pytest-relaxed

pytest-relaxed provides 'relaxed' test discovery for pytest.

It is the spiritual successor to https://pypi.python.org/pypi/spec, but is built for pytest instead of nosetests, and rethinks some aspects of the design (such as increased ability to opt-in to various behaviors.)

For a development roadmap, see the maintainer's roadmap page.

Rationale

Has it ever felt strange to you that we put our tests in tests/, then name the files test_foo.py, name the test classes TestFoo, and finally name the test methods test_foo_bar? Especially when almost all of the code inside of tests/ is, well, tests?

This pytest plugin takes a page from the rest of Python, where you don't have to explicitly note public module/class members, but only need to hint as to which ones are private. By default, all files and objects pytest is told to scan will be considered tests; to mark something as not-a-test, simply prefix it with an underscore.

Relaxed discovery

The "it's a test by default unless underscored" approach works for files:

tests
    ├── _util.py
    ├── one_module.py
    └── another_module.py

It's applied to module members:

def _helper():
    pass

def one_thing():
    assert True

def another_thing():
    assert False

def yet_another():
    assert _helper() == 'something'

And to class members:

class SomeObject:
    def behavior_one(self):
        assert True

    def another_behavior(self):
        assert False

    def _helper(self):
        pass

    def it_does_things(self):
        assert self._helper() == 'whatever'

Special cases

As you might expect, there are a few more special cases around discovery to avoid fouling up common test extensions:

• Files named conftest.py aren't treated as tests, because they do special pytest things;
• Module and class members named setup_(module|class|method|function) are not considered tests, as they are how pytest implements classic/xunit style setup and teardown;
• Objects decorated as fixtures with @pytest.fixture are, of course, also skipped.

Backwards compatibility

If you like the idea of pytest-relaxed but have a large test suite, it may be daunting to think about "upgrading" it all in one go. It's relatively simple to arrive at a 'hybrid' test suite where your legacy tests still run normally (as long as they're already pytest-compatible, which is true for most unittest suites) but 'relaxed' style tests also work as expected.

• The only change you'll still have to make is renaming 'helper' files (any whose name doesn't start with test_) so their names begin with an underscore; then, of course, search and replace any imports of such files.

• pytest-relaxed explicitly sidesteps around anything that looks like "classic" test files (i.e. named test_*), allowing pytest's native collection to take effect. Such files should not need any alteration.

• Our reporter (display) functionality still works pretty well with legacy style tests; test prefixes and suffixes are stripped at display time, so TestMyThing.test_something still shows up as if it was written in relaxed style: MyThing w/ nested something.

  • However, because we don't collect such tests, nesting and other features we offer won't work until you've renamed the files to not start with test_, and changed any classes to not inherit from unittest.TestCase or similar.

Nested class organization

On top of the relaxed discovery algorithm, pytest-relaxed also lets you organize tests in a nested fashion, again like the spec nose plugin or the tools that inspired it, such as Ruby's rspec.

This is purely optional, but we find it's a nice middle ground between having a proliferation of files or suffering a large, flat test namespace making it hard to see which feature areas have been impacted by a bug (or whatnot).

The feature is enabled by using nested/inner classes, like so:

class SomeObject:
    def basic_behavior(self):
        assert True

    class init:
        "__init__"

        def no_args_required(self):
            assert True

        def accepts_some_arg(self):
            assert True

        def sets_up_config(self):
            assert False

    class some_method:
        def accepts_whatever_params(self):
            assert False

        def base_behavior(self):
            assert True

        class when_config_says_foo:
            def it_behaves_like_this(self):
                assert False

        class when_config_says_bar:
            def it_behaves_like_this(self):
                assert True

Test discovery on these inner classes is recursive, so you can nest them as deeply as you like. Naturally, as with all Python code, sometimes you can have too much of a good thing...but that's up to you.

Note

If writing Python-2-old-style classes makes you uncomfortable, you can write them as e.g. class SomethingUnderTest(object): - pytest-relaxed doesn't actually care. This is (naturally) moot under Python 3.

Nested class attributes

If you're namespacing your tests via nested classes, you may find yourself wanting to reference the enclosing "scope" of the outer classes they live in, such as class attributes. pytest-relaxed automatically copies such attributes onto inner classes during the test collection phase, allowing you to write code like this:

class Outer:
    behavior_one = True

    def outer_test(self):
        assert self.behavior_one

    class Inner:
        behavior_two = True

        def inner_test(self):
            assert self.behavior_one and self.behavior_two

Notably:

• The behavior is nested, infinitely, as you might expect;
• Attributes that look like test classes or methods themselves, are not copied (though others, i.e. ones named with a leading underscore, are);
• Only attributes _not_ already present on the inner class are copied; thus inner classes may naturally "override" attributes, just as with class inheritance.

Other test helpers

pytest-relaxed offers a few other random lightweight test-related utilities that don't merit their own PyPI entries (most ported from spec), such as:

• trap, a decorator for use on test functions and/or test helpers/subroutines which is similar to pytest's own capsys/capfd fixtures in that it allows capture of stdout/err.

  • It offers a slightly simpler API: it replaces sys.(stdout|stderr) with IO objects which can be getvalue()'d as needed.
  • More importantly, it can wrap arbitrary callables, which is useful for code-sharing use cases that don't easily fit into the design of fixtures.

• raises, a wrapper around pytest.raises which works as a decorator, similar to the Nose testing tool of the same name.

Nested output display

Continuing in the "port of spec / inspired by RSpec and friends" vein, pytest-relaxed greatly enhances pytest's verbose display mode:

• Tests are shown in a nested, tree-like fashion, with 'header' lines shown for modules, classes (including nested classes) and so forth.
• The per-test-result lines thus consist of just the test names, and are colorized (similar to the built-in verbose mode) based on success/failure/skip.
• Headers and test names are massaged to look more human-readable, such as replacing underscores with spaces.

Unlike spec, this functionality doesn't affect normal/non-verbose output at all, and can be disabled entirely, allowing you to use the relaxed test discovery alongside normal pytest verbose display or your favorite pytest output plugins (such as pytest-sugar.)

Installation & use

As with most pytest plugins, it's quite simple:

• pip install pytest-relaxed;

• Tell pytest where your tests live via the testpaths option; otherwise pytest-relaxed will cause pytest to load all of your non-test code as tests!

• Not required, but strongly recommended: configure pytest's default filename pattern (python_files) to be an unqualified glob (*).

  • This doesn't impact (our) test discovery, but pytest's assertion 'rewriting' (the feature that turns assert var == othervar into assert 17 == 2 during error display) reuses this setting when determining which files to manipulate.

• Thus, a recommended setup.cfg (or pytest.ini, sans the header) is:

  [tool:pytest]
  testpaths = tests
  python_files = *
  

• Write some tests, as exampled above;

• pytest to run the tests, and you're done!