PyContracts

PyContracts is a Python package that allows to declare constraints on function parameters and return values. It supports a basic type system, variables binding, arithmetic constraints, and has several specialized contracts (notably for Numpy arrays).

As a quick intro, please see this presentation about PyContracts.

Why: The purpose of PyContracts is not to turn Python into a statically-typed language (albeit you can be as strict as you wish), but, rather, to avoid the time-consuming and obfuscating checking of various preconditions. In fact, more than the type constraints, I found useful the ability to impose value and size constraints. For example, "I need a list of at least 3 positive numbers" can be expressed as list[>=3](number, >0)). If you find that PyContracts is overkill for you, you might want to try a simpler alternative, such as typecheck. If you find that PyContracts is not enough for you, you probably want to be using Haskell instead of Python.

Specifying contracts: Contracts can be specified in three ways:

1. Using the ``@contract`` decorator: :

   @contract(a='int,>0', b='list[N],N>0', returns='list[N]')
   def my_function(a, b):
       ...
   

2. Using annotations (for Python 3): :

   @contract
   def my_function(a : 'int,>0', b : 'list[N],N>0') -> 'list[N]': 
        # Requires b to be a nonempty list, and the return 
        # value to have the same length.
        ...
   

3. Using docstrings, with the :type: and :rtype: tags: :

   @contract
   def my_function(a, b): 
       """ Function description.
           :type a: int,>0
           :type b: list[N],N>0
           :rtype: list[N]
       """
       ...
   

Deployment: In production, all checks can be disabled using the function contracts.disable_all(), so the performance hit is 0.

Extensions: You can extend PyContracts with new contracts types: :

new_contract('valid_name', lambda s: isinstance(s, str) and len(s)>0)
@contract(names='dict(int: (valid_name, int))')
def process_accounting(records):
    ...

Any Python type is a contract: :

@contract(a=int, # simple contract
          b='int,>0' # more complicated
          )
def f(a, b):
    ...

Enforcing interfaces: ContractsMeta is a metaclass, like ABCMeta, which propagates contracts to the subclasses: :

from contracts import contract, ContractsMeta, with_metaclass

class Base(with_metaclass(ContractsMeta, object)):

    @abstractmethod
    @contract(probability='float,>=0,<=1')
    def sample(self, probability):
        pass

class Derived(Base):
    # The contract above is automatically enforced, 
    # without this class having to know about PyContracts at all!
    def sample(self, probability):
        ....

Numpy: There is special support for Numpy: :

@contract(image='array[HxWx3](uint8),H>10,W>10')
def recolor(image):
    ...

Status: The syntax is stable and it won't be changed. PyContracts is very well tested on Python 2.x.

Status on Python 3.x: We reached feature parity! Everything works on Python 3 now.

Contributors:

• Chris Beaumont (Harvard-Smithsonian Center for Astrophysics): $var syntax; kwargs/args for extensions.
• Brett Graham (Rowland Institute at Harvard University): attr(name:type) syntax for checking types of attributes.
• William Furr: bug reports and performance improvements
• Karol Kuczmarski (Google Zurich): implementation of "string" and "unicode" contracts
• Maarten Derickx (Leiden U.): documentation fixes
• Calen Pennington (EdX): disabling checks inside check() function.
• Adam Palay (EdX): implementation of environment variable enabling/disabling override.
• Ryan Heimbuch: bug reports
• Bernhard Biskup: bug reports
• asharp: bug fixes
• Dennis Kempin (Google mothership): Sphinx-style constraints specs
• Andy Hayden: Python 3 support, more efficient Numpy checks
• Jonathan Sharpe: contracts for file-like objects, not operator

(Please let me know if I forgot anybody.)