Really move verify tests to Python 3 #9571
Conversation
|
@mvollmer: For debugging your tests, either cherry-pick "test: Fix checkSuccess() for Python 3" here, or simply run the test through python2 for the time being ( |
|
Great work! |
martinpitt
changed the title
martinpitt
added
the
blocked
martinpitt
removed
needswork
blocked
`b64encode()` returns bytes, convert it to a string for inserting it into a string.
Python 3's TestCase.run() already does this by itself.
Python 3's unittest calls `tearDown()` before `TestCase.result` gets updated. So look at `_outcome` instead. This does not exist in Python 2, so make it version dependent.
In Python3, this exception appears as `testlib.Error:` in tracebacks, while it is just `Error:` in Python 2. Normalize it so that our existing naughty overrides continue to work.
Commit 6939ef2 missed run-tests, so while running individual check-* tests was using Python 3, running the entire suite was not. Closes cockpit-project#9571
|
Only one well-known flake, I think this is green enough. (We can't run tests ATM) |
| @@ -255,13 +255,13 @@ class TestConnection(MachineCase): | |||
|
|
|||
| # login handler: wrong password | |||
| headers = m.execute("curl -s --head --header 'Authorization: Basic {}' http://172.27.0.15:9090/cockpit/login".format( | |||
| base64.b64encode(b"admin:hahawrong"))) | |||
| base64.b64encode(b"admin:hahawrong").decode())) | |||
There was a problem hiding this comment.
b64encode returns bytes?! this is some kind of awful :(
(I checked, and indeed, it seems that there is no better way to handle this)
| @@ -286,9 +286,10 @@ account required pam_succeed_if.so user ingroup %s""" % m.get_admin_group | |||
| self.allow_restart_journal_messages() | |||
|
|
|||
| def curl_auth(self, url, userpass): | |||
| header = "Authorization: Basic " + base64.b64encode(userpass) | |||
| header = "Authorization: Basic " + base64.b64encode(userpass.encode()).decode() | |||
| @@ -221,6 +221,9 @@ def normalize_traceback(trace): | |||
| # replace noise in SELinux violations | |||
| trace = re.sub('audit\([0-9.:]+\)', 'audit(0)', trace) | |||
| trace = re.sub(r'\b(pid|ino)=[0-9]+ ', r'\1=0 ', trace) | |||
|
|
|||
| # in Python 3, testlib.Error is shown with namespace | |||
| trace = re.sub('testlib\.Error', 'Error', trace) | |||
There was a problem hiding this comment.
I'd strongly consider normalising this in the other direction (and updating the naughties). It seems that py3 is the expected norm now, so why not treat it that way?
There was a problem hiding this comment.
I filed issue #9591 about it. I'm still hoping that the actually printed format can be fixed to just Error: again, I just didn't see a good way of doing this. I don't want to change all the naughties until this settled down and we also port the selenium, avocado, and container tests.
checkSuccessis always true even on failures)check-ostree OstreeCase.testRemoteManagement_PY3in testlib.py (PR test: Robustify filterOutput() for non-ASCII characters #9582)