Skip to main content

Extensible methods a la Haskell's typeclasses.

Project description

Extensible methods for Python mimicking typeclasses in Haskell.

Build status: Linux and OSX Build status: Windows Latest PyPI version Python versions supported


Some statically typed languages have ad hoc polymorphism where a function can have multiple implementations depending on the types of its arguments. In languages like C++ and Java, it is called function overloading. In Haskell, it is accomplished with type classes.

Consider an example of writing a toJson function in C++. The function takes a single value and returns a string, but it must be implemented differently for each different type of value:

std::string toJson(int i);
std::string toJson(double d);
std::string toJson(std::string s);

Some implementations may be “recursive”, and call the implementation for another type:

template <typename T>
std::string toJson(std::vector<T> const& xs) {
    for (T const& x : xs) {
        ... toJson(x) ...

Many dynamically typed languages, like Python and JavaScript, lack ad hoc polymorphism in the language, but developers can implement it by hand by inspecting the argument types and dispatching to implementations accordingly:

def to_json(value):
    if isinstance(value, int):
        return ...
    if isinstance(value, float):
        return ...
    if isinstance(value, str):
        return ...
    if isinstance(value, list):
        return '[' + ','.join(to_json(x) for x in value) + ']'

In addition to being a little uglier, this technique suffers from a limitation: once we’ve defined the function, we can’t add any more overloads. Imagine we want to define a JSON serialization for our user-defined type:

from ... import to_json

class Person:
    name: str

def to_json_person(person):
    return f'{{"name":{to_json(}}}'

While this example works for serializing Person, we won’t be able to serialize a list of Person because the implementation of to_json for list won’t call to_json_person.

Type Classes

In many languages, e.g. C++ and Java, two functions with the same name but different types are called overloads of the name. In Haskell, these overloads are not permitted: no two functions (or any other values for that matter) can have the same name in the same scope. However, type classes offer a way around this limitation.

A type class in Haskell is a group of polymorphic functions, called methods, parameterized by a single type variable. The type class only needs to declare the method signatures; it does not need to provide any definitions.

An instance for a type class defines all the methods of the type class for a specific type argument in the place of the type variable. In other words, a type class has exactly one polymorphic declaration, but many monomorphic instances, one for every possible type argument. Thus, a method can have many definitions (i.e. implementations), one from each instance, which means it can be overloaded.

At a method call site, how does Haskell know which overload, from which instance, to use? Haskell requires that the signature of the method in the type class declaration mentions the type variable in one of its parameters or its return type. It tries to unify that polymorphic declaration signature with the call site to fill in the type variable; if it succeeds, then it selects the monomorphic instance for that type argument.


How can we replicate type classes in Python?

Decorate a method signature with a call to typeclass, giving it the name of a type variable. The decorator will check the signature to make sure that the type variable appears at least once in the type annotations of the parameters. Unlike Haskell, Python cannot infer the return type at a call site, so that path to instance discovery is impossible; the type variable must be used as the type of at least one parameter.

T = typing.TypeVar('T')
def to_json(value: T) -> str:
    """Serialize a value to JSON."""

We may optionally provide a default implementation. If we do not, the default behavior is to raise a NotImplementedError diagnosing a missing instance for the specific type variable.

The typeclass decorator will add an instance attribute to the method. Use that to decorate monomorphic implementations, giving it the type argument:

def _to_json_str(s):
    return f'"{s}"'

We can decorate an implementation multiple times if it can serve multiple instances:

def _to_json_number(n):
    return str(n)

We can define an implementation for all types structurally matching a protocol. Because it is presently impossible to infer the difference between a protocol and a type, we must differentiate it for the decorator:

@to_json.instance(typing.Iterable, protocol=True)
def _to_json_iterable(xs):
   return '[' + ','.join(to_json(x) for x in xs) + ']'

If a type argument matches multiple protocols, the instance that was first defined will be chosen.

Now we can define instances for types whether we defined the type or imported it.

def _to_json_person(person):
    return f'{{"name":{to_json(}}}'
>>> to_json([Person(name='John')])

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

typeclasses-0.2.0.tar.gz (10.4 kB view hashes)

Uploaded source

Built Distribution

typeclasses-0.2.0-py3-none-any.whl (9.3 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page