remember/return returnvalues of test functions

[pytestbot]

Originally reported by: Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt)

---

this would come in handy for things like the twisted integration,
which currently has to copy the pyfunc_call hook of pytest to get the function return values in order to block on a defered

an alternative would be to expose the "call a test function" machinery in a way that allows to use the returnvalue without copying the machinery

---

• Bitbucket: https://bitbucket.org/pytest-dev/pytest/issue/207

[Droogans]

@nicoddemus you appear to be authoring 100% of the recent commits on master.

When would something like this ever become a generally accepted best practice in the project? I'd like to be able to return arbitrary objects after I've finished writing several assert statements, and have that returned object appear later in hook function arguments (item perhaps?).

[RonnyPfannschmidt]

the main problem with this one is that it requires some refactorings in the internals that aren't clearly defined yet

[nicoddemus]

Hi @Droogans,

Right now we don't have plans to add this feature to the core, but I you can easily do it yourself by attaching an object to request.node, which is actually an "item", something like:

def test(request):
    obj = SomeObject()
    request.node.obj = obj

Also to be honest copying the pytest_pyfunc_call does not seem too bad, here it is after cleaning it up a bit by removing support for yield tests:

@hookimpl(trylast=True)
def pytest_pyfunc_call(pyfuncitem):
    funcargs = pyfuncitem.funcargs
    testargs = {arg: funcargs[arg] arg in pyfuncitem._fixtureinfo.argnames}
    testfunction(**testargs)
    return True

Here you can even do some custom logic before doing something with the return value, for example check for a mark for example.

As @RonnyPfannschmidt mentioned, it is not clear yet how we could refactor the internals in order to support this more clearly.

[Droogans]

@nicoddemus aren't yield based tests going away in 4.x anyway?

[nicoddemus]

Yep, that's why I didn't bother copying their logic above 😉

[Droogans]

Ok I went ahead and tried this today, this was what I had to do:

@pytest.hookimpl(trylast=True, hookwrapper=True)
def pytest_pyfunc_call(pyfuncitem):
    yield
    funcargs = pyfuncitem.funcargs
    testargs = { arg: funcargs[arg] for arg in pyfuncitem._fixtureinfo.argnames }
    element_of_interest = pyfuncitem.obj(**testargs)
    selenium = funcargs.get('selenium')

    if selenium and element_of_interest:
        selenium.execute_script('arguments[0].scrollIntoView(true);', element_of_interest)
        # pass through to the end of the selenium fixture's teardown section, will take screenshot there

[nicoddemus]

In the snippet above you will end up calling the test function twice, is that what you intended?

[Droogans]

No, but that's why I posted it here. It appears to be doing some extra work at the end, I'll see if I can't put a debugging statement in a few places to catch those things as I continue to iron out this snippet. Thanks.

[nicoddemus]

If you take out the hook-wrapper declaration it should do what you want, I think:

def pytest_pyfunc_call(pyfuncitem):
    funcargs = pyfuncitem.funcargs
    ...

[Zac-HD]

Closing in favor of #7337, where the plan is to issue a warning and eventually error on non-None return values. Specific use-cases like async functions can be, and already are, handled by plugins which insert a run-the-thing shim.

[RonnyPfannschmidt]

Reopening as this issue is about enabling other plugins to actually handle this without copying the surrounding machinery, which inherently adds a structural fragility

[RonnyPfannschmidt]

closing this as too broad, will create new card