Parametrized nodeid not matching with parametrized IDs

[vbarbaresi]

This is about this issue in pytest-repeat: pytest-dev/pytest-repeat#21

It seems that the bug doesn't come from the plugin: I managed to reproduce with a simple test here.
It affects all pytest versions, at least since 2.8

I wrote a test and a fix attempt here: vbarbaresi@12769f0

The issue is the following: with parametrized IDs and parametrized fixtures, one can't run a test using its node ID with a [parameter] argument

I tracked it down to the x.name vs name matching, which is in my example:
test_foo[1-param] vs test_foo[param]

My fix attempt drops everything after [ on both sides. However, we lose the supplied parameter value so it's not satisfying. I don't know where to fix this properly.

[blueyed]

@vbarbaresi
Can you please turn this into a PR (assuming the test is passing)?

[vbarbaresi]

The test is passing but doesn't check that [param] is honored.
With my workaround it will just match the test name, ignoring everything inside []. That's not satisfying

Now that I look at it again, the issue seems fundamentally hard to fix.
Plus, there are no tests on running a parametrized nodeid with [parameter] from the command line so it's hard to know what should work.
Consider my branch as a failed attempt.

---

To sum up:

import pytest

@pytest.fixture(autouse=True)
def __bar(request):
    return request.param

def pytest_generate_tests(metafunc):
    metafunc.parametrize('__bar', [1], indirect=True)

class TestClass(object):
    @pytest.fixture(params=['param'])
    def foo(self, request):
        pass

    def test_foo(self, foo):
        pass

metafunc.parametrize doesn't play well with nodeid + parameter
The nodeid becomes TestClass::test_foo[1-param]

---

Note that there is an old TODO in _pytest/main.py saying

# TODO: remove parametrized workaround once collection structure contains parametrization
if x.name == name or x.name.split("[")[0] == name:

I feel like it has to do with this issue

[blueyed]

Thanks for the update!

Now that I look at it again, the issue seems fundamentally hard to fix.
Plus, there are no tests on running a parametrized nodeid with [parameter] from the command line so it's hard to know what should work.

I feel your pain, having worked at collection recently.
(just as a side-note, the extra "::()" here and there annoyed me quite some time, until I dug into it to get it removed via #4358 - turned out to be not that hard then after all)

Given there are no tests I'd argue that you are free to clean this up to match your expectations.

Surely @RonnyPfannschmidt has a rant in this regard also.. :)

I'd say that a WIP-PR is still something good to have, where you could add your summarized example as a failing test to start from. Use [skip appveyor] in the commit title then to only get the baseline test reports from Travis - this can be very useful to see if and how much breaks from the current tests.

[RonnyPfannschmidt]

@blueyed i actually have a fix on the horizon, but it will at least need 3-4 major releases to sort nodes out - given that outline no more ranting needed

[DavidAntliff]

@RonnyPfannschmidt I'd be interested in the fix, if it's available?

[RonnyPfannschmidt]

The node refactoring needed is still in very early stages

[bschnurr]

Not sure if it helps but here is the work around I used to change the generated pytest id's when the parameter contained characters like "::" and "\n"

I need to change the Id so that both Visual Studio's Test Explorer could parse it correctly and pytest was having issues running a single test when trying to call it using the ID

microsoft/PTVS@caa5383

[1Mark]

Any progress on this?

[nicoddemus]

The node refactoring has been progressing, but I don't think there's news about this issue in particular (@RonnyPfannschmidt might give more details here).

[RonnyPfannschmidt]

nothing new there, after the rest of the path refactoring lands i'm going to work on setting up the function-definition node, afterwards we can do it more easily