- Contributing to distro identification currently doesn't have any specific standards and rather depends on the specific implementation.
- A 100% coverage is expected for each PR unless explicitly authorized by the reviewer.
- Please try to maintain maximum code-health (via landscape.io).
Distro's tests are implemented via a standardized framework under tests/test_distro.py
For each distribution, tests should be added in the relevant class according to which distribution file(s) exists on it, so, for example, tests should be added under TestOSRelease
where /etc/os-release
is available.
The tests must be self-contained, meaning that the release files for the distribution should be maintained in the repository under tests/resources/distros/distribution_name+distribution_version
.
A tests method would like somewhat like this:
def test_centos7_os_release(self):
desired_outcome = {
'id': 'centos',
'name': 'CentOS Linux',
'pretty_name': 'CentOS Linux 7 (Core)',
'version': '7',
'pretty_version': '7 (Core)',
'best_version': '7',
'like': 'rhel fedora',
'codename': 'Core'
}
self._test_outcome(desired_outcome)
The framework will automatically try to pick up the relevant file according to the method's name (centos7
meaning the folder should be named centos7
as well) and compare the desired_outcome
with the parsed files found under the test dir.
The exception to the rule is under the TestDistroRelease
test class which should look somewhat like this:
def test_centos5_dist_release(self):
desired_outcome = {
'id': 'centos',
'name': 'CentOS',
'pretty_name': 'CentOS 5.11 (Final)',
'version': '5.11',
'pretty_version': '5.11 (Final)',
'best_version': '5.11',
'codename': 'Final',
'major_version': '5',
'minor_version': '11'
}
self._test_outcome(desired_outcome, 'centos', '5')
Where the name of the method is not indicative of the lookup folder but rather the two last arguments in _test_outcome
.
A test case is mandatory under TestOverall
for a PR to be complete.