CI tests are failing (xvfb issue?)

Conda-based CI tests are failing, as seen in PR #755. This this is new behavior. I checked on my local branch and this appears to be happening on the main branch => appears that the issue is outside of this repository. One possibility is an issue with slycot.

Doing a bit of digging. It doesn't look like slycot is the issue, since Python 3.7 without Slycot without cvxopt also crashes. The error messages seem to point to coverage (?). I'm try to isolate things here.

I don't think that it is coverage.

Looks Qt related which is the default backend for Matplotlib. My first guess is that this is a conda-forge packaging bug on the MPL or Qt side.

More things to try:

pytest-xvfb
set MPLBACKEND
try QT_QPA_PLATFORM=offscreen

Using pytest-xvfb fixes the original crash (which was happening in config_test.py), but introduces a new test failure:

def test_root_locus_zoom(self):
[102](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:103)
        """Check the zooming functionality of the Root locus plot"""
[103](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:104)
        system = TransferFunction([1000], [1, 25, 100, 0])
[104](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:105)
        plt.figure()
[105](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:106)
        root_locus(system)
[106](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:107)
        fig = plt.gcf()
[107](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:108)
        ax_rlocus = fig.axes[0]
[108](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:109)
    
[109](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:110)
        event = type('test', (object,), {'xdata': 14.7607954359,
[110](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:111)
                                         'ydata': -35.6171379864,
[111](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:112)
                                         'inaxes': ax_rlocus.axes})()
[112](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:113)
        ax_rlocus.set_xlim((-10.813628105112421, 14.760795435937652))
[113](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:114)
        ax_rlocus.set_ylim((-35.61713798641108, 33.879716621220311))
[114](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:115)
>       plt.get_current_fig_manager().toolbar.mode = 'zoom rect'
[115](https://github.com/murrayrm/python-control/runs/7679958935?check_suite_focus=true#step:5:116)
E       AttributeError: 'NoneType' object has no attribute 'mode'

There is a note saying that we were using xvfb-run to get the right MPL back end, so that is probably related somehow.

One thing that is odd is that the other test suites pass and they are using xvfb-run. In particular, control-slycot-src.yml seems fine and it runs the full test suite...

There is a note saying that we were using xvfb-run to get the right MPL back end, so that is probably related somehow.

Yes, it is related. The plain Agg backend does not have the toolbar for the figure manager. We have seen this before: #504 (comment)

So this looks more and more like a qt backend issue in the MPL or Qt packages in conda-forge.

Are you able to determine package version differences in the control-slycot-src vs. the python-package-conda runs?

It looks like the error occurs when the installation uses a specialized conda environment (versus just installing things in the conda base environment). There are quite a few differences in the various packages that are installed in each case.

Now that the conda error is resolved, there is a new X11 error on the main branch. Something in the GHA runners must have changed between the check run for #755 and after the merge of it.

The failures with the invalid DISPLAY variable are totally random. We might just have been lucky in the builds for #755 and #757.

https://github.com/bnavigator/python-control/commits/update-CI

For the record: the random DISPLAY failure is back. (Maybe due to #796)

Navigation Menu

Repository navigation

CI tests are failing (xvfb issue?) #756

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

CI tests are failing (xvfb issue?) #756

Description

Activity

murrayrm commented on Aug 4, 2022

bnavigator commented on Aug 4, 2022

murrayrm commented on Aug 4, 2022

murrayrm commented on Aug 4, 2022

bnavigator commented on Aug 4, 2022

murrayrm commented on Aug 5, 2022

bnavigator commented on Aug 6, 2022

bnavigator commented on Aug 6, 2022

bnavigator commented on Nov 20, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions