autosig 0.8.0

Define signatures to create beautiful APIs

Go straight to the documentation. Install with pip install autosig. Python 3 only.

Motivation

When I look at a great API I always observe a great level of consistency: similarly named and ordered arguments at a syntactic level; similar defaults, range of allowable values etc. on the semantic side. When looking at the code, one doesn’t see these regularities represented very explicitly.

Imagine we are starting to develop a library with three entry points, map, reduce and filter:

from collections import Iterable


def map(function, iterable):
    assert callable(function)
    assert isinstance(iterable, Iterable)
    return (function(x) for x in iterable)


def reduce(function, iterable):
    total = next(iterable)
    for x in iterable:
        total = function(total, x)
    return total


def filter(iterable, fun):
    if not isinstance(iterable, Iterable):
        iterable = [iterable]
    if isinstance(fun, set):
        fun = lambda x: x in fun
    return (x for x in iterable if fun(x))

But this is hardly well crafted. The order and naming of arguments isn’t consistent. One function checks its argument right away. The next doesn’t. The third attempts certain conversions to try and work with arguments that are not iterables or functions. There are reasons to build strict or tolerant APIs, but it’s unlikely that mixing the two within the same API is a good idea, unless it’s done deliberately (for instance offering a strict and tolerant version of every function). It wouldn’t be difficult to fix these problems in this small API but we would end up with duplicated logic that we need to keep aligned for the foreseeable future. Let’s do it instead the autosig way:

from autosig import param, Signature, autosig, check
from collections import Iterable


def to_callable(x):
    return (lambda y: y in x) if isinstance(x, set) else x


def to_iterable(x):
    return x if isinstance(x, Iterable) else [x]


API_signature = Signature(
    function=param(converter=to_callable, validator=check(callable)),
    iterable=param(converter=to_iterable, validator=check(Iterable)))


@autosig(API_signature)
def map(function, iterable):
    return (function(x) for x in iterable)


@autosig(API_signature)
def reduce(function, iterable):
    total = next(iterable)
    for x in iterable:
        total = function(total, x)
    return total


@autosig(API_signature)
def filter(function, iterable):
    return (x for x in iterable if function(x))

Let’s go through it step by step. First we defined 2 simple conversion functions. This is a good first step independent of autosig. Next we create a signature object, with two parameters. These are intialized with objects that define the checking and conversion that need to be performed on those parameters, independent of which function is going to use that signature. check creates a function that uses its argument, a Callable or a type, to validate an argument (the callable called with the argument as its own argument must return true; the type must be the type of the argument). Finally, we repeat the definition of our three API function, attaching the signature just defined with a decorator and then skipping all the checking and conversion logic and going straight to the meat of the function!

At the cost of a little code we have gained a lot:

• Explicit definition of the desired API signature, in a single place — DRY principle;

• association of that signature with API functions, checked at load time — no room for error;

• uniform application of conversion and validation logic without repeating it;

autosig is the pro tool for the API designer! If you want to take a look at a real package that uses autosig, check out altair_recipes.

Features

• Define reusable parameters with defaults, conversion and validation logic, documentation, preferred position in the signature and whether keyword-only.

• Define reusable signatures as ordered maps from names to parameters.

• Combine signatures to create complex ones on top of simple ones.

• Decorate functions and methods with their signatures. Enforced at load time. Conversion and validation logic executed at call time.

• Not hot about signatures? You can just use parameters as in:

  @autosig
  def reduce(function = param(...), iterable=param(...)):

  for more free-form APIs.

• Open source (BSD license)

• Extensive property-based testing, excellent coverage

Credits

This package is heavily based on attrs. While that may change in the future, for now it must be said this is a thin layer over that, with a bit of reflection sprinkled over. It is, I suppose, a quite original direction to take attrs into.

This package was created with Cookiecutter and the elgertam/cookiecutter-pipenv project template, based on audreyr/cookiecutter-pypackage.

History

0.8.0 (2019-08-27)

• autosig can decorate methods (exclude self from signature)

0.7.0 (2018-09-25)

• Argumentless autosig decorator for the use case of reusable parameter definitions but no reusable signatures.

0.6.0 (2018-09-24)

• Added check the quick validator generator. check(int) checks an argument is integer. check(\lambda x: x>0) checks an argument is positive. Behind the scenes it creates uses an assert statement which hopefully prints a useful message.

0.5.0 (2018-09-21)

• All new API, many breaking changes (sorry)

• signature decorator is gone

• create signatures directly withe the Signature constructor (it is no longer a base class to inherit from)

• do not use inheritance to define new signatures form old ones. It was a dead end as far as controlling the order of arguments. Use instead the + operator to combine two signatures, analogous to inheriting from one while adding new attributes.

• the new approach gives control over order of arguments, allows to mix mandatory and default arguments in one signature yet allow to reuse it (“stick” new mandatory arguments in between the arguments of the old signature)

0.4.1 (2018-09-05)

• Close abstraction holes revealing dependency on attr (which is gratefully acknowledged, but could be confusing).

0.3.1 (2018-08-30)

• Improved docstring generation

0.3.0 (2018-08-30)

• Compose docstring from param docstrings

0.2.3 (2018-08-28)

• Better and passing tests.

0.2.2 (2018-08-27)

• More stringent enforcement of signatures including defaults. Fixed build.

0.1.0 (2018-04-25)

• First release on PyPI.