Keyword selection expressions are limited to python syntax, which excludes some nodeids

[jaysw]

Hi there.

py.test is fantastic!

I'm wondering if the following is an error, or, expected behaviour:

With the following:

@pytest.mark.parametrize('p', ['spam eggs', 'ham'])
def test_me(p):
    assert 1 == 1

This works fine:

py.test -k 'test_me[ham]'

While the following triggers an internal error / syntax error:

py.test -k 'test_me[spam eggs]'

Looks like condition here:

https://github.com/pytest-dev/pytest/blob/master/_pytest/mark.py#L154

prevents any matching on parameters having spaces in them (and falls through to an eval() call).

[RonnyPfannschmidt]

The current mark keyword evaluator uses python syntax,
There is a plan to create a more free evaluator, but its much work and not a priority

[jaysw]

So, in the above example, is there a way to select and run just the test, "test_me[spam eggs]" ?

[RonnyPfannschmidt]

The exact test can be selected by nodeid just use py.test $nodeid

[Zac-HD]

To summarise, as of pytest 3.9.1:

pytest 'test.py::test_me[spam eggs]'    # Selection by nodeid works
pytest test.py -k 'test_me[spam eggs]'  # UsageError in _pytest/mark/legacy.py

The root cause is this code (updated link), which chokes when trying to eval a string which is not valid Python. The only obvious solution is to upgrade that function, as using the repr rather than the str in nodeids would have severe compatibility issues. Unclear if we want to bother though, since it's in a legacy module that may be removed soon?

[RonnyPfannschmidt]

We don't have a replacement lined up yet