sumtypes

sumtypes provides Algebraic Data Types for Python. The main benefit is the implementation of Sum Types (aka Tagged Unions), which Python doesn’t have any native representation for. Product Types are just objects with multiple attributes.

Documentation is at https://sumtypes.readthedocs.org/

This module uses the attrs library to provide features like attribute validation and defaults.

Example

Decorate your classes to make them a sum type:

import attr
from sumtypes import sumtype, constructor, match

@sumtype
class MyType(object):
    # constructors specify names for their arguments
    MyConstructor = constructor('x')
    AnotherConstructor = constructor('x', 'y')

    # You can also make use of any feature of the attrs
    # package by using attr.ib in constructors
    ThirdConstructor = constructor(
        one=attr.ib(default=42),
        two=attr.ib(validator=attr.validators.instance_of(int)))

(attrs package, and attr.ib documentation)

Then construct them by calling the constructors:

v = MyType.MyConstructor(1)
v2 = MyType.AnotherConstructor('foo', 2)

You can get the values from the tagged objects:

assert v.x == 1
assert v2.x == 'foo'
assert v2.y == 2

You check the constructor used:

assert type(v) is MyType.MyConstructor

And, like Scala case classes, the constructor type is a subclass of the main type:

assert isinstance(v, MyType)

And the tagged objects support equality:

assert v == MyType.MyConstructor(1)
assert v != MyType.MyConstructor(2)

Simple pattern matching is also supported. To write a function over all the cases of a sum type:

@match(MyType)
class get_number(object):
    def MyConstructor(x): return x
    def AnotherConstructor(x, y): return y
    def ThirdConstructor(one, two): return one + two

assert get_number(v) == 1
assert get_number(v2) == 2

match ensures that all cases are handled. If you really want to write a ‘partial function’ (i.e. one that doesn’t cover all cases), use match_partial.

See Also

Over the past few years, the ecosystem of libraries to help with functional programming in Python has exploded. Here are some libraries I recommend:

• effect - a library for isolating side-effects

• pyrsistent - persistent (optimized immutable) data structures in Python

• toolz - a general library of pure FP functions

• fn.py - a Scala-inspired set of tools, including a weird lambda syntax, option type, and monads