3.6b2 Breaks NoseXCover

[nedbat]

Originally reported by Kevin Stone (Bitbucket: kevinastone, GitHub: kevinastone)

---

Upgrading from 3.5.3 to 3.6b2 (via PyPI) breaks the coverage calculations generated by the nose plug-in, nosexcover.

---

• Bitbucket: https://bitbucket.org/ned/coveragepy/issue/224
• This issue had attachments: 3.6-debug.txt, 3.5.2-debug.txt, coverage-3.5.3.xml, coverage-3.6b1.xml, coverage-3.6b2.xml. See the original issue for details.

[nedbat]

Kevin, can you provide more details? How are they broken? Did you run them under 3.6b1?

[nedbat]

Original comment by Kevin Stone (Bitbucket: kevinastone, GitHub: kevinastone)

---

Sorry about the lack of details, just wanted to send a quick report since you pushed 3.6b2 to PyPI so it will likely affect many others as well who just pip install coverage like I had been doing without a version guard.

I've been deploying from PyPI, so my code runs fine on 3.5.3. I haven't tried running under 3.6b1 (will try now). Basically, all my tests showed 0% coverage under 3.6b2.

It looks like it works fine under 3.6b1 but not 3.6b2.

Because I'm running coverage via several nose plugins (with are then run via django-nose), I don't have a lot of direct experience with coverage (sorry).

Find attached some coverage.xml files running the same tests under different version of coverage.

Here's my setup.cfg for running nosetests:

[nosetests]
with-doctest=1
#verbosity=2
with-xcoverage=1
cover-erase=
cover-package=subblime
with-xunit=1
exclude-dir=functional
nocapture=
with-progressive=1
with-yanc=1
with-xtraceback=1

and my testing_requirements.txt (with the coverage version guard added):

# For Testing
factory_boy
mock
django-nose
coverage<3.5.99
nose-exclude
nose-progressive
nosexcover
yanc
xtraceback

[nedbat]

Original comment by Tres Seaver (Bitbucket: tseaver, GitHub: tseaver)

---

I'm seeing a regression from 3.5.2 for my 'compoze' package.

With coverage 3.5.2

:::bash

    $ cd /tmp
    $ git clone git@github.com:tseaver/compoze.git
    ...
    $ cd compoze
    $ /opt/Python-2.6.8/bin/virtualenv /tmp/cv352
    ...
    $ /tmp/cv352/bin/python setup.py develop
    ...
    $ /tmp/cv352/bin/easy_install nose coverage==3.5.2
    ...
    $ /tmp/cv352/bin/nosetests --with-coverage
    ...
    TOTAL                568      0   100%   
    ----------------------------------------------------------------------
    Ran 142 tests in 0.535s

    $ /tmp/cv3.5.2/bin/coverage debug sys > /tmp/3.5.2-debug.txt
    $ /tmp/cv3.5.2/bin/coverage debug data >> /tmp/3.5.2-debug.txt

With coverage 3.6.0b2::

:::bash

    $ /opt/Python-2.6.8/bin/virtualenv /tmp/cv3.6
    ...
    $ /tmp/cv36/bin/python setup.py develop
    ...
    $ /tmp/cv36/bin/easy_install nose coverage
    ...
    $ /tmp/cv36/bin/nosetests --with-coverage
    ...
    TOTAL                568    568     0%   
    ----------------------------------------------------------------------
    Ran 142 tests in 0.309s

    $ /tmp/cv3.6/bin/coverage debug sys > /tmp/3.6-debug.txt
    $ /tmp/cv3.6/bin/coverage debug data >> /tmp/3.6-debug.txt

Output from the 'debug' steps for each environment attached.

[nedbat]

Original comment by Kevin Stone (Bitbucket: kevinastone, GitHub: kevinastone)

---

I did a bisect of the repo trying to isolate the changeset that breaks, hope that helps to isolate the issue.

The first bad revision is:
changeset:   1374:4c7cc10d396f
user:        Ned Batchelder <ned@nedbatchelder.com>
date:        Sun Dec 09 15:50:39 2012 -0500
summary:     Get meta-coverage working on sub-processes.

[nedbat]

Original comment by Kevin Stone (Bitbucket: kevinastone, GitHub: kevinastone)

---

The breakage is due to the following addition in coverage/control.py:468:

+        # Combining is a kind of harvesting.
+        self._harvested = True

Reverting/removing this change fixes the issue.

[nedbat]

Original comment by Diogo Baeder (Bitbucket: diogobaeder, GitHub: diogobaeder)

---

Hi,

This is affecting me as well. I get these two behaviours:

1. If I use --cover-erase, it wipes out the coverage and reports 0% coverage;
2. If I use the "coverage run" command before nosetests, generate some reports, and then run nose with coverage again, it keeps using the same old reports no matter what I change in my project.

These issues started to happen in 3.6b2; If I downgrade coverage to 3.6b1, it works fine (as expected).

[nedbat]

Fixed in changeset ebced10fc675 (bb), but now maybe double warnings are possible?

[nedbat]

Issue #225 was marked as a duplicate of this issue.

[nedbat]

kitsunde: this issue is marked resolved. Could you open a new issue with instructions how I can reproduce the problem? Please include the version of coverage.py that you are using.