fix util.misc.available_cpu_count()

[notestaff]

Fixes util.misc.available_cpu_count() . Also adds some generally useful pytest utils.

[yesimon]

I don't think this should be a fixture and contextmanager. The pytest philosophy doesn't want to use with mock(obj) as m... lines in the test cases because it breaks up the flow. https://pypi.org/project/pytest-mock/. For your actual test on the CPU counts the simple mock.return_value = '1' would suffice. You can perhaps monkeypatch the MockFixture module to add this functionality if necessary.

[notestaff]

The available_cpu_count tests needs to monkeypatch different return values of slurp_file for different arguments. Also, a contextmanager is needed if you want to test more than one thing within one test function. monkeypatch with context managers is built into pytest, so pytest philosophy does allow for that.

[dpark01]

I'm not 100% following what's going on in the monkeypatch code, but I like all of the other changes in the rest of the PR and am interested in getting this merged in soon. I'm fine with introducing a new fixture that, so far, is only used in one place, though I'm not quite clear exactly which code paths are being tested by it. Wouldn't the new tests-docker.sh sufficiently test the available_cpu_count function, the main drawback being that it doesn't get credit for it in coveralls?

[notestaff]

The situation in #911, where cfs_quota_us is reporting -1, doesn’t exist on travis, so the moneypatching simulates it.

[dpark01]

Hm, yes worth mocking all the possible values, I'm just not used to this way of mocking it. Let's just go with this.

I do see -1 from that cgroup file in the usual PAPIv2 setup, my guess is that test-docker.sh is actually getting -1 if --cpus was not passed to docker run (but then you wouldn't know what the expected return value of the function would be).