Super-duper low cognitive overhead matching for Python developers reading or writing tests. Implemented fully in modern & typed Python, without any dependencies. Runs and passes its tests on most Pythons after 3.6. PyPy works fine too.
pychoir has mostly been developed for use with pytest, but nothing prevents from using it in any other test framework (like vanilla unittest) or even outside of testing, if you feel like it.
- With pip:
pip install pychoir - With pipenv:
pipenv install --dev pychoir - With poetry:
poetry add --dev pychoir - With PDM:
pdm add -dG test pychoir - With uv:
uv add --dev pychoir
Check out the API Reference on readthedocs for detailed info and examples of all the available Matchers https://pychoir.readthedocs.io/en/stable/api.html
You have probably written quite a few tests where you assert something like
assert thing_under_test() == {'some_fields': 'some values'}However, sometimes you do not expect exact equivalence. So you start
result = thing_under_test()
result_number = result.pop('number', None)
assert result_number is None or result_number < 3
result_list_of_strings = result.pop('list_of_strings', None)
assert (
result_list_of_strings is not None
and len(result_list_of_strings) == 5
and all(isinstance(s, str) for s in result_list_of_strings)
)
assert result == {'some_fields': 'some values'}…but this is not very convenient for anyone in the long run.
This is where pychoir comes in with matchers:
from pychoir import LessThan, All, HasLength, IsNoneOr, IsInstance
assert thing_under_test() == {
'number': IsNoneOr(LessThan(3)),
'list_of_strings': HasLength(5) & All(IsInstance(str)),
'some_fields': 'some values',
}It can also be cumbersome to check mocked calls without using matchers:
assert mock.call_args[0][0] < 3
assert isinstance(mock.call_args[0][1], str)
assert len(mock.call_args[0][2]) == 3but simple and easy when using them:
from pychoir import LessThan, IsInstance, HasLength
mock.assert_called_with(LessThan(3), IsInstance(str), HasLength(3))You can also check many things about the same value: for example IsInstance(int) & 5 will make sure that the value is not only equal to 5, but is also an int (goodbye to accidental 5.0).
You can place a matcher almost anywhere where a value can be. pychoir matchers work well inside lists, tuples, dicts, dataclasses, mock call assertions… You can also place normal values inside matchers, and they will match as with traditional == or !=.
A core principle is that pychoir Matchers are composable and can be used freely in various combinations. For example [LessThan(3) | 5] is “equal to” a list with one item, holding a value equal to 5 or any value less than 3.
Yes, you can! pychoir Matcher baseclass has been designed to be usable by code outside the library. It also takes care of most of the generic plumbing, so your custom matcher typically needs very little code.
Here is the implementation of IsInstance as an example:
from typing import Any, Type
from pychoir import Matcher
class IsInstance(Matcher):
def __init__(self, type_: Type[Any]):
super().__init__()
self.type = type_
def _matches(self, other: Any) -> bool:
return isinstance(other, self.type)
def _description(self) -> str:
return self.type.__name__All you need to take care of is defining the parameters (if any) in __init__(), the match itself in _matches(), and a description of the parameters in _description().
Here is an even simpler Anything matcher that does not take parameters and matches literally anything:
from typing import Any
from pychoir import Matcher
class Anything(Matcher):
def _matches(self, _: Any) -> bool:
return True
def _description(self) -> str:
return ''If your custom matcher is generic enough to be useful for everyone, please contribute (fork and make a pull request for now) and have it included in pychoir!
Nothing wrong with hamcrest as such, but pychoir aims to be better integrated with natural Python syntax, meaning for example that you do not need to use a custom assert function. pychoir matchers are drop-in replacements for your normal values alone or inside structures, even deeply nested ones. You can use hamcrest matchers through pychoir if you like, wrapping them in the Matches(my_hamcrest_matcher) matcher, although the idea is that pychoir would soon come with an equivalent set of matchers.
What a nice fluent API for matching, allowing matching multiple things at once. However, you can only match one value at a time. With pychoir you’ll be matching the whole result at once, be it a single value, list, tuple, dict, dataclass, you name it. Let’s see if pychoir gets some of that fluent stuff going forward as well.
I’d be happy to hear from you about other similar libraries.
Python has a rather peculiar way of handling equivalence, which allows customizing it in wild and imaginative ways. This is a very powerful feature, which one should usually avoid overusing. pychoir is built around the idea of using this power to build a lean and mean matcher implementation that looks like a custom DSL but is actually completely standard Python 3.
pychoir has quite a vast range of Matchers built in as well as basic API Reference documenting them. New ideas are still plenty and more can be discussed in Discussions. Documentation will receive updates as well. Most remarkably fancy examples are missing. Making pychoir easier to contribute to is also on the list.
It comes from the French word pochoir which means a drawing technique using templates. For some reason this method of matching in tests reminds me of drawing with those. A French word was chosen because it happens to start with a p and a vowel 😉
