I am currently working on a project where we are running a large suite of parameterized tests (>1M). The tests are randomly generated use-cases and in this large test space, it is expected that in each run certain edge cases will fail, ~1-2%. Is there a implementation for Pytest where you can pass a failure-rate argument, or handle this behavior?
I guess what you want is modify exit status of pytest command, there is a nonpublic hook, named pytest_sessionfinish, could do this.
consider you have following tests:
def test_spam():
assert 0
def test_ham():
pass
def test_eggs():
pass
and a hook in conftest.py:
import pytest, _pytest
ACCEPTABLE_FAILURE_RATE = 50
#pytest.hookimpl()
def pytest_sessionfinish(session, exitstatus):
if exitstatus != _pytest.main.EXIT_TESTSFAILED:
return
failure_rate = (100.0 * session.testsfailed) / session.testscollected
if failure_rate <= ACCEPTABLE_FAILURE_RATE:
session.exitstatus = 0
then invoke pytest:
$ pytest --tb=no -q tests.py
F.. [100%]
1 failed, 2 passed in 0.06 seconds
here the failure rate is 1 / 3 == 33.3%, below 50%:
$ echo $?
0
you could see the exit status of pytest is 0.
Related
I am trying to execute a function decorated with Hypothesis' #strategy.composite decorator.
I know I can test functions using the #given decorator, such as -
from hypothesis import given
from hypothesis import strategies as st
#given(st.integers(min_value=1))
def test_bar(x):
assert x > 0
with pytest using - pytest <filename.py>.
But in the case of a function with the #strategy.composite decorator like -
from hypothesis import strategies as st
from hypothesis import given
import pytest
#st.composite
def test_foo(draw):
arg1 = draw(st.integers(min_value=1))
arg2 = draw(st.lists(st.integers(), min_size=arg1, max_size=arg1))
print(arg1, " ", arg2)
assert(len(arg2) == arg1)
I am unable to execute the tests in a similar way.
When using pytest I am unable to execute the tests (using python to execute the python file does nothing) -
[reik#reik-msi tests]$ pytest testhypo.py
==================================== test session starts ====================================
platform linux -- Python 3.8.3, pytest-5.4.3, py-1.8.1, pluggy-0.13.1
rootdir: /home/reik/tests
plugins: hypothesis-5.16.0, lazy-fixture-0.6.3
collected 1 item
testhypo.py F [100%]
========================================= FAILURES ==========================================
_________________________________________ test_foo __________________________________________
item = <Function test_foo>
#pytest.hookimpl(hookwrapper=True)
def pytest_runtest_call(item):
if not hasattr(item, "obj"):
yield
elif not is_hypothesis_test(item.obj):
# If #given was not applied, check whether other hypothesis
# decorators were applied, and raise an error if they were.
if getattr(item.obj, "is_hypothesis_strategy_function", False):
> raise InvalidArgument(
"%s is a function that returns a Hypothesis strategy, but pytest "
"has collected it as a test function. This is useless as the "
"function body will never be executed. To define a test "
"function, use #given instead of #composite." % (item.nodeid,)
)
E hypothesis.errors.InvalidArgument: testhypo.py::test_foo is a function that returns a Hypothesis strategy, but pytest has collected it as a test function. This is useless as the function body will never be executed. To define a test function, use #given instead of #composite.
/usr/lib/python3.8/site-packages/hypothesis/extra/pytestplugin.py:132: InvalidArgument
================================== short test summary info ==================================
FAILED testhypo.py::test_foo - hypothesis.errors.InvalidArgument: testhypo.py::test_foo is...
===================================== 1 failed in 0.06s =====================================
I tried adding the function call test_foo() but I got the same error.
Then I tried adding #given above the function and got a different error -
========================================== ERRORS ===========================================
________________________________ ERROR at setup of test_foo _________________________________
file /usr/lib/python3.8/site-packages/hypothesis/core.py, line 903
def run_test_as_given(test):
E fixture 'test' not found
> available fixtures: cache, capfd, capfdbinary, caplog, capsys, capsysbinary, doctest_namespace, monkeypatch, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory
> use 'pytest --fixtures [testpath]' for help on them.
/usr/lib/python3.8/site-packages/hypothesis/core.py:903
================================== short test summary info ==================================
ERROR testhypo.py::test_foo
if I do #given() - note the extra braces, instead, I get another error -
========================================= FAILURES ==========================================
_________________________________________ test_foo __________________________________________
arguments = (), kwargs = {}
def wrapped_test(*arguments, **kwargs):
> raise InvalidArgument(message)
E hypothesis.errors.InvalidArgument: given must be called with at least one argument
/usr/lib/python3.8/site-packages/hypothesis/core.py:234: InvalidArgument
================================== short test summary info ==================================
FAILED testhypo.py::test_foo - hypothesis.errors.InvalidArgument: given must be called wit...
I tried wrapping the code inside another function -
from hypothesis import strategies as st
from hypothesis import given
import pytest
def demo():
#st.composite
def test_foo(draw):
arg1 = draw(st.integers(min_value=1))
arg2 = draw(st.lists(st.integers(), min_size=arg1, max_size=arg1))
print(arg1, " ", arg2)
assert(len(arg2) == arg1)
but that did not work either -
[reik#reik-msi tests]$ python testhypo.py
[reik#reik-msi tests]$ pytest testhypo.py
==================================== test session starts ====================================
platform linux -- Python 3.8.3, pytest-5.4.3, py-1.8.1, pluggy-0.13.1
rootdir: /home/reik/tests
plugins: hypothesis-5.16.0, lazy-fixture-0.6.3
collected 0 items
=================================== no tests ran in 0.00s ===================================
(I also tried putting the demo function call at the end of the file, but that did not change the testing behaviour in any way)
The Hypothesis quick start guide says that calling the function would work, but it clearly does not. (To be fair, the documentation does not specify how to run tests with #composite)
How do I test the functions that are decorated with #strategy.composite? I do not have to use pytest - I would prefer not having to use it actually, but it seemed the easiest way to test the functions (which were decorated with #given) so I decided to go that route.
#st.composite is a helper function for defining custom strategies, not tests.
What you're trying to do can be accomplished by using #given and st.data():
#given(st.data())
def test_foo(data):
x = data.draw(st.integers())
...
https://hillelwayne.com/post/property-testing-complex-inputs/ gives a good overview of how these techniques are used.
I had the same issue. Here is a minimal and complete working example of using #composite.
#dataclass
class Container():
id: int
value: str
#st.composite
def generate_containers(draw):
_id = draw(st.integers())
_value = draw(st.characters())
return Container(id=_id, value=_value)
#given(generate_containers())
def test_container(container):
assert isinstance(container, Container)
assert isinstance(container.id, int)
assert isinstance(container.value, str)
My setup is such that; pytest test.py executes nothing while pytest --first-test test.py executes the target function test_skip.
In order to determine whether a certain test should be conducted or not, this is what I have been using:
skip_first = pytest.mark.skipif(
not (
pytest.config.getoption("--first-test")
or os.environ.get('FULL_AUTH_TEST')
), reason="Skipping live"
)
#skip_first
def test_skip():
assert_something
Now that, pytest.config.getoption is being deprecated, I am trying to update my code. This is what I have written:
#pytest.fixture
def skip_first(request):
def _skip_first():
return pytest.mark.skipif(
not (
request.config.getoption("--first-test")
or os.environ.get('FULL_AUTH_TEST')),
reason="Skipping"
)
return _skip_first()
# And, to call:
def test_skip(skip_first):
assert 1==2
However, whether I do pytest test.py or pytest --first-test test.py, test_skip will always execute. But, the skip_first seems to be working fine - Inserting a print statement shows skip_first = MarkDecorator(mark=Mark(name='skipif', args=(False,), kwargs={'reason': 'Skipping first'})), when --first-test is given, and args=(True) when given. (Same thing was observed when using the first setup).
Am I missing something?? I even tried to return the function _skip_first instead of it's output in the def skip_first but no difference.
When using a test class, the manual indicates we need to use #pytest.mark.usefixtures("fixturename") but that proved to be of no use either (with classes).
Ideas? This is my system: platform linux -- Python 3.6.7, pytest-4.0.2, py-1.7.0, pluggy-0.8.0
In order to cause a SKIP from a fixture, you must raise pytest.skip. Here's an example using your code above:
import os
import pytest
#pytest.fixture
def skip_first(request):
if (
request.config.getoption("--first-test")
or os.environ.get('FULL_AUTH_TEST')
):
raise pytest.skip('Skipping!')
# And, to call:
def test_skip(skip_first):
assert 1==2
If you want, you can almost replace your original code by doing:
#pytest.fixture
def skip_first_fixture(request): ...
skip_first = pytest.mark.usefixtures('skip_first_fixture')
#skip_first
def test_skip(): ...
Here's the execution showing this working:
$ pytest t.py -q
F [100%]
=================================== FAILURES ===================================
__________________________________ test_skip ___________________________________
skip_first = None
def test_skip(skip_first):
> assert 1==2
E assert 1 == 2
E -1
E +2
t.py:16: AssertionError
1 failed in 0.03 seconds
$ pytest t.py --first-test -q
s [100%]
1 skipped in 0.01 seconds
Dependent on the overall test result of a pytest test run I would like to execute conditional tear down. This means the access to the overall test result must happen after all tests have been executed but before the test runner has been left. How can I achieve this?
I could not find a suitable pytest hook to access the overall test result yet.
You don't need one; just collect the test results yourself. This is the blueprint I usually use when in need of accessing the test results in batch:
# conftest.py
import pytest
def pytest_sessionstart(session):
session.results = dict()
#pytest.hookimpl(tryfirst=True, hookwrapper=True)
def pytest_runtest_makereport(item, call):
outcome = yield
result = outcome.get_result()
if result.when == 'call':
item.session.results[item] = result
Now all test results are stored under session.results dict; example usage:
# conftest.py (continued)
def pytest_sessionfinish(session, exitstatus):
print()
print('run status code:', exitstatus)
passed_amount = sum(1 for result in session.results.values() if result.passed)
failed_amount = sum(1 for result in session.results.values() if result.failed)
print(f'there are {passed_amount} passed and {failed_amount} failed tests')
Running the tests will yield:
$ pytest -sv
================================== test session starts ====================================
platform darwin -- Python 3.6.4, pytest-3.7.1, py-1.5.3, pluggy-0.7.1 -- /Users/hoefling/.virtualenvs/stackoverflow/bin/python3.6
cachedir: .pytest_cache
rootdir: /Users/hoefling/projects/private/stackoverflow/so-51711988, inifile:
collected 3 items
test_spam.py::test_spam PASSED
test_spam.py::test_eggs PASSED
test_spam.py::test_fail FAILED
run status code: 1
there are 2 passed and 1 failed tests
======================================== FAILURES =========================================
_______________________________________ test_fail _________________________________________
def test_fail():
> assert False
E assert False
test_spam.py:10: AssertionError
=========================== 1 failed, 2 passed in 0.05 seconds ============================
In case the overall pytest exit code (exitstatus) is sufficient info (info about # passed, # failed, etc. not required) use the following:
# conftest.py
def pytest_sessionfinish(session, exitstatus):
print()
print('run status code:', exitstatus)
Accessing the error
You can access the error details from the call.excinfo object:
#pytest.hookimpl(tryfirst=True, hookwrapper=True)
def pytest_runtest_makereport(item, call):
outcome = yield
result = outcome.get_result()
if result.outcome == "failed":
exception = call.excinfo.value
exception_class = call.excinfo.type
exception_class_name = call.excinfo.typename
exception_type_and_message_formatted = call.excinfo.exconly()
exception_traceback = call.excinfo.traceback
When a test fails, there's an output indicating the context of the test, e.g.
=================================== FAILURES ===================================
______________________________ Test.test_sum_even ______________________________
numbers = [2, 4, 6]
#staticmethod
def test_sum_even(numbers):
assert sum(numbers) % 2 == 0
> assert False
E assert False
test_preprocessing.py:52: AssertionError
What if I want the same thing for passed tests as well? so that I can have a quick check on the parameters that get passed to the tests are correct?
I tried command line options line --full-trace, -l, --tb long, and -rpP, but none of them works.
Any idea?
Executing pytest with the --verbose flag will cause it to list the fully qualified name of every test as it executes, e.g.,:
tests/dsl/test_ancestor.py::TestAncestor::test_finds_ancestor_nodes
tests/dsl/test_and.py::TestAnd::test_aliased_as_ampersand
tests/dsl/test_and.py::TestAnd::test_finds_all_nodes_in_both_expressions
tests/dsl/test_anywhere.py::TestAnywhere::test_finds_all_nodes_when_no_arguments_given_regardless_of_the_context
tests/dsl/test_anywhere.py::TestAnywhere::test_finds_multiple_kinds_of_nodes_regardless_of_the_context
tests/dsl/test_anywhere.py::TestAnywhere::test_finds_nodes_regardless_of_the_context
tests/dsl/test_axis.py::TestAxis::test_finds_nodes_given_the_xpath_axis
tests/dsl/test_axis.py::TestAxis::test_finds_nodes_given_the_xpath_axis_without_a_specific_tag
If you are just asking for standard output from passed test cases, then you need to pass the -s option to pytest to prevent capturing of standard output. More info about standard output suppression is available in the pytest docs.
pytest doesn't have this functionality. What it does is showing you the error from the exception when an assertion fails.
A workaround is to explicitly include the information you want to see from the passing tests by using python's logging module and then use the caplog fixture from pytest.
For example one version of a func.py could be:
import logging
logging.basicConfig(level=logging.DEBUG)
log = logging.getLogger('my-even-logger')
def is_even(numbers):
res = sum(numbers) % 2 == 0
if res is True:
log.warning('Sum is Even')
else:
log.warning('Sum is Odd')
#... do stuff ...
and then a test_func.py:
import logging
import pytest
from func import is_even
#pytest.fixture
def my_list():
numbers = [2, 4, 6]
return numbers
def test_is_even(caplog, my_list):
with caplog.at_level(logging.DEBUG, logger='my-even-logger'):
is_even(my_list)
assert 'Even' in caplog.text
If you run pytest -s test_even.py and since the test passes, the logger shows you the following message:
test_even.py WARNING:my-sum-logger:Sum is Even
i run
nosetests -v --nocapture --nologcapture tests/reward/test_stuff.py
i get
----------------------------------------------------------------------
Ran 7 tests in 0.005s
OK
the tests without decorators run fine, however i have some fixture tests setup liike so which aren't running in the above command:
#use_fixtures(fixtures.ap_user_data, fixtures.ap_like_data, fixtures.other_reward_data)
def test_settings_are_respected(data):
"""
Tests setting takes effect
"""
res = func()
assert len(res) > 0
the decorator is:
def use_fixtures(*fixtures):
"""
Decorator for tests, abstracts the build-up required when using fixtures
"""
def real_use_fixtures(func):
def use_fixtures_inner(*args, **kwargs):
env = {}
for cls in fixtures:
name = cls.__name__
table = name[:-5] #curtails _data
env[table] = get_orm_class(table)
fixture = SQLAlchemyFixture(env=env, engine=engine, style=TrimmedNameStyle(suffix="_data"))
return fixture.with_data(*fixtures)(func)()
return use_fixtures_inner
return real_use_fixtures
is my use of decorators stopping nosetests from running my tests ?
I don't know if your decorator is causing nose to miss tests, but what if you take a test with a decorator, and remove the decorator long enough to see if it's still missed?