PySingature is a python package that provides a set of utilities for simple function parameter typechecking, without modifying the target function and with minimal boilerplate or impact on performance.

PySignature provides two main components: a small set of classes that represent type assertions and a decorator for defining the function's signature.

Let's take a look at a (very) simple function:

def fn(x, y):
    return x * y

We can add parameter typechecking using the PySignature in the following way:

from pysignature import typechecked
from pysignature.types import Numeric

@typechecked(x=Numeric, y=Numeric)
def fn(x, y):
    return x + y

Now if we call fn with non-numeric types (neither ints nor floats) we get an exception FunctionTypeCheckError with detailed information about the wrong parameters in its property errors.

We can go beyond numeric validation and use complex type assertions provided by this module, or we can use any callable that uppon error returns a TypeError (such as float).

To expand our previous example, let's consider the fact that the * operator works if the left operand is a list or a string and the right operand is an integer (Numeric accepts floats). We can make use of built-in function int and PySignature's List, Or, List and Any type assertions.

from pysignature import typechecked
from pysignature.types import String, Or, List, Any

@typechecked(x=Or(String, List(Any)), y=int)
def fn(x, y):
    return x + y

PySignature supports typechecking for *args and **kwargs argument types. The only limitation is that they all must conform to a single type. They are specified in the following way:

@typechecked(x=String, _variadic=Numeric, _named=List(List(Any)))
def fn(x, *extra, **options):
    # ... Do something
    return

Each function decorated with pysignature.typechecked gets a property named untyped that ignores the typechecking functionality. In our original example, that would mean we can use fn.untyped(None, 1).

PySignature was created to cover a very specific need: allow for robust handling of parameters coming from the external world (the internet) without burdening the data scientists at Intelimétrica with the fact that their functions might be called with unsafe arguments.

The idea is to use PySignature in the (normally) small subset of functions that are exposed from a certain module to the outside world, so the mantainers of the function can develop without extra cognitive burden and the clients of such function can use them with extra confidence and robustness.

PySignature is not (neither tries to be) an extension to Python's type system. It does not attempt to bring full static typying to the languge and while it is not slow, it is not recommended for functions that are too be called in very tight loops.

Simply install via pip with:

$ pip install pysignature

Or, directly download the package and run

$ python setup.py install

Yes please! Help us with documentation, reporting bugs, implementing new type assertions or improving code

The development requirements are minimal. The only consideration is that we use pytest for testing.

PySignature uses the Apache License 2.0.