pyguarantees 0.0.13

Guarantee unittests for functions and methods, guarantee types and their properties

pyguarantees

This package has two components:

• test_guarantees: guarantee unittest-coverage for functions and methods.
• functional_guarantees: guarantee types and other properties for parameters and return values of callables—functions and methods.

test_guarantees

"I will have to write a unittest for this function later", you say. This package ensures that you won't forget.

"Why does this function fail? I've tested it... omg I didn't even call it in my TestCase." Use this package to make sure that a function/method will be called or a class instance constructed in its respective TestCase.

Example

import unittest
from src.pyguarantees import test_guarantees as tg
from some_package import some_fct_with_test_guarantee


class ExampleClass:
  @tg.guarantee_test()  # tg.main will raise exception if there is not test for this method
  def foo(self):
    return self

  @classmethod  # works for classmethods
  @tg.guarantee_test()  # @tg.guarantee_usage possible in any of these methods, but optional
  def class_method(cls):
    return cls

  @staticmethod  # works for staticmethods
  @tg.guarantee_test()
  def static_method():
    return "static!"


@tg.guarantee_test()
@tg.guarantee_usage()  # make sure that this is called in all its tests
def add_one(a):
  return a + 1


@tg.guarantee_test()
@tg.guarantee_usage()  # Makes sure that __init__ is called in the test
class RegularClass:
  def __init__(self):
    self.x = 2


class ExampleTest(unittest.TestCase):
  @tg.implements_test_for(
    ExampleClass.foo,
    ExampleClass.class_method,
    ExampleClass.static_method
  )
  def test_foo(self):
    ExampleClass().foo()
    ExampleClass.class_method()
    ExampleClass.static_method()

  @tg.implements_test_for(add_one, some_fct_with_test_guarantee)
  def test_other(self):
    val = add_one(1)
    self.assertEqual(val, 2)

  @tg.implements_test_for(RegularClass)
  def test_regular_class(self):
    regular_class = RegularClass()
    ...


if __name__ == '__main__':
  tg.main()

As in the example, pyguarantees.test_guarantees will be abbreviated with tg from here on out.

Failing to use an @tg.implements_test_for for a function or method decorated with @tg.guarantee_test leads to a tg.exceptions.TestsNotImplementedError, while failing to use this function or method in the corresponding test will lead to a tg.exceptions.NotUsedInTestsError if it is decorated by @tg.guarantee_usage. These exceptions are only raised if the unittest.TestCase are called first and then checked by tg.enforce, or tg.main is called to do both automatically.

Currently doesn't work with nested functions (defined inside of other callables). This might be fixed at some point.

The package consists of three decorators and two functions, as well as two Exceptions. All are explained below.

Decorators

The three decorators shown below have no effect without the functions of this package.

guarantee_test

Takes no arguments.

Any function, method, or class (except, for the moment, ones nested inside of other callables) decorated with @tg.guarantee_test that is in the scope of unittest will force unittest to throw an exception should it not be in an @tg.implements_test_for.

Currently, it is necessary to include the brackets—()— so that the function is registered. This executes the decorator once but not the callable that it decorates, making it computationally inexpensive.

Having a function (or method) decorated like follows:

@tg.guarantee_test()
def foo():
    pass

but not having a test in your unittest.TestCase decorated by @implements_test_for(foo) would lead to a TestsNotImplementedError being raised.

The same works for classes.

guarantee_usage

Takes no arguments.

Must be used below @tg.guarantee_test, otherwise it is ignored.

Just like with @tg.guarantee_test, brackets are not optional, but the execution of the decorator is computationally inexpensive.

A function decorated as follows:

@tg.guarantee_test()
@tg.guarantee_usage()
def foo():
    pass

with a unittest that looks something like this:

class TestExample(unittest.TestCase):
    @tg.implements_test_for(foo)
    def test_foo(self):
        ...   # some code that doesn't call foo

would lead to a NotUsedInTestsError being raised.

In this scenario, if foo is an argument in several @tg.implements_test_for, @tg.guarantee_usage makes certain that foo is used in every test-function decorated in such a way.

Special case: For classes, @tg.guarantee_usage guarantees that __init__ is called. For a callable class, this still holds; to guarantee that the __call__-method is called in the test, it has to be decorated by @tg.guarantee_usage itself, not the class it belongs to.

implements_test_for

• args: Give any function or method that the corresponding test is meant for.
• kwargs: The value will be used like an arg, while the key will be ignored.

Functions and methods that weren't decorated by @tg.guarantee_test lead to a user-warning but are ignored otherwise.

Usage might look as follows:

class TestExample(unittest.TestCase):
    @implements_test_for(function1, function2, this_key_is_ignored=function3)
    def test_example(self):
        ...

Functions

Two functions are provided by pyguarantees.test_guarantees, both directly under tg. At least one has to be used for the decorators to have an effect.

enforce

Takes no arguments.

Run this after running all your unittests. This runs additional unittests that check which functions violated their guarantees and raise exceptions accordingly.

It is recommended to only use this function when using a complicated unittest-setup. When using unittest.main(), it is recommended to use tg.main() instead.

main

Takes no arguments.

Calls unittest.main() followed by tg.enforce.

Exceptions

Exceptions are located under tg.exceptions.

There are two custom Exceptions as presented below.

TestsNotImplementedError

Arguments of tg.exceptions.TestsNotImplementedError:

• functions (type callable): The callables that weren't mentioned in a @tg.implements_test_for.

Members of tg.exceptions.TestsNotImplementedError:

• functions (type callable): The callables that weren't mentioned in a @tg.implements_test_for.
• description (type str): The error string printed when the exception is raised and not caught.

The output of raising this exception might look something like:

<Traceback...>

pyguarantees.test_guarantees.exceptions.TestsNotImplementedError: 

    No tests were implemented for the following methods and functions: 

    1. Missing test-case for the following callable: 
        Name: 		foo
        Module: 	__main__
    2. Missing test-case for the following callable: 
        Name: 		bar
        Module: 	__main__

NotUsedInTestsError

Arguments of tg.exceptions.NotUsedInTestsError:

• functions (type: callable): The callables that were mentioned in a @tg.implements_test_for but not used in the corresponding test.

Members of tg.exceptions.NotUsedInTestsError:

• functions (type: callable): The callables that were mentioned in a @tg.implements_test_for but not used in the corresponding test.
• description (type: str): The error string printed when the exception is raised and not caught.

A possible error message might look like the following:

<Traceback...>

pyguarantees.test_guarantees.exceptions.NotUsedInTestsError:

The following objects were not used in their assigned tests: 

1. The following callable was not called in its assigned tests: 
	Name: 		foo
	Module: 	__main__
	This callable is tested but not called in the following test-cases: 
				- Name: 	TestFoo.test_foo1
				   Module: 	__main__
				- Name: 	TestFoo.test_foo2
				   Module: 	__main__
2. The following callable was not called in its assigned tests: 
	Name: 		bar
	Module: 	some_module
	This callable is tested but not called in the following test-cases: 
				- Name: 	TestBar.test_bar
				   Module: 	test_some_module

functional_guarantees

Few things are more useful in programming than the ability to constrain a program's possible behaviors and communicate those constraints clearly in code. Statically typed languages do this with types, scope modifiers, and lifetime modifiers, among others (int, static, private, const, etc.). These are static constraints—or static guarantees—in that they are evaluated statically, before runtime.

Oftentimes, a program also has dynamic guarantees, evaluated during runtime—assertions, for example. A function dealing with division, for example, has to deal with the special case of division by zero.

pyguarantees.functional_guarantees, abbreviated with fg from here on out, enables both types of guarantees to be defined in Python where they should happen: function (or method) signatures. This is where statically typed languages put their static guarantees (a typical function signature looks something like scope-etc-modifiers return-type function-name(parameter-type parameter-name)) and where in my opinion, dynamic guarantees belong as well.

This might have the following advantages:

• Make code more readable by having constraints in a predefined place.
• Make code easier to write by providing important information about APIs in a glancable way.
• Make it possible to include information on dynamic constraints in automatically generated documentation.
• Encourage programmers to think about these constraints while writing the functions—a type of test-driven development directly at the function (seeing parts of the "tests" in the function-signature might assist readability of code, as well).

This package is an attempt to open up at least some of these advantages to Python-users at least partially, given the constraints of the Python-language.

Example 1

This example shows basic usage of the fg package for parameters of functions.

import numpy as np
from pyguarantees import functional_guarantees as fg

from your_module import your_custom_error_callback


# One of many built-in guarantees using one of many built-in options
@fg.add_guarantees(param_guarantees=[fg.IsInt("num", minimum=3)])
def add_one(num):
    return num + 1


# Use fg.IsClass to guarantee all types and classes that don't have specific guarantees 
#  in fg. If they do, it is recommended to use those specific guarantees.
@fg.add_guarantees(
    param_guarantees=[
        fg.IsClass(
            "X",
            class_type=np.ndarray,
            dynamic_checks=[
                fg.DynamicCheck(check=lambda x: x.min() > 0, description="min: 0"),
                fg.DynamicCheck(check=lambda x: x.var() < 5, description="var < 5"),
                fg.DynamicCheck(check=lambda x: x.shape == (3, 80, 80), description="shape (3, 80, 80")
            ],
            error_callback=your_custom_error_callback
        ),
        fg.IsClass("mean", class_type=np.ndarray),
        fg.IsClass("std", class_type=np.ndarray)
    ],
    return_guarantee=fg.IsClass("", class_type=np.ndarray)
)
def normalize(X, mean, std):
    return (X - mean) / std

Example 2

This example shows the use of the fg-package for testing.

import unittest
from pyguarantees import functional_guarantees as fg


@fg.add_guarantees(return_guarantee=fg.IsInt("a"))
def foo(a):
    return a 


@fg.add_guarantees(return_guarantee=fg.IsInt("a"))
def bar(a):
    return a 


class TestExample(unittest.TestCase):
    def test_example(self):
        # chain functions and automatically raise exceptions if at any point
        #   the guarantees for parameters and/or return values aren't fulfilled.
        foo(bar(1))   

add_guarantees

Add guarantees to a callable.

Arguments:

• param_guarantees A list of Guarantees that will work on the parameters of the callable decorated by add_guarantees. Must fulfill the following properties:
  • Be in the order that the arguments are defined in the callable-signature. If you have for example a function def foo(a, b), where a should be an int and b a str, param_guarantees has to be something like [fg.IsInt("a"), fg.IsStr("b").
• return_guarantee A single Guarantee on the return values.
• is_staticmethod Needed to deal with staticmethods (classmethod and regular methods work automatically).

Guarantees

There are several pre-defined guarantees. Most are based on the TypeGuarantee.

Guarantee

The base class of all guarantees in fg.

TypeGuarantee

this README is currently under development. More is coming.