Closed cinabars closed 4 years ago
Thanks for this! We should definitely fix this to standardize the behavior across the different compute
functions and to match the docstring.
I just noticed that we use motchallenge_metrics
as the default in MetricsHost.compute()
and self.names
as the default in MetricsHost.compute_overall()
(and I can't see where this is set).
Should we also replace self.names
with motchallenge_metrics
at the same time? @cheind
Also strange that appveyor failed with "Build execution time has reached the maximum allowed time for your plan (60 minutes)." It seems highly unlikely that this is due to the content of this PR.
Sorry for the late response, I'm on vacation. Yes I believe that is an inconsistency. self.names
seems to be are oversight from a previous refactoring and should be removed in factor of motchallenge_metrics
.
We've seen these random AppVeyor fails before. As far as I can tell it seems to be related to uploading to pypi-test
. We might be lucky and the error goes away without changes in a future attempt. Trying to restart the build, now.
Seems not to have helped. I will update twine upload tomorrow. Keeping this https://packaging.python.org/guides/using-testpypi/ as a reference.
The Appveyor issue is now fixed (df8dbfa) on develop. @cinabars would you mind rebasing your PR on the latest changes?
For the record: the way we handled version numbers worked only for master and development branches, but not for PR and feature branches. Uploading failed because of conflicts with already existing non-development package names (i.e no .dev###
suffix) on PyPi.
Pulled down the fix, looks like the appveyor build is still failing though.
Sorry for the inconvenience. I've disabled pypi uploading for PRs for now. Would you mind rebasing on the latest develop? 2436d4f
Hah, that will also not work as Appveyor marks pull requests as develop. APPVEYOR_REPO_BRANCH - build branch. For Pull Request commits it is base branch PR is merging int
So, according to docs it seems like before_deploy
scripts are preferrable. Those are never run for PRs. If you find the time, please rebase one more time :) If that doesn't help we will probably accept your PR anyways as all the tests pass.
It seems to work now. The main issue seems to be this: building a PR does not decrypt sensitive variables, causing PyPi to fail to access our account.
Thanks and sorry for the problems with our continous integration services.
when calling
compute_many
, if kwargmetrics
is left to its default valueNone
, it will not calculate all metrics. This does not align with the Kwargs documentation on this function, nor the behavior of similar functioncompute
.