fail-under should show more precision; confusing when just a small value below threshold

[nedbat]

Originally reported by Jon Dufresne (Bitbucket: jdufresne, GitHub: jdufresne)

---

If the test coverage is just a small fraction below the fail-under value, the report will fail with a non-zero exit status (good and expected) but the report appears as if everything is ok. As an example I received the following report:

$ coverage report --fail-under=70
Name                                    Stmts   Miss  Cover   Missing
---------------------------------------------------------------------
...  
---------------------------------------------------------------------
TOTAL                                    1816    551    70%   
$ echo $?
2

Doing the math:

>>> (1816.0 - 551.0) / 1816.0
0.6965859030837004

This is indeed below 70%. However at first glance, coverage reports this as 70% coverage (due to rounding I assume) creating confusion for me the user. In the case of failure, I would like to see report provide a more precise number for coverage so that the failure is immediately obvious.

---

• Bitbucket: https://bitbucket.org/ned/coveragepy/issue/284

[nedbat]

Hmm, it seems odd to me to change the reporting based on whether --fail-under is used or not. Better I think would be to use the rounded number in the fail-under calculation.

[nedbat]

Original comment by Jon Dufresne (Bitbucket: jdufresne, GitHub: jdufresne)

---

Hmm, it seems odd to me to change the reporting based on whether --fail-under is used or not. Better I think would be to use the rounded number in the fail-under calculation.

Sure. I think this solution makes sense and would remove the confusion.

[nedbat]

Original comment by Alexander Walters (Bitbucket: sdamon, GitHub: sdamon)

---

Or just use floor instead of round. 0.9999 would round to 100%, and floor to 99%, and would match the expectations of --fail-under.

[nedbat]

Original comment by Jon Dufresne (Bitbucket: jdufresne, GitHub: jdufresne)

---

Or just use floor instead of round. 0.9999 would round to 100%, and floor to 99%, and would match the expectations of --fail-under.

I like this idea. I created a pull request:

https://bitbucket.org/ned/coveragepy/pull-request/36/use-mathfloor-instead-of-round-when/diff

[nedbat]

Fixed in 0a9cb6a6fc61 (bb)