pytest-dev/pytest-cov

condition was never false

Opened this issue · 2 comments

Summary

The target function is below.(This is code that occurs the issue.)

def main(value):
    result = None
    try:
        try:
            result = 1 / value
        finally:
            if result is not None:
                print(result)
    except ZeroDivisionError as ex:
        print(ex)

The test function is below.

import pytest

from target import main

@pytest.mark.parametrize('value', [
    1,
    0,
])
def test(value):
    main(value)

Expected vs actual result

  • Expected
    • I want complete branch coverage on line 7.
  • actual
    • line 7 didn't return from function 'main', because the condition on line 7 was never false
    • line 7 didn't return from function 'main', because the condition on line 7 was always true (5.0.0)

Reproducer

Versions

title value
Python 3.9.18
Platform Linux-5.4.0-181-generic-x86_64-with-glibc2.31
Packages pytest: 7.4.2
pluggy: 1.3.0
Plugins html: 4.0.2
cov: 4.1.0
metadata: 3.0.0
mock: 3.12.0
only: 2.0.0

Config

not use.

What has been tried to solve the problem

Add dummy code to line 9.

This is a complicated situation. In theory, the condition on line 7 could be False and the if could return from the function. We can tell by looking at the code and thinking about it that if that condition is False it won't return from the function, it will branch to the except ZeroDivisionError on line 9. But coverage.py can't do that level of analysis, so it decides there is a possibility that has been missed. The message is a bit misleading, the missing branch isn't "because the condition on line 7 was never false", it's more complicated than that. (BTW: that message in the latest code has been flipped to "because the condition on line 7 was always true".)

You should had a partial branch pragma to the condition to quiet the message.

@nedbat

Thanks for your reply.

We can tell by looking at the code and thinking about it that if that condition is False it won't return from the function, it will branch to the except ZeroDivisionError on line 9. But coverage.py can't do that level of analysis, so it decides there is a possibility that has been missed.

It seems difficult to check such code correctly...

I know this code is bad, but I cannot refactor it right now.
So I wanted to know how to properly check it with coverage...