Composite Column Types

Sets of columns can be associated with a single user-defined datatype, which in modern use is normally a Python dataclass. The ORM provides a single attribute which represents the group of columns using the class you provide.

A simple example represents pairs of Integer columns as a Point object, with attributes .x and .y. Using a dataclass, these attributes are defined with the corresponding int Python type:

import dataclasses

class Point:
    x: int
    y: int

Non-dataclass forms are also accepted, but require additional methods to be implemented. For an example using a non-dataclass class, see the section Using Legacy Non-Dataclasses.

New in version 2.0: The composite() construct fully supports Python dataclasses including the ability to derive mapped column datatypes from the composite class.

We will create a mapping to a table vertices, which represents two points as x1/y1 and x2/y2. The Point class is associated with the mapped columns using the composite() construct.

The example below illustrates the most modern form of composite() as used with a fully Annotated Declarative Table configuration. mapped_column() constructs representing each column are passed directly to composite(), indicating zero or more aspects of the columns to be generated, in this case the names; the composite() construct derives the column types (in this case int, corresponding to Integer) from the dataclass directly:

from sqlalchemy.orm import DeclarativeBase, Mapped
from sqlalchemy.orm import composite, mapped_column

class Base(DeclarativeBase):

class Vertex(Base):
    __tablename__ = "vertices"

    id: Mapped[int] = mapped_column(primary_key=True)

    start: Mapped[Point] = composite(mapped_column("x1"), mapped_column("y1"))
    end: Mapped[Point] = composite(mapped_column("x2"), mapped_column("y2"))

    def __repr__(self):
        return f"Vertex(start={self.start}, end={self.end})"


In the example above the columns that represent the composites (x1, y1, etc.) are also accessible on the class but are not correctly understood by type checkers. If accessing the single columns is important they can be explicitly declared, as shown in Map columns directly, pass attribute names to composite.

The above mapping would correspond to a CREATE TABLE statement as:

>>> from sqlalchemy.schema import CreateTable
>>> print(CreateTable(Vertex.__table__))

Working with Mapped Composite Column Types

With a mapping as illustrated in the top section, we can work with the Vertex class, where the .start and .end attributes will transparently refer to the columns referenced by the Point class, as well as with instances of the Vertex class, where the .start and .end attributes will refer to instances of the Point class. The x1, y1, x2, and y2 columns are handled transparently:

  • Persisting Point objects

    We can create a Vertex object, assign Point objects as members, and they will be persisted as expected:

    >>> v = Vertex(start=Point(3, 4), end=Point(5, 6))
    >>> session.add(v)
    >>> session.commit()
    BEGIN (implicit) INSERT INTO vertices (x1, y1, x2, y2) VALUES (?, ?, ?, ?) [generated in ...] (3, 4, 5, 6) COMMIT
  • Selecting Point objects as columns

    composite() will allow the Vertex.start and Vertex.end attributes to behave like a single SQL expression to as much an extent as possible when using the ORM Session (including the legacy Query object) to select Point objects:

    >>> stmt = select(Vertex.start, Vertex.end)
    >>> session.execute(stmt).all()
    SELECT vertices.x1, vertices.y1, vertices.x2, vertices.y2 FROM vertices [...] ()
    [(Point(x=3, y=4), Point(x=5, y=6))]
  • Comparing Point objects in SQL expressions

    The Vertex.start and Vertex.end attributes may be used in WHERE criteria and similar, using ad-hoc Point objects for comparisons:

    >>> stmt = select(Vertex).where(Vertex.start == Point(3, 4)).where(Vertex.end < Point(7, 8))
    >>> session.scalars(stmt).all()
    SELECT, vertices.x1, vertices.y1, vertices.x2, vertices.y2 FROM vertices WHERE vertices.x1 = ? AND vertices.y1 = ? AND vertices.x2 < ? AND vertices.y2 < ? [...] (3, 4, 7, 8)
    [Vertex(Point(x=3, y=4), Point(x=5, y=6))]

    New in version 2.0: composite() constructs now support “ordering” comparisons such as <, >=, and similar, in addition to the already-present support for ==, !=.


    The “ordering” comparison above using the “less than” operator (<) as well as the “equality” comparison using ==, when used to generate SQL expressions, are implemented by the Comparator class, and don’t make use of the comparison methods on the composite class itself, e.g. the __lt__() or __eq__() methods. From this it follows that the Point dataclass above also need not implement the dataclasses order=True parameter for the above SQL operations to work. The section Redefining Comparison Operations for Composites contains background on how to customize the comparison operations.

  • Updating Point objects on Vertex Instances

    By default, the Point object must be replaced by a new object for changes to be detected:

    >>> v1 = session.scalars(select(Vertex)).one()
    SELECT, vertices.x1, vertices.y1, vertices.x2, vertices.y2 FROM vertices [...] ()
    >>> v1.end = Point(x=10, y=14) >>> session.commit()
    UPDATE vertices SET x2=?, y2=? WHERE = ? [...] (10, 14, 1) COMMIT

    In order to allow in place changes on the composite object, the Mutation Tracking extension must be used. See the section Establishing Mutability on Composites for examples.

Other mapping forms for composites

The composite() construct may be passed the relevant columns using a mapped_column() construct, a Column, or the string name of an existing mapped column. The following examples illustrate an equivalent mapping as that of the main section above.

Map columns directly, then pass to composite

Here we pass the existing mapped_column() instances to the composite() construct, as in the non-annotated example below where we also pass the Point class as the first argument to composite():

from sqlalchemy import Integer
from sqlalchemy.orm import mapped_column, composite

class Vertex(Base):
    __tablename__ = "vertices"

    id = mapped_column(Integer, primary_key=True)
    x1 = mapped_column(Integer)
    y1 = mapped_column(Integer)
    x2 = mapped_column(Integer)
    y2 = mapped_column(Integer)

    start = composite(Point, x1, y1)
    end = composite(Point, x2, y2)

Map columns directly, pass attribute names to composite

We can write the same example above using more annotated forms where we have the option to pass attribute names to composite() instead of full column constructs:

from sqlalchemy.orm import mapped_column, composite, Mapped

class Vertex(Base):
    __tablename__ = "vertices"

    id: Mapped[int] = mapped_column(primary_key=True)
    x1: Mapped[int]
    y1: Mapped[int]
    x2: Mapped[int]
    y2: Mapped[int]

    start: Mapped[Point] = composite("x1", "y1")
    end: Mapped[Point] = composite("x2", "y2")

Imperative mapping and imperative table

When using imperative table or fully imperative mappings, we have access to Column objects directly. These may be passed to composite() as well, as in the imperative example below:

        "start": composite(Point, vertices_table.c.x1, vertices_table.c.y1),
        "end": composite(Point, vertices_table.c.x2, vertices_table.c.y2),

Using Legacy Non-Dataclasses

If not using a dataclass, the requirements for the custom datatype class are that it have a constructor which accepts positional arguments corresponding to its column format, and also provides a method __composite_values__() which returns the state of the object as a list or tuple, in order of its column-based attributes. It also should supply adequate __eq__() and __ne__() methods which test the equality of two instances.

To illustrate the equivalent Point class from the main section not using a dataclass:

class Point:
    def __init__(self, x, y):
        self.x = x
        self.y = y

    def __composite_values__(self):
        return self.x, self.y

    def __repr__(self):
        return f"Point(x={self.x!r}, y={self.y!r})"

    def __eq__(self, other):
        return isinstance(other, Point) and other.x == self.x and other.y == self.y

    def __ne__(self, other):
        return not self.__eq__(other)

Usage with composite() then proceeds where the columns to be associated with the Point class must also be declared with explicit types, using one of the forms at Other mapping forms for composites.

Tracking In-Place Mutations on Composites

In-place changes to an existing composite value are not tracked automatically. Instead, the composite class needs to provide events to its parent object explicitly. This task is largely automated via the usage of the MutableComposite mixin, which uses events to associate each user-defined composite object with all parent associations. Please see the example in Establishing Mutability on Composites.

Redefining Comparison Operations for Composites

The “equals” comparison operation by default produces an AND of all corresponding columns equated to one another. This can be changed using the comparator_factory argument to composite(), where we specify a custom Comparator class to define existing or new operations. Below we illustrate the “greater than” operator, implementing the same expression that the base “greater than” does:

import dataclasses

from sqlalchemy.orm import composite
from sqlalchemy.orm import CompositeProperty
from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import mapped_column
from sqlalchemy.sql import and_

class Point:
    x: int
    y: int

class PointComparator(CompositeProperty.Comparator):
    def __gt__(self, other):
        """redefine the 'greater than' operation"""

        return and_(
                a > b
                for a, b in zip(

class Base(DeclarativeBase):

class Vertex(Base):
    __tablename__ = "vertices"

    id: Mapped[int] = mapped_column(primary_key=True)

    start: Mapped[Point] = composite(
        mapped_column("x1"), mapped_column("y1"), comparator_factory=PointComparator
    end: Mapped[Point] = composite(
        mapped_column("x2"), mapped_column("y2"), comparator_factory=PointComparator

Since Point is a dataclass, we may make use of dataclasses.astuple() to get a tuple form of Point instances.

The custom comparator then returns the appropriate SQL expression:

>>> print(Vertex.start > Point(5, 6))
vertices.x1 > :x1_1 AND vertices.y1 > :y1_1

Nesting Composites

Composite objects can be defined to work in simple nested schemes, by redefining behaviors within the composite class to work as desired, then mapping the composite class to the full length of individual columns normally. This requires that additional methods to move between the “nested” and “flat” forms are defined.

Below we reorganize the Vertex class to itself be a composite object which refers to Point objects. Vertex and Point can be dataclasses, however we will add a custom construction method to Vertex that can be used to create new Vertex objects given four column values, which will will arbitrarily name _generate() and define as a classmethod so that we can make new Vertex objects by passing values to the Vertex._generate() method.

We will also implement the __composite_values__() method, which is a fixed name recognized by the composite() construct (introduced previously at Using Legacy Non-Dataclasses) that indicates a standard way of receiving the object as a flat tuple of column values, which in this case will supersede the usual dataclass-oriented methodology.

With our custom _generate() constructor and __composite_values__() serializer method, we can now move between a flat tuple of columns and Vertex objects that contain Point instances. The Vertex._generate method is passed as the first argument to the composite() construct as the source of new Vertex instances, and the __composite_values__() method will be used implicitly by composite().

For the purposes of the example, the Vertex composite is then mapped to a class called HasVertex, which is where the Table containing the four source columns ultimately resides:

from __future__ import annotations

import dataclasses
from typing import Any
from typing import Tuple

from sqlalchemy.orm import composite
from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import mapped_column

class Point:
    x: int
    y: int

class Vertex:
    start: Point
    end: Point

    def _generate(cls, x1: int, y1: int, x2: int, y2: int) -> Vertex:
        """generate a Vertex from a row"""
        return Vertex(Point(x1, y1), Point(x2, y2))

    def __composite_values__(self) -> Tuple[Any, ...]:
        """generate a row from a Vertex"""
        return dataclasses.astuple(self.start) + dataclasses.astuple(self.end)

class Base(DeclarativeBase):

class HasVertex(Base):
    __tablename__ = "has_vertex"
    id: Mapped[int] = mapped_column(primary_key=True)
    x1: Mapped[int]
    y1: Mapped[int]
    x2: Mapped[int]
    y2: Mapped[int]

    vertex: Mapped[Vertex] = composite(Vertex._generate, "x1", "y1", "x2", "y2")

The above mapping can then be used in terms of HasVertex, Vertex, and Point:

hv = HasVertex(vertex=Vertex(Point(1, 2), Point(3, 4)))


stmt = select(HasVertex).where(HasVertex.vertex == Vertex(Point(1, 2), Point(3, 4)))

hv = session.scalars(stmt).first()

Composite API

Object Name Description

composite([_class_or_attr], *attrs, [group, deferred, raiseload, comparator_factory, active_history, init, repr, default, default_factory, compare, kw_only, info, doc], **__kw)

Return a composite column-based property for use with a Mapper.

function sqlalchemy.orm.composite(_class_or_attr: None | Type[_CC] | Callable[..., _CC] | _CompositeAttrType[Any] = None, /, *attrs: _CompositeAttrType[Any], group: str | None = None, deferred: bool = False, raiseload: bool = False, comparator_factory: Type[Composite.Comparator[_T]] | None = None, active_history: bool = False, init: _NoArg | bool = _NoArg.NO_ARG, repr: _NoArg | bool = _NoArg.NO_ARG, default: Any | None = _NoArg.NO_ARG, default_factory: _NoArg | Callable[[], _T] = _NoArg.NO_ARG, compare: _NoArg | bool = _NoArg.NO_ARG, kw_only: _NoArg | bool = _NoArg.NO_ARG, info: _InfoType | None = None, doc: str | None = None, **__kw: Any) Composite[Any]

Return a composite column-based property for use with a Mapper.

See the mapping documentation section Composite Column Types for a full usage example.

The MapperProperty returned by composite() is the Composite.

  • class_ – The “composite type” class, or any classmethod or callable which will produce a new instance of the composite object given the column values in order.

  • *attrs

    List of elements to be mapped, which may include:

    • Column objects

    • mapped_column() constructs

    • string names of other attributes on the mapped class, which may be any other SQL or object-mapped attribute. This can for example allow a composite that refers to a many-to-one relationship

  • active_history=False – When True, indicates that the “previous” value for a scalar attribute should be loaded when replaced, if not already loaded. See the same flag on column_property().

  • group – A group name for this property when marked as deferred.

  • deferred – When True, the column property is “deferred”, meaning that it does not load immediately, and is instead loaded when the attribute is first accessed on an instance. See also deferred().

  • comparator_factory – a class which extends Comparator which provides custom SQL clause generation for comparison operations.

  • doc – optional string that will be applied as the doc on the class-bound descriptor.

  • info – Optional data dictionary which will be populated into the attribute of this object.

  • init – Specific to Declarative Dataclass Mapping, specifies if the mapped attribute should be part of the __init__() method as generated by the dataclass process.

  • repr – Specific to Declarative Dataclass Mapping, specifies if the mapped attribute should be part of the __repr__() method as generated by the dataclass process.

  • default_factory – Specific to Declarative Dataclass Mapping, specifies a default-value generation function that will take place as part of the __init__() method as generated by the dataclass process.

  • compare

    Specific to Declarative Dataclass Mapping, indicates if this field should be included in comparison operations when generating the __eq__() and __ne__() methods for the mapped class.

    New in version 2.0.0b4.

  • kw_only – Specific to Declarative Dataclass Mapping, indicates if this field should be marked as keyword-only when generating the __init__().