Test tweaking makefile for travis #1437
Conversation
|
Could you show me a recent manifestation of the problem? I haven't seen it since @ogrisel changes. But you are right, fixing the spectral clustering stuff is probably more important first. |
| @@ -31,7 +31,7 @@ inplace: | |||
| test-code: in | |||
| $(NOSETESTS) -s sklearn | |||
| test-doc: | |||
| $(NOSETESTS) -s doc/ doc/modules/ doc/datasets/ \ | |||
| $(NOSETESTS) -s --doctest-fixtures=_fixture doc/ doc/modules/ doc/datasets/ \ | |||
| doc/developers doc/tutorial/basic doc/tutorial/statistical_inference | |||
There was a problem hiding this comment.
On my box (with nosetests version 1.2.1) it does not seem to be necessary as long as it's already there in the setup.cfg file of the project. Furthermore it used to be there in the Makefile initially when travis use to fail taking the fixtures into account so I don't think re-adding those commandline options here will help.
Maybe the travis runner has also some kind of ~/.nose.cfg outside our control that overrides those options? I wonder how we could check for that.
|
@amueller Just have a look at https://github.com/scikit-learn/scikit-learn/. The travis status button is red. |
|
@mblondel Yes, but the errors are unrelated. And there are more than I would have expected :( |
|
and the last failure reported on this page is not the state of the current master but is a commit from a incoming PR not yet merged in the master branch of the main repo. |
|
Oh, I didn't know that. Thanks @ogrisel. Can we mention that somewhere (or did I overlook it?) |
|
What is strange is that build failure reported in the badge of the project is from this PR: the branch name of the source PR is not "master" while travis is using it to set the build status of the "master" branch on: https://secure.travis-ci.org/#!/scikit-learn/scikit-learn/branch_summary |
|
https://travis-ci.org/scikit-learn/scikit-learn/builds/3462077 This one is related and occured after @ogrisel's fix. The latest failures are bit strange (those tests used to pass). Do you know what commit introduced them? The fact that Travis doesn't deterministically fail is really bothering. |
|
@ogrisel: Yes, I also find that strange. In the README.rst file, I specified explicitly the branch name, I don't know if that helps. https://secure.travis-ci.org/scikit-learn/scikit-learn.png?branch=master |
|
We should get help from Travis people (if there's a ML or message board). |
|
@mblondel ok, I believe you now ;) |
Try to force travis to see the doctest fixtures, to avoid the mldata problem.