
+    >>> )
+
+This makes it easy to check if a movie is rated better or worse than its peers.
+
+You may want to apply multiple expressions over the same window, i.e., the
+same partition and frame. For example, you could modify the previous example
+to also include the best and worst rating in each movie's group (same studio,
+genre, and release year) by using three window functions in the same query. The
+partition and ordering from the previous example is extracted into a dictionary
+to reduce repetition::
+
+    >>> from django.db.models import Avg, ExtractYear, F, Max, Min, Window
+    >>> window = {
+    >>>    'partition': [F('studio'), F('genre')],
+    >>>    'order_by': ExtractYear('released').asc(),
+    >>> }
+    >>> Movie.objects.annotate(
+    >>>     avg_rating=Window(
+    >>>         expression=Avg('rating'), **window,
+    >>>     ),
+    >>>     best=Window(
+    >>>         expression=Max('rating'), **window,
+    >>>     ),
+    >>>     worst=Window(
+    >>>         expression=Min('rating'), **window,
+    >>>     ),
+    >>> )
+
+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
+:meth:`~django.db.models.Expression.asc` and
+:meth:`~django.db.models.Expression.desc` may not be supported. Consult the
+documentation for your database as needed.
+
+.. _window-frames:
+
+Frames
+~~~~~~
+
+For a window frame, you can choose either a range-based sequence of rows or an
+ordinary sequence of rows.
+
+.. class:: ValueRange(start=None, end=None)
+
+    .. attribute:: frame_type
+
+        This attribute is set to ``'RANGE'``.
+
+    PostgreSQL has limited support for ``ValueRange`` and only supports use of
+    the standard start and end points, such as ``CURRENT ROW`` and ``UNBOUNDED
+    FOLLOWING``.
+
+.. class:: RowRange(start=None, end=None)
+
+    .. attribute:: frame_type
+
+        This attribute is set to ``'ROWS'``.
+
+Both classes return SQL with the template::
+
+    %(frame_type)s BETWEEN %(start)s AND %(end)s
+
+Frames narrow the rows that are used for computing the result. They shift from
+some start point to some specified end point. Frames can be used with and
+without partitions, but it's often a good idea to specify an ordering of the
+window to ensure a deterministic result. In a frame, a peer in a frame is a row
+with an equivalent value, or all rows if an ordering clause isn't present.
+
+The default starting point for a frame is ``UNBOUNDED PRECEDING`` which is the
+first row of the partition. The end point is always explicitly included in the
+SQL generated by the ORM and is by default ``UNBOUNDED FOLLOWING``. The default
+frame includes all rows from the partition to the last row in the set.
+
+The accepted values for the ``start`` and ``end`` arguments are ``None``, an
+integer, or zero. A negative integer for ``start`` results in ``N preceding``,
+while ``None`` yields ``UNBOUNDED PRECEDING``. For both ``start`` and ``end``,
+zero will return ``CURRENT ROW``. Positive integers are accepted for ``end``.
+
+There's a difference in what ``CURRENT ROW`` includes. When specified in
+``ROWS`` mode, the frame starts or ends with the current row. When specified in
+``RANGE`` mode, the frame starts or ends at the first or last peer according to
+the ordering clause. Thus, ``RANGE CURRENT ROW`` evaluates the expression for
+rows which have the same value specified by the ordering. Because the template
+includes both the ``start`` and ``end`` points, this may be expressed with::
+
+    ValueRange(start=0, end=0)
+
+If a movie's "peers" are described as movies released by the same studio in the
+same genre in the same year, this ``RowRange`` example annotates each movie
+with the average rating of a movie's two prior and two following peers::
+
+    >>> from django.db.models import Avg, ExtractYear, F, RowRange, Window
+    >>> Movie.objects.annotate(
+    >>>     avg_rating=Window(
+    >>>         expression=Avg('rating'),
+    >>>         partition_by=[F('studio'), F('genre')],
+    >>>         order_by=ExtractYear('released').asc(),
+    >>>         frame=RowRange(start=-2, end=2),
+    >>>     ),
+    >>> )
+
+If the database supports it, you can specify the start and end points based on
+values of an expression in the partition. If the ``released`` field of the
+``Movie`` model stores the release month of each movies, this ``ValueRange``
+example annotates each movie with the average rating of a movie's peers
+released between twelve months before and twelve months after the each movie.
+
+    >>> from django.db.models import Avg, ExpressionList, F, ValueRange, Window
+    >>> Movie.objects.annotate(
+    >>>     avg_rating=Window(
+    >>>         expression=Avg('rating'),
+    >>>         partition_by=[F('studio'), F('genre')],
+    >>>         order_by=F('released').asc(),
+    >>>         frame=ValueRange(start=-12, end=12),
+    >>>     ),
+    >>> )
+
 .. currentmodule:: django.db.models
 
 Technical Information
@@ -677,6 +862,30 @@ calling the appropriate methods on the wrapped expression.
         Tells Django that this expression contains an aggregate and that a
         ``GROUP BY`` clause needs to be added to the query.
 
+    .. attribute:: contains_over_clause
+
+        .. versionadded:: 2.0
+
+        Tells Django that this expression contains a
+        :class:`~django.db.models.expressions.Window` expression. It's used,
+        for example, to disallow window function expressions in queries that
+        modify data. Defaults to ``True``.
+
+    .. attribute:: filterable
+
+        .. versionadded:: 2.0
+
+        Tells Django that this expression can be referenced in
+        :meth:`.QuerySet.filter`. Defaults to ``True``.
+
+    .. attribute:: window_compatible
+
+        .. versionadded:: 2.0
+
+        Tells Django that this expression can be used as the source expression
+        in :class:`~django.db.models.expressions.Window`. Defaults to
+        ``False``.
+
     .. method:: resolve_expression(query=None, allow_joins=True, reuse=None, summarize=False, for_save=False)
 
         Provides the chance to do any pre-processing or validation of

docs/ref/models/querysets.txt

@@ -1681,6 +1681,11 @@ raised if ``select_for_update()`` is used in autocommit mode.
     ``select_for_update()`` you should use
     :class:`~django.test.TransactionTestCase`.
 
+.. admonition:: Certain expressions may not be supported
+
+    PostgreSQL doesn't support ``select_for_update()`` with
+    :class:`~django.db.models.expressions.Window` expressions.
+
 .. versionchanged:: 1.11
 
     The ``skip_locked`` argument was added.

docs/releases/2.0.txt

@@ -52,6 +52,14 @@ Mobile-friendly ``contrib.admin``
 The admin is now responsive and supports all major mobile devices.
 Older browser may experience varying levels of graceful degradation.
 
+Window expressions
+------------------
+
+The new :class:`~django.db.models.expressions.Window` expression allows
+adding an ``OVER`` clause to querysets. You can use :ref:`window functions
+<window-functions>` and :ref:`aggregate functions <aggregation-functions>` in
+the expression.
+
 Minor features
 --------------
 
@@ -404,6 +412,11 @@ backends.
   requires that the arguments to ``OF`` be columns rather than tables, set
   ``DatabaseFeatures.select_for_update_of_column = True``.
 
+* To enable support for :class:`~django.db.models.expressions.Window`
+  expressions, set ``DatabaseFeatures.supports_over_clause`` to ``True``. You
+  may need to customize the ``DatabaseOperations.window_start_rows_start_end()``
+  and/or ``window_start_range_start_end()`` methods.
+
 * Third-party database backends should add a
   ``DatabaseOperations.cast_char_field_without_max_length`` attribute with the
   database data type that will be used in the

tests/backends/base/test_operations.py

@@ -0,0 +1,10 @@
+from django.db import NotSupportedError, connection
+from django.test import SimpleTestCase, skipIfDBFeature
+
+
+class DatabaseOperationTests(SimpleTestCase):
+    @skipIfDBFeature('supports_over_clause')
+    def test_window_frame_raise_not_supported_error(self):
+        msg = 'This backend does not support window expressions.'
+        with self.assertRaisesMessage(NotSupportedError, msg):
+            connection.ops.window_frame_rows_start_end()

tests/db_functions/test_window.py

@@ -0,0 +1,39 @@
+from django.db.models.functions import Lag, Lead, NthValue, Ntile
+from django.test import SimpleTestCase
+
+
+class ValidationTests(SimpleTestCase):
+    def test_nth_negative_nth_value(self):
+        msg = 'NthValue requires a positive integer as for nth'
+        with self.assertRaisesMessage(ValueError, msg):
+            NthValue(expression='salary', nth=-1)
+
+    def test_nth_null_expression(self):
+        msg = 'NthValue requires a non-null source expression'
+        with self.assertRaisesMessage(ValueError, msg):
+            NthValue(expression=None)
+
+    def test_lag_negative_offset(self):
+        msg = 'Lag requires a positive integer for the offset'
+        with self.assertRaisesMessage(ValueError, msg):
+            Lag(expression='salary', offset=-1)
+
+    def test_lead_negative_offset(self):
+        msg = 'Lead requires a positive integer for the offset'
+        with self.assertRaisesMessage(ValueError, msg):
+            Lead(expression='salary', offset=-1)
+
+    def test_null_source_lead(self):
+        msg = 'Lead requires a non-null source expression'
+        with self.assertRaisesMessage(ValueError, msg):
+            Lead(expression=None)
+
+    def test_null_source_lag(self):
+        msg = 'Lag requires a non-null source expression'
+        with self.assertRaisesMessage(ValueError, msg):
+            Lag(expression=None)
+
+    def test_negative_num_buckets_ntile(self):
+        msg = 'num_buckets must be greater than 0'
+        with self.assertRaisesMessage(ValueError, msg):
+            Ntile(num_buckets=-1)

tests/expressions/tests.py

@@ -10,8 +10,8 @@ from django.db.models.aggregates import (
     Avg, Count, Max, Min, StdDev, Sum, Variance,
 )
 from django.db.models.expressions import (
-    Case, Col, Exists, ExpressionWrapper, F, Func, OrderBy, OuterRef, Random,
-    RawSQL, Ref, Subquery, Value, When,
+    Case, Col, Exists, ExpressionList, ExpressionWrapper, F, Func, OrderBy,
+    OuterRef, Random, RawSQL, Ref, Subquery, Value, When,
 )
 from django.db.models.functions import (
     Coalesce, Concat, Length, Lower, Substr, Upper,
@@ -1330,6 +1330,11 @@ class ValueTests(TestCase):
         self.assertNotEqual(value, other_value)
         self.assertNotEqual(value, no_output_field)
 
+    def test_raise_empty_expressionlist(self):
+        msg = 'ExpressionList requires at least one expression'
+        with self.assertRaisesMessage(ValueError, msg):
+            ExpressionList()
+
 
 class ReprTests(TestCase):
 
@@ -1355,6 +1360,14 @@ class ReprTests(TestCase):
         self.assertEqual(repr(RawSQL('table.col', [])), "RawSQL(table.col, [])")
         self.assertEqual(repr(Ref('sum_cost', Sum('cost'))), "Ref(sum_cost, Sum(F(cost)))")
         self.assertEqual(repr(Value(1)), "Value(1)")
+        self.assertEqual(
+            repr(ExpressionList(F('col'), F('anothercol'))),
+            'ExpressionList(F(col), F(anothercol))'
+        )
+        self.assertEqual(
+            repr(ExpressionList(OrderBy(F('col'), descending=False))),
+            'ExpressionList(OrderBy(F(col), descending=False))'
+        )
 
     def test_functions(self):
         self.assertEqual(repr(Coalesce('a', 'b')), "Coalesce(F(a), F(b))")

tests/expressions_window/models.py

@@ -0,0 +1,11 @@
+from django.db import models
+
+
+class Employee(models.Model):
+    name = models.CharField(max_length=40, blank=False, null=False)
+    salary = models.PositiveIntegerField()
+    department = models.CharField(max_length=40, blank=False, null=False)
+    hire_date = models.DateField(blank=False, null=False)
+
+    def __str__(self):
+        return '{}, {}, {}, {}'.format(self.name, self.department, self.salary, self.hire_date)

tests/expressions_window/tests.py

@@ -0,0 +1,783 @@
+import datetime
+from unittest import skipIf, skipUnless
+
+from django.core.exceptions import FieldError
+from django.db import NotSupportedError, connection
+from django.db.models import (
+    F, RowRange, Value, ValueRange, Window, WindowFrame,
+)
+from django.db.models.aggregates import Avg, Max, Min, Sum
+from django.db.models.functions import (
+    CumeDist, DenseRank, ExtractYear, FirstValue, Lag, LastValue, Lead,
+    NthValue, Ntile, PercentRank, Rank, RowNumber, Upper,
+)
+from django.test import SimpleTestCase, TestCase, skipUnlessDBFeature
+
+from .models import Employee
+
+
+@skipUnlessDBFeature('supports_over_clause')
+class WindowFunctionTests(TestCase):
+    @classmethod
+    def setUpTestData(cls):
+        Employee.objects.bulk_create([
+            Employee(name=e[0], salary=e[1], department=e[2], hire_date=e[3])
+            for e in [
+                ('Jones', 45000, 'Accounting', datetime.datetime(2005, 11, 1)),
+                ('Williams', 37000, 'Accounting', datetime.datetime(2009, 6, 1)),
+                ('Jenson', 45000, 'Accounting', datetime.datetime(2008, 4, 1)),
+                ('Adams', 50000, 'Accounting', datetime.datetime(2013, 7, 1)),
+                ('Smith', 55000, 'Sales', datetime.datetime(2007, 6, 1)),
+                ('Brown', 53000, 'Sales', datetime.datetime(2009, 9, 1)),
+                ('Johnson', 40000, 'Marketing', datetime.datetime(2012, 3, 1)),
+                ('Smith', 38000, 'Marketing', datetime.datetime(2009, 10, 1)),
+                ('Wilkinson', 60000, 'IT', datetime.datetime(2011, 3, 1)),
+                ('Moore', 34000, 'IT', datetime.datetime(2013, 8, 1)),
+                ('Miller', 100000, 'Management', datetime.datetime(2005, 6, 1)),
+                ('Johnson', 80000, 'Management', datetime.datetime(2005, 7, 1)),
+            ]
+        ])
+
+    def test_dense_rank(self):
+        qs = Employee.objects.annotate(rank=Window(
+            expression=DenseRank(),
+            order_by=ExtractYear(F('hire_date')).asc(),
+        ))
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 1),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 1),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 1),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 2),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 3),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 4),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 4),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 4),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 5),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 6),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 7),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 7),
+        ], lambda entry: (entry.name, entry.salary, entry.department, entry.hire_date, entry.rank), ordered=False)
+
+    def test_department_salary(self):
+        qs = Employee.objects.annotate(department_sum=Window(
+            expression=Sum('salary'),
+            partition_by=F('department'),
+            order_by=[F('hire_date').asc()],
+        )).order_by('department', 'department_sum')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', 45000, 45000),
+            ('Jenson', 'Accounting', 45000, 90000),
+            ('Williams', 'Accounting', 37000, 127000),
+            ('Adams', 'Accounting', 50000, 177000),
+            ('Wilkinson', 'IT', 60000, 60000),
+            ('Moore', 'IT', 34000, 94000),
+            ('Miller', 'Management', 100000, 100000),
+            ('Johnson', 'Management', 80000, 180000),
+            ('Smith', 'Marketing', 38000, 38000),
+            ('Johnson', 'Marketing', 40000, 78000),
+            ('Smith', 'Sales', 55000, 55000),
+            ('Brown', 'Sales', 53000, 108000),
+        ], lambda entry: (entry.name, entry.department, entry.salary, entry.department_sum))
+
+    def test_rank(self):
+        """
+        Rank the employees based on the year they're were hired. Since there
+        are multiple employees hired in different years, this will contain
+        gaps.
+        """
+        qs = Employee.objects.annotate(rank=Window(
+            expression=Rank(),
+            order_by=ExtractYear(F('hire_date')).asc(),
+        ))
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 1),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 1),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 1),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 4),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 5),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 6),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 6),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 6),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 9),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 10),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 11),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 11),
+        ], lambda entry: (entry.name, entry.salary, entry.department, entry.hire_date, entry.rank), ordered=False)
+
+    def test_row_number(self):
+        """
+        The row number window function computes the number based on the order
+        in which the tuples were inserted. Depending on the backend,
+
+        Oracle requires an ordering-clause in the Window expression.
+        """
+        qs = Employee.objects.annotate(row_number=Window(
+            expression=RowNumber(),
+            order_by=F('pk').asc(),
+        )).order_by('pk')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', 1),
+            ('Williams', 'Accounting', 2),
+            ('Jenson', 'Accounting', 3),
+            ('Adams', 'Accounting', 4),
+            ('Smith', 'Sales', 5),
+            ('Brown', 'Sales', 6),
+            ('Johnson', 'Marketing', 7),
+            ('Smith', 'Marketing', 8),
+            ('Wilkinson', 'IT', 9),
+            ('Moore', 'IT', 10),
+            ('Miller', 'Management', 11),
+            ('Johnson', 'Management', 12),
+        ], lambda entry: (entry.name, entry.department, entry.row_number))
+
+    @skipIf(connection.vendor == 'oracle', "Oracle requires ORDER BY in row_number, ANSI:SQL doesn't")
+    def test_row_number_no_ordering(self):
+        """
+        The row number window function computes the number based on the order
+        in which the tuples were inserted.
+        """
+        # Add a default ordering for consistent results across databases.
+        qs = Employee.objects.annotate(row_number=Window(
+            expression=RowNumber(),
+        )).order_by('pk')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', 1),
+            ('Williams', 'Accounting', 2),
+            ('Jenson', 'Accounting', 3),
+            ('Adams', 'Accounting', 4),
+            ('Smith', 'Sales', 5),
+            ('Brown', 'Sales', 6),
+            ('Johnson', 'Marketing', 7),
+            ('Smith', 'Marketing', 8),
+            ('Wilkinson', 'IT', 9),
+            ('Moore', 'IT', 10),
+            ('Miller', 'Management', 11),
+            ('Johnson', 'Management', 12),
+        ], lambda entry: (entry.name, entry.department, entry.row_number))
+
+    def test_avg_salary_department(self):
+        qs = Employee.objects.annotate(avg_salary=Window(
+            expression=Avg('salary'),
+            order_by=F('department').asc(),
+            partition_by='department',
+        )).order_by('department', '-salary', 'name')
+        self.assertQuerysetEqual(qs, [
+            ('Adams', 50000, 'Accounting', 44250.00),
+            ('Jenson', 45000, 'Accounting', 44250.00),
+            ('Jones', 45000, 'Accounting', 44250.00),
+            ('Williams', 37000, 'Accounting', 44250.00),
+            ('Wilkinson', 60000, 'IT', 47000.00),
+            ('Moore', 34000, 'IT', 47000.00),
+            ('Miller', 100000, 'Management', 90000.00),
+            ('Johnson', 80000, 'Management', 90000.00),
+            ('Johnson', 40000, 'Marketing', 39000.00),
+            ('Smith', 38000, 'Marketing', 39000.00),
+            ('Smith', 55000, 'Sales', 54000.00),
+            ('Brown', 53000, 'Sales', 54000.00),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.avg_salary))
+
+    def test_lag(self):
+        """
+        Compute the difference between an employee's salary and the next
+        highest salary in the employee's department. Return None if the
+        employee has the lowest salary.
+        """
+        qs = Employee.objects.annotate(lag=Window(
+            expression=Lag(expression='salary', offset=1),
+            partition_by=F('department'),
+            order_by=[F('salary').asc(), F('name').asc()],
+        )).order_by('department')
+        self.assertQuerysetEqual(qs, [
+            ('Williams', 37000, 'Accounting', None),
+            ('Jenson', 45000, 'Accounting', 37000),
+            ('Jones', 45000, 'Accounting', 45000),
+            ('Adams', 50000, 'Accounting', 45000),
+            ('Moore', 34000, 'IT', None),
+            ('Wilkinson', 60000, 'IT', 34000),
+            ('Johnson', 80000, 'Management', None),
+            ('Miller', 100000, 'Management', 80000),
+            ('Smith', 38000, 'Marketing', None),
+            ('Johnson', 40000, 'Marketing', 38000),
+            ('Brown', 53000, 'Sales', None),
+            ('Smith', 55000, 'Sales', 53000),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.lag))
+
+    def test_first_value(self):
+        qs = Employee.objects.annotate(first_value=Window(
+            expression=FirstValue('salary'),
+            partition_by=F('department'),
+            order_by=F('hire_date').asc(),
+        )).order_by('department', 'hire_date')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 45000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 45000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 45000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 45000),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 60000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 60000),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 100000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 100000),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 38000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 38000),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 55000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 55000),
+        ], lambda row: (row.name, row.salary, row.department, row.hire_date, row.first_value))
+
+    def test_last_value(self):
+        qs = Employee.objects.annotate(last_value=Window(
+            expression=LastValue('hire_date'),
+            partition_by=F('department'),
+            order_by=F('hire_date').asc(),
+        ))
+        self.assertQuerysetEqual(qs, [
+            ('Adams', 'Accounting', datetime.date(2013, 7, 1), 50000, datetime.date(2013, 7, 1)),
+            ('Jenson', 'Accounting', datetime.date(2008, 4, 1), 45000, datetime.date(2008, 4, 1)),
+            ('Jones', 'Accounting', datetime.date(2005, 11, 1), 45000, datetime.date(2005, 11, 1)),
+            ('Williams', 'Accounting', datetime.date(2009, 6, 1), 37000, datetime.date(2009, 6, 1)),
+            ('Moore', 'IT', datetime.date(2013, 8, 1), 34000, datetime.date(2013, 8, 1)),
+            ('Wilkinson', 'IT', datetime.date(2011, 3, 1), 60000, datetime.date(2011, 3, 1)),
+            ('Miller', 'Management', datetime.date(2005, 6, 1), 100000, datetime.date(2005, 6, 1)),
+            ('Johnson', 'Management', datetime.date(2005, 7, 1), 80000, datetime.date(2005, 7, 1)),
+            ('Johnson', 'Marketing', datetime.date(2012, 3, 1), 40000, datetime.date(2012, 3, 1)),
+            ('Smith', 'Marketing', datetime.date(2009, 10, 1), 38000, datetime.date(2009, 10, 1)),
+            ('Brown', 'Sales', datetime.date(2009, 9, 1), 53000, datetime.date(2009, 9, 1)),
+            ('Smith', 'Sales', datetime.date(2007, 6, 1), 55000, datetime.date(2007, 6, 1)),
+        ], transform=lambda row: (row.name, row.department, row.hire_date, row.salary, row.last_value), ordered=False)
+
+    def test_function_list_of_values(self):
+        qs = Employee.objects.annotate(lead=Window(
+            expression=Lead(expression='salary'),
+            order_by=[F('hire_date').asc(), F('name').desc()],
+            partition_by='department',
+        )).values_list('name', 'salary', 'department', 'hire_date', 'lead')
+        self.assertNotIn('GROUP BY', str(qs.query))
+        self.assertSequenceEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 45000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 37000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 50000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), None),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 34000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), None),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 80000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), None),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 40000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), None),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 53000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), None),
+        ])
+
+    def test_min_department(self):
+        """An alternative way to specify a query for FirstValue."""
+        qs = Employee.objects.annotate(min_salary=Window(
+            expression=Min('salary'),
+            partition_by=F('department'),
+            order_by=[F('salary').asc(), F('name').asc()]
+        )).order_by('department', 'salary', 'name')
+        self.assertQuerysetEqual(qs, [
+            ('Williams', 'Accounting', 37000, 37000),
+            ('Jenson', 'Accounting', 45000, 37000),
+            ('Jones', 'Accounting', 45000, 37000),
+            ('Adams', 'Accounting', 50000, 37000),
+            ('Moore', 'IT', 34000, 34000),
+            ('Wilkinson', 'IT', 60000, 34000),
+            ('Johnson', 'Management', 80000, 80000),
+            ('Miller', 'Management', 100000, 80000),
+            ('Smith', 'Marketing', 38000, 38000),
+            ('Johnson', 'Marketing', 40000, 38000),
+            ('Brown', 'Sales', 53000, 53000),
+            ('Smith', 'Sales', 55000, 53000),
+        ], lambda row: (row.name, row.department, row.salary, row.min_salary))
+
+    def test_max_per_year(self):
+        """
+        Find the maximum salary awarded in the same year as the
+        employee was hired, regardless of the department.
+        """
+        qs = Employee.objects.annotate(max_salary_year=Window(
+            expression=Max('salary'),
+            order_by=ExtractYear('hire_date').asc(),
+            partition_by=ExtractYear('hire_date')
+        )).order_by(ExtractYear('hire_date'), 'salary')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', 45000, 2005, 100000),
+            ('Johnson', 'Management', 80000, 2005, 100000),
+            ('Miller', 'Management', 100000, 2005, 100000),
+            ('Smith', 'Sales', 55000, 2007, 55000),
+            ('Jenson', 'Accounting', 45000, 2008, 45000),
+            ('Williams', 'Accounting', 37000, 2009, 53000),
+            ('Smith', 'Marketing', 38000, 2009, 53000),
+            ('Brown', 'Sales', 53000, 2009, 53000),
+            ('Wilkinson', 'IT', 60000, 2011, 60000),
+            ('Johnson', 'Marketing', 40000, 2012, 40000),
+            ('Moore', 'IT', 34000, 2013, 50000),
+            ('Adams', 'Accounting', 50000, 2013, 50000),
+        ], lambda row: (row.name, row.department, row.salary, row.hire_date.year, row.max_salary_year))
+
+    def test_cume_dist(self):
+        """
+        Compute the cumulative distribution for the employees based on the
+        salary in increasing order. Equal to rank/total number of rows (12).
+        """
+        qs = Employee.objects.annotate(cume_dist=Window(
+            expression=CumeDist(),
+            order_by=F('salary').asc(),
+        )).order_by('salary', 'name')
+        # Round result of cume_dist because Oracle uses greater precision.
+        self.assertQuerysetEqual(qs, [
+            ('Moore', 'IT', 34000, 0.0833333333),
+            ('Williams', 'Accounting', 37000, 0.1666666667),
+            ('Smith', 'Marketing', 38000, 0.25),
+            ('Johnson', 'Marketing', 40000, 0.3333333333),
+            ('Jenson', 'Accounting', 45000, 0.5),
+            ('Jones', 'Accounting', 45000, 0.5),
+            ('Adams', 'Accounting', 50000, 0.5833333333),
+            ('Brown', 'Sales', 53000, 0.6666666667),
+            ('Smith', 'Sales', 55000, 0.75),
+            ('Wilkinson', 'IT', 60000, 0.8333333333),
+            ('Johnson', 'Management', 80000, 0.9166666667),
+            ('Miller', 'Management', 100000, 1),
+        ], lambda row: (row.name, row.department, row.salary, round(row.cume_dist, 10)))
+
+    def test_nthvalue(self):
+        qs = Employee.objects.annotate(
+            nth_value=Window(expression=NthValue(
+                expression='salary', nth=2),
+                order_by=[F('hire_date').asc(), F('name').desc()],
+                partition_by=F('department'),
+            )
+        ).order_by('department', 'hire_date', 'name')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', datetime.date(2005, 11, 1), 45000, None),
+            ('Jenson', 'Accounting', datetime.date(2008, 4, 1), 45000, 45000),
+            ('Williams', 'Accounting', datetime.date(2009, 6, 1), 37000, 45000),
+            ('Adams', 'Accounting', datetime.date(2013, 7, 1), 50000, 45000),
+            ('Wilkinson', 'IT', datetime.date(2011, 3, 1), 60000, None),
+            ('Moore', 'IT', datetime.date(2013, 8, 1), 34000, 34000),
+            ('Miller', 'Management', datetime.date(2005, 6, 1), 100000, None),
+            ('Johnson', 'Management', datetime.date(2005, 7, 1), 80000, 80000),
+            ('Smith', 'Marketing', datetime.date(2009, 10, 1), 38000, None),
+            ('Johnson', 'Marketing', datetime.date(2012, 3, 1), 40000, 40000),
+            ('Smith', 'Sales', datetime.date(2007, 6, 1), 55000, None),
+            ('Brown', 'Sales', datetime.date(2009, 9, 1), 53000, 53000),
+        ], lambda row: (row.name, row.department, row.hire_date, row.salary, row.nth_value))
+
+    def test_lead(self):
+        """
+        Determine what the next person hired in the same department makes.
+        Because the dataset is ambiguous, the name is also part of the
+        ordering clause. No default is provided, so None/NULL should be
+        returned.
+        """
+        qs = Employee.objects.annotate(lead=Window(
+            expression=Lead(expression='salary'),
+            order_by=[F('hire_date').asc(), F('name').desc()],
+            partition_by='department',
+        )).order_by('department')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 45000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 37000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 50000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), None),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 34000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), None),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 80000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), None),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 40000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), None),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 53000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), None),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.lead))
+
+    def test_lead_offset(self):
+        """
+        Determine what the person hired after someone makes. Due to
+        ambiguity, the name is also included in the ordering.
+        """
+        qs = Employee.objects.annotate(lead=Window(
+            expression=Lead('salary', offset=2),
+            partition_by='department',
+            order_by=F('hire_date').asc(),
+        ))
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 37000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 50000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), None),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), None),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), None),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), None),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), None),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), None),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), None),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), None),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), None),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), None),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.lead),
+            ordered=False
+        )
+
+    def test_lead_default(self):
+        qs = Employee.objects.annotate(lead_default=Window(
+            expression=Lead(expression='salary', offset=5, default=60000),
+            partition_by=F('department'),
+            order_by=F('department').asc(),
+        ))
+        self.assertEqual(list(qs.values_list('lead_default', flat=True).distinct()), [60000])
+
+    def test_ntile(self):
+        """
+        Compute the group for each of the employees across the entire company,
+        based on how high the salary is for them. There are twelve employees
+        so it divides evenly into four groups.
+        """
+        qs = Employee.objects.annotate(ntile=Window(
+            expression=Ntile(num_buckets=4),
+            order_by=F('salary').desc(),
+        )).order_by('ntile', '-salary', 'name')
+        self.assertQuerysetEqual(qs, [
+            ('Miller', 'Management', 100000, 1),
+            ('Johnson', 'Management', 80000, 1),
+            ('Wilkinson', 'IT', 60000, 1),
+            ('Smith', 'Sales', 55000, 2),
+            ('Brown', 'Sales', 53000, 2),
+            ('Adams', 'Accounting', 50000, 2),
+            ('Jenson', 'Accounting', 45000, 3),
+            ('Jones', 'Accounting', 45000, 3),
+            ('Johnson', 'Marketing', 40000, 3),
+            ('Smith', 'Marketing', 38000, 4),
+            ('Williams', 'Accounting', 37000, 4),
+            ('Moore', 'IT', 34000, 4),
+        ], lambda x: (x.name, x.department, x.salary, x.ntile))
+
+    def test_percent_rank(self):
+        """
+        Calculate the percentage rank of the employees across the entire
+        company based on salary and name (in case of ambiguity).
+        """
+        qs = Employee.objects.annotate(percent_rank=Window(
+            expression=PercentRank(),
+            order_by=[F('salary').asc(), F('name').asc()],
+        )).order_by('percent_rank')
+        # Round to account for precision differences among databases.
+        self.assertQuerysetEqual(qs, [
+            ('Moore', 'IT', 34000, 0.0),
+            ('Williams', 'Accounting', 37000, 0.0909090909),
+            ('Smith', 'Marketing', 38000, 0.1818181818),
+            ('Johnson', 'Marketing', 40000, 0.2727272727),
+            ('Jenson', 'Accounting', 45000, 0.3636363636),
+            ('Jones', 'Accounting', 45000, 0.4545454545),
+            ('Adams', 'Accounting', 50000, 0.5454545455),
+            ('Brown', 'Sales', 53000, 0.6363636364),
+            ('Smith', 'Sales', 55000, 0.7272727273),
+            ('Wilkinson', 'IT', 60000, 0.8181818182),
+            ('Johnson', 'Management', 80000, 0.9090909091),
+            ('Miller', 'Management', 100000, 1.0),
+        ], transform=lambda row: (row.name, row.department, row.salary, round(row.percent_rank, 10)))
+
+    def test_nth_returns_null(self):
+        """
+        Find the nth row of the data set. None is returned since there are
+        fewer than 20 rows in the test data.
+        """
+        qs = Employee.objects.annotate(nth_value=Window(
+            expression=NthValue('salary', nth=20),
+            order_by=F('salary').asc()
+        ))
+        self.assertEqual(list(qs.values_list('nth_value', flat=True).distinct()), [None])
+
+    def test_multiple_partitioning(self):
+        """
+        Find the maximum salary for each department for people hired in the
+        same year.
+        """
+        qs = Employee.objects.annotate(max=Window(
+            expression=Max('salary'),
+            partition_by=[F('department'), ExtractYear(F('hire_date'))],
+        )).order_by('department', 'hire_date', 'name')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 45000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 45000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 37000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 50000),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 60000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 34000),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 100000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 100000),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 38000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 40000),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 55000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 53000),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.max))
+
+    def test_multiple_ordering(self):
+        """
+        Accumulate the salaries over the departments based on hire_date.
+        If two people were hired on the same date in the same department, the
+        ordering clause will render a different result for those people.
+        """
+        qs = Employee.objects.annotate(sum=Window(
+            expression=Sum('salary'),
+            partition_by='department',
+            order_by=[F('hire_date').asc(), F('name').asc()],
+        )).order_by('department', 'sum')
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 45000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 90000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 127000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 177000),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 60000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 94000),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 100000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 180000),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 38000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 78000),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 55000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 108000),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.sum))
+
+    @skipIf(connection.vendor == 'postgresql', 'n following/preceding not supported by PostgreSQL')
+    def test_range_n_preceding_and_following(self):
+        qs = Employee.objects.annotate(sum=Window(
+            expression=Sum('salary'),
+            order_by=F('salary').asc(),
+            partition_by='department',
+            frame=ValueRange(start=-2, end=2),
+        ))
+        self.assertIn('RANGE BETWEEN 2 PRECEDING AND 2 FOLLOWING', str(qs.query))
+        self.assertQuerysetEqual(qs, [
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 37000),
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 90000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 90000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 50000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 53000),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 55000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 40000),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 38000),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 60000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 34000),
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 100000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 80000),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.sum), ordered=False)
+
+    def test_range_unbound(self):
+        """A query with RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING."""
+        qs = Employee.objects.annotate(sum=Window(
+            expression=Sum('salary'),
+            partition_by='department',
+            order_by=[F('hire_date').asc(), F('name').asc()],
+            frame=ValueRange(start=None, end=None),
+        )).order_by('department', 'hire_date', 'name')
+        self.assertIn('RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING', str(qs.query))
+        self.assertQuerysetEqual(qs, [
+            ('Jones', 'Accounting', 45000, datetime.date(2005, 11, 1), 177000),
+            ('Jenson', 'Accounting', 45000, datetime.date(2008, 4, 1), 177000),
+            ('Williams', 'Accounting', 37000, datetime.date(2009, 6, 1), 177000),
+            ('Adams', 'Accounting', 50000, datetime.date(2013, 7, 1), 177000),
+            ('Wilkinson', 'IT', 60000, datetime.date(2011, 3, 1), 94000),
+            ('Moore', 'IT', 34000, datetime.date(2013, 8, 1), 94000),
+            ('Miller', 'Management', 100000, datetime.date(2005, 6, 1), 180000),
+            ('Johnson', 'Management', 80000, datetime.date(2005, 7, 1), 180000),
+            ('Smith', 'Marketing', 38000, datetime.date(2009, 10, 1), 78000),
+            ('Johnson', 'Marketing', 40000, datetime.date(2012, 3, 1), 78000),
+            ('Smith', 'Sales', 55000, datetime.date(2007, 6, 1), 108000),
+            ('Brown', 'Sales', 53000, datetime.date(2009, 9, 1), 108000),
+        ], transform=lambda row: (row.name, row.department, row.salary, row.hire_date, row.sum))
+
+    def test_row_range_rank(self):
+        """
+        A query with ROWS BETWEEN UNBOUNDED PRECEDING AND 3 FOLLOWING.
+        The resulting sum is the sum of the three next (if they exist) and all
+        previous rows according to the ordering clause.
+        """
+        qs = Employee.objects.annotate(sum=Window(
+            expression=Sum('salary'),
+            order_by=[F('hire_date').asc(), F('name').desc()],
+            frame=RowRange(start=None, end=3),
+        )).order_by('sum', 'hire_date')
+        self.assertIn('ROWS BETWEEN UNBOUNDED PRECEDING AND 3 FOLLOWING', str(qs.query))
+        self.assertQuerysetEqual(qs, [
+            ('Miller', 100000, 'Management', datetime.date(2005, 6, 1), 280000),
+            ('Johnson', 80000, 'Management', datetime.date(2005, 7, 1), 325000),
+            ('Jones', 45000, 'Accounting', datetime.date(2005, 11, 1), 362000),
+            ('Smith', 55000, 'Sales', datetime.date(2007, 6, 1), 415000),
+            ('Jenson', 45000, 'Accounting', datetime.date(2008, 4, 1), 453000),
+            ('Williams', 37000, 'Accounting', datetime.date(2009, 6, 1), 513000),
+            ('Brown', 53000, 'Sales', datetime.date(2009, 9, 1), 553000),
+            ('Smith', 38000, 'Marketing', datetime.date(2009, 10, 1), 603000),
+            ('Wilkinson', 60000, 'IT', datetime.date(2011, 3, 1), 637000),
+            ('Johnson', 40000, 'Marketing', datetime.date(2012, 3, 1), 637000),
+            ('Adams', 50000, 'Accounting', datetime.date(2013, 7, 1), 637000),
+            ('Moore', 34000, 'IT', datetime.date(2013, 8, 1), 637000),
+        ], transform=lambda row: (row.name, row.salary, row.department, row.hire_date, row.sum))
+
+    @skipUnlessDBFeature('can_distinct_on_fields')
+    def test_distinct_window_function(self):
+        """
+        Window functions are not aggregates, and hence a query to filter out
+        duplicates may be useful.
+        """
+        qs = Employee.objects.annotate(
+            sum=Window(
+                expression=Sum('salary'),
+                partition_by=ExtractYear('hire_date'),
+                order_by=ExtractYear('hire_date')
+            ),
+            year=ExtractYear('hire_date'),
+        ).values('year', 'sum').distinct('year').order_by('year')
+        results = [
+            {'year': 2005, 'sum': 225000}, {'year': 2007, 'sum': 55000},
+            {'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):
+            with self.subTest(result=val):
+                self.assertEqual(qs[idx], val)
+
+    def test_fail_update(self):
+        """Window expressions can't be used in an UPDATE statement."""
+        msg = 'Window expressions are not allowed in this query'
+        with self.assertRaisesMessage(FieldError, msg):
+            Employee.objects.filter(department='Management').update(
+                salary=Window(expression=Max('salary'), partition_by='department'),
+            )
+
+    def test_fail_insert(self):
+        """Window expressions can't be used in an INSERT statement."""
+        msg = 'Window expressions are not allowed in this query'
+        with self.assertRaisesMessage(FieldError, msg):
+            Employee.objects.create(
+                name='Jameson', department='Management', hire_date=datetime.date(2007, 7, 1),
+                salary=Window(expression=Sum(Value(10000), order_by=F('pk').asc())),
+            )
+
+    def test_invalid_start_value_range(self):
+        msg = "start argument must be a negative integer, zero, or None, but got '3'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=ValueRange(start=3),
+            )))
+
+    def test_invalid_end_value_range(self):
+        msg = "end argument must be a positive integer, zero, or None, but got '-3'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=ValueRange(end=-3),
+            )))
+
+    def test_invalid_type_end_value_range(self):
+        msg = "end argument must be a positive integer, zero, or None, but got 'a'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=ValueRange(end='a'),
+            )))
+
+    def test_invalid_type_start_value_range(self):
+        msg = "start argument must be a negative integer, zero, or None, but got 'a'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                frame=ValueRange(start='a'),
+            )))
+
+    def test_invalid_type_end_row_range(self):
+        msg = "end argument must be a positive integer, zero, or None, but got 'a'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                frame=RowRange(end='a'),
+            )))
+
+    @skipUnless(connection.vendor == 'postgresql', 'Frame construction not allowed on PostgreSQL')
+    def test_postgresql_illegal_range_frame_start(self):
+        msg = 'PostgreSQL only supports UNBOUNDED together with PRECEDING and FOLLOWING.'
+        with self.assertRaisesMessage(NotSupportedError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=ValueRange(start=-1),
+            )))
+
+    @skipUnless(connection.vendor == 'postgresql', 'Frame construction not allowed on PostgreSQL')
+    def test_postgresql_illegal_range_frame_end(self):
+        msg = 'PostgreSQL only supports UNBOUNDED together with PRECEDING and FOLLOWING.'
+        with self.assertRaisesMessage(NotSupportedError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=ValueRange(end=1),
+            )))
+
+    def test_invalid_type_start_row_range(self):
+        msg = "start argument must be a negative integer, zero, or None, but got 'a'."
+        with self.assertRaisesMessage(ValueError, msg):
+            list(Employee.objects.annotate(test=Window(
+                expression=Sum('salary'),
+                order_by=F('hire_date').asc(),
+                frame=RowRange(start='a'),
+            )))
+
+
+class NonQueryWindowTests(SimpleTestCase):
+    def test_window_repr(self):
+        self.assertEqual(
+            repr(Window(expression=Sum('salary'), partition_by='department')),
+            '<Window: Sum(F(salary)) OVER (PARTITION BY F(department))>'
+        )
+        self.assertEqual(
+            repr(Window(expression=Avg('salary'), order_by=F('department').asc())),
+            '<Window: Avg(F(salary)) OVER (ORDER BY OrderBy(F(department), descending=False))>'
+        )
+
+    def test_window_frame_repr(self):
+        self.assertEqual(
+            repr(RowRange(start=-1)),
+            '<RowRange: ROWS BETWEEN 1 PRECEDING AND UNBOUNDED FOLLOWING>'
+        )
+        self.assertEqual(
+            repr(ValueRange(start=None, end=1)),
+            '<ValueRange: RANGE BETWEEN UNBOUNDED PRECEDING AND 1 FOLLOWING>'
+        )
+        self.assertEqual(
+            repr(ValueRange(start=0, end=0)),
+            '<ValueRange: RANGE BETWEEN CURRENT ROW AND CURRENT ROW>'
+        )
+        self.assertEqual(
+            repr(RowRange(start=0, end=0)),
+            '<RowRange: ROWS BETWEEN CURRENT ROW AND CURRENT ROW>'
+        )
+
+    def test_empty_group_by_cols(self):
+        window = Window(expression=Sum('pk'))
+        self.assertEqual(window.get_group_by_cols(), [])
+        self.assertFalse(window.contains_aggregate)
+
+    def test_frame_empty_group_by_cols(self):
+        frame = WindowFrame()
+        self.assertEqual(frame.get_group_by_cols(), [])
+
+    def test_frame_window_frame_notimplemented(self):
+        frame = WindowFrame()
+        msg = 'Subclasses must implement window_frame_start_end().'
+        with self.assertRaisesMessage(NotImplementedError, msg):
+            frame.window_frame_start_end(None, None, None)
+
+    def test_invalid_filter(self):
+        msg = 'Window is disallowed in the filter clause'
+        with self.assertRaisesMessage(NotSupportedError, msg):
+            Employee.objects.annotate(dense_rank=Window(expression=DenseRank())).filter(dense_rank__gte=1)
+
+    def test_invalid_order_by(self):
+        msg = 'order_by must be either an Expression or a sequence of expressions'
+        with self.assertRaisesMessage(ValueError, msg):
+            Window(expression=Sum('power'), order_by='-horse')
+
+    def test_invalid_source_expression(self):
+        msg = "Expression 'Upper' isn't compatible with OVER clauses."
+        with self.assertRaisesMessage(ValueError, msg):
+            Window(expression=Upper('name'))