Coverage gets confused when encoding comment is in the wrong place

[nedbat]

Originally reported by David MacIver (Bitbucket: david_maciver_, GitHub: Unknown)

---

It seems that code which incorrectly has things before the encoding comment confuses coverage.

Reproduction:

conventions.py:

from __future__ import division, print_function, absolute_import


# coding=utf-8

class UniqueIdentifier(object):

    def __init__(self):
        pass


not_set = UniqueIdentifier()

test_conventions.py:

import conventions

Then running the following:

python -m coverage run test_conventions.py && coverage report --show-missing

Causes coverage to report that the file conventions.py has not got 100% coverage (it does), but also gives nonsensical line numbers:

Name                  Stmts   Miss  Cover   Missing
---------------------------------------------------
conventions.py            5      3    40%   4, 7-10
test_conventions.py       1      0   100%   
---------------------------------------------------
TOTAL                     6      3    50%   

Line 4 is the encoding comment, and lines 7-10 correspond to the class body.

I have tested this on Python 3.4.3 running coverage 4.0.2. I noticed this when my CI started breaking (I erroneously had some code like this due to a confused formatter) on a coverage upgrade, so I think it's probably introduced in a recent version.

---

• Bitbucket: https://bitbucket.org/ned/coveragepy/issue/443

[nedbat]

It's definitely reproducible, and it definitely appeared in 4.0.2. Thanks for the edge cases!

[nedbat]

Fixed in a51232f360ef (bb), aiming for 4.0.3