How can I get my assertions in pytest to stop being abbreviated with ellipsis?

Could you show an example? This is how the output looks for me:

    def test_foo():
>       assert "a"*100 == "a"*100 + "b"
E       assert 'aaaaaaaaaaaa...aaaaaaaaaaaaa' == 'aaaaaaaaaaaaa...aaaaaaaaaaaab'
E         Skipping 89 identical leading characters in diff, use -v to show
E         - aaaaaaaaaaa
E         + aaaaaaaaaaab
E         ?            +

I think the diff makes it perfectly clear what's failing - but with -v (as the message suggests) I get the full diff:

    def test_foo():
>       assert "a"*100 == "a"*100 + "b"
E       assert 'aaaaaaaaaaaa...aaaaaaaaaaaaa' == 'aaaaaaaaaaaaa...aaaaaaaaaaaab'
E         - aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
E         + aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaab
E         ?

The other solution doesn't work in general. In our use case, for example, pytest truncates parameters passed to functions called from assertion statements:

E                   AssertionError: assert None is not None
E                    +  where None = search('\\n\\*\\s.*\\b[Tt]upla item 0\\b',
'@beartyped pep_hinted() parameter pep_hinted_param=((<function <lambda> at
0x7fe278838830>,),) violates PEP type hint...7fe278838830>,):\n  * Not int, bool, or
float.\n  * Tuple item 0 value <function <lambda> at 0x7fe278838830> not str.')

beartype_test/unit/pep/p484/test_p484.py:124: AssertionError

Note the ellipses (i.e., ...) that pytest interjected without our permission between the substrings violates PEP type hint and 7fe278838830>,):\n in the above output, replacing a mission-critical substring needed to make sense of that output.

Thanks, overly helpful pytest developers. You shouldn't have. Really.

what's the problem, bruh?

To corroborate both @NicholasDiPiazza and @MatthiasUrlichs, Pytest sadly ignores -v options when truncating parameters passed to functions called from assertion statements – which is more than mildly infuriating.

Pytest developers, if we ask for verbosity four friggin' times, it means: "Yes, we really do want verbosity, are well-apprised of the potential pitfalls, and could care less because uncensored error reporting is much more important than minimizing those nebulous theoretical harms that frankly don't apply to real-world continuous integration."

what's the solution, bruh?

Our working solution is as follows:

  • Pass the --showlocals option to the addopts setting in the project's top-level pytest.ini file: e.g.,
addopts = -vvvv --showlocals -p no:xvfb -r a --doctest-glob=
  • Assign the verbose string (or object representation, or whatever) being truncated to a local variable in the problematic test. In our use case, for example, we assign the verbose exception message being truncated to a local variable in that test: e.g.,
def test_problem_function():
    with raises(Exception) as exception_info:
        raise ValueError("a"*1024)

    exception_str = str(exception_info.value)   # <--- This is where the magic happens.
    assert problem_function("a"*1024, exception_str)

def problem_function(substr, bigstr):
    return substr not in bigstr
  • Manually inspect pytest output for that variable's value instead: e.g.,
========================================================== FAILURES ===========================================================
______________________________________________ test_truncated_exception_message _______________________________________________

    def test_truncated_exception_message():
        with raises(Exception) as exception_info:
            raise ValueError("a"*1024)
    
        exception_str = str(exception_info.value)
>       assert problem_function("a"*1024, exception_str)
E       AssertionError: assert False
E        +  where False = problem_function(('a' * 1024), 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa...aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa')

exception_info = <ExceptionInfo ValueError('aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa...aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa') tblen=1>
exception_str = 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaa'

beartype_test/unit/pep/p484/test_p484.py:39: AssertionError

While the AssertionError itself remains truncated, the --showlocals option correctly preserves the untruncated value of the local exception_str test variable. Voilà!

This actually works as advertised, with the caveat that this only works as advertised for strings (or object representations, or whatever) that are directly assignable to local variables in tests. That's typically but not always the case – which is why pytest really needs to start respecting -vvvvvvvvvvvvvv options in its error reporting.

Tags:

Python

Pytest

Recent Posts