Azure pipeline is failing randomly when using `plt.subplots` because of `Tcl`

3rd Party Packages
1

Hello,

After a lot of test… I still can not figure out why the Azure pipeline for a repository I maintain is randomly, and I insist on the randomly, failing. I would appreciate any help from experienced dev!

I’m running a pipeline on the windows-latest image, with python versions ranging from 3.7 to 3.10. Here is the pipeline: pycrostates/azure-pipelines.yml at 555ef8db65755dd415ae075e53b7cd4b03f69edf · vferat/pycrostates · GitHub
It’s nothing fancy, it retrieves from cache a testing dataset or downloads it and then runs the unit tests with pytest. But, the windows runner will randomly fail during the tests (usually 1 or 2 out of the 4 python versions) with 2 errors related to matplotlib and Tcl.

The first one: _tkinter.TclError: invalid command name "tcl_findLibrary"

    def test_plot_segmentation(ModK, inst):
        """Test the plot of a segmentation."""
        segmentation = ModK.predict(inst)
    
        segmentation.plot()
        plt.close("all")
>       segmentation.plot(cmap="plasma")

pycrostates\segmentation\tests\test_segmentation.py:194: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
pycrostates\segmentation\segmentation.py:395: in plot
    verbose=verbose,
pycrostates\viz\segmentation.py:77: in plot_raw_segmentation
    **kwargs,
pycrostates\viz\segmentation.py:205: in _plot_segmentation
    fig, axes = plt.subplots(1, 1)
C:\hostedtoolcache\windows\Python\3.7.9\x64\lib\site-packages\matplotlib\pyplot.py:1455: in subplots
    fig = figure(**fig_kw)
C:\hostedtoolcache\windows\Python\3.7.9\x64\lib\site-packages\matplotlib\pyplot.py:811: in figure
    FigureClass=FigureClass, **kwargs)
C:\hostedtoolcache\windows\Python\3.7.9\x64\lib\site-packages\matplotlib\pyplot.py:327: in new_figure_manager
    return _get_backend_mod().new_figure_manager(*args, **kwargs)
C:\hostedtoolcache\windows\Python\3.7.9\x64\lib\site-packages\matplotlib\backend_bases.py:3494: in new_figure_manager
    return cls.new_figure_manager_given_figure(num, fig)
C:\hostedtoolcache\windows\Python\3.7.9\x64\lib\site-packages\matplotlib\backends\_backend_tk.py:949: in new_figure_manager_given_figure
    window = tk.Tk(className="matplotlib")
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <tkinter.Tk object .>, screenName = None, baseName = 'pytest'
className = 'matplotlib', useTk = 1, sync = 0, use = None

    def __init__(self, screenName=None, baseName=None, className='Tk',
                 useTk=1, sync=0, use=None):
        """Return a new Toplevel widget on screen SCREENNAME. A new Tcl interpreter will
        be created. BASENAME will be used for the identification of the profile file (see
        readprofile).
        It is constructed from sys.argv[0] without extensions if None is given. CLASSNAME
        is the name of the widget class."""
        self.master = None
        self.children = {}
        self._tkloaded = 0
        # to avoid recursions in the getattr code in case of failure, we
        # ensure that self.tk is always _something_.
        self.tk = None
        if baseName is None:
            import os
            baseName = os.path.basename(sys.argv[0])
            baseName, ext = os.path.splitext(baseName)
            if ext not in ('.py', '.pyc'):
                baseName = baseName + ext
        interactive = 0
>       self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)
E       _tkinter.TclError: invalid command name "tcl_findLibrary"

Note that it does not always occur in the same test, e.g. same error in a different run:

    def test_check_axes():
        """Test _check_axes checker."""
        # test valid inputs
>       _, ax = plt.subplots(1, 1)

pycrostates\utils\tests\test_checks.py:114: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
C:\hostedtoolcache\windows\Python\3.8.10\x64\lib\site-packages\matplotlib\pyplot.py:1455: in subplots
    fig = figure(**fig_kw)
C:\hostedtoolcache\windows\Python\3.8.10\x64\lib\site-packages\matplotlib\pyplot.py:808: in figure
    manager = new_figure_manager(
C:\hostedtoolcache\windows\Python\3.8.10\x64\lib\site-packages\matplotlib\pyplot.py:327: in new_figure_manager
    return _get_backend_mod().new_figure_manager(*args, **kwargs)
C:\hostedtoolcache\windows\Python\3.8.10\x64\lib\site-packages\matplotlib\backend_bases.py:3494: in new_figure_manager
    return cls.new_figure_manager_given_figure(num, fig)
C:\hostedtoolcache\windows\Python\3.8.10\x64\lib\site-packages\matplotlib\backends\_backend_tk.py:949: in new_figure_manager_given_figure
    window = tk.Tk(className="matplotlib")
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <tkinter.Tk object .>, screenName = None, baseName = 'pytest'
className = 'matplotlib', useTk = 1, sync = 0, use = None

    def __init__(self, screenName=None, baseName=None, className='Tk',
                 useTk=1, sync=0, use=None):
        """Return a new Toplevel widget on screen SCREENNAME. A new Tcl interpreter will
        be created. BASENAME will be used for the identification of the profile file (see
        readprofile).
        It is constructed from sys.argv[0] without extensions if None is given. CLASSNAME
        is the name of the widget class."""
        self.master = None
        self.children = {}
        self._tkloaded = False
        # to avoid recursions in the getattr code in case of failure, we
        # ensure that self.tk is always _something_.
        self.tk = None
        if baseName is None:
            import os
            baseName = os.path.basename(sys.argv[0])
            baseName, ext = os.path.splitext(baseName)
            if ext not in ('.py', '.pyc'):
                baseName = baseName + ext
        interactive = 0
>       self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)
E       _tkinter.TclError: invalid command name "tcl_findLibrary"

And the second error: _tkinter.TclError: Can't find a usable init.tcl

    def test_check_axes():
        """Test _check_axes checker."""
        # test valid inputs
>       _, ax = plt.subplots(1, 1)

pycrostates\utils\tests\test_checks.py:114: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
C:\hostedtoolcache\windows\Python\3.9.13\x64\lib\site-packages\matplotlib\pyplot.py:1455: in subplots
    fig = figure(**fig_kw)
C:\hostedtoolcache\windows\Python\3.9.13\x64\lib\site-packages\matplotlib\pyplot.py:808: in figure
    manager = new_figure_manager(
C:\hostedtoolcache\windows\Python\3.9.13\x64\lib\site-packages\matplotlib\pyplot.py:327: in new_figure_manager
    return _get_backend_mod().new_figure_manager(*args, **kwargs)
C:\hostedtoolcache\windows\Python\3.9.13\x64\lib\site-packages\matplotlib\backend_bases.py:3494: in new_figure_manager
    return cls.new_figure_manager_given_figure(num, fig)
C:\hostedtoolcache\windows\Python\3.9.13\x64\lib\site-packages\matplotlib\backends\_backend_tk.py:949: in new_figure_manager_given_figure
    window = tk.Tk(className="matplotlib")
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <tkinter.Tk object .>, screenName = None, baseName = 'pytest'
className = 'matplotlib', useTk = True, sync = False, use = None

    def __init__(self, screenName=None, baseName=None, className='Tk',
                 useTk=True, sync=False, use=None):
        """Return a new Toplevel widget on screen SCREENNAME. A new Tcl interpreter will
        be created. BASENAME will be used for the identification of the profile file (see
        readprofile).
        It is constructed from sys.argv[0] without extensions if None is given. CLASSNAME
        is the name of the widget class."""
        self.master = None
        self.children = {}
        self._tkloaded = False
        # to avoid recursions in the getattr code in case of failure, we
        # ensure that self.tk is always _something_.
        self.tk = None
        if baseName is None:
            import os
            baseName = os.path.basename(sys.argv[0])
            baseName, ext = os.path.splitext(baseName)
            if ext not in ('.py', '.pyc'):
                baseName = baseName + ext
        interactive = False
>       self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)
E       _tkinter.TclError: Can't find a usable init.tcl in the following directories: 
E           {C:\hostedtoolcache\windows\Python\3.9.13\x64\tcl\tcl8.6}
E       
E       C:/hostedtoolcache/windows/Python/3.9.13/x64/tcl/tcl8.6/init.tcl: couldn't read file "C:/hostedtoolcache/windows/Python/3.9.13/x64/tcl/tcl8.6/init.tcl": No error
E       couldn't read file "C:/hostedtoolcache/windows/Python/3.9.13/x64/tcl/tcl8.6/init.tcl": No error
E           while executing
E       "uplevel #0 [list source $tclfile]"
E       
E       
E       This probably means that Tcl wasn't installed properly.

I am completely out of idea, does anyone has a good suggestion on how to fix it and make this CI workflow stable and reliable?

2

I will add one more piece of information I forgot… I did see it fail on the second call to a plt.subplots() in a test, meaning the first one actually worked! e.g.

    def test_plot_epoch_segmentation():
        """Test segmentation plots for epochs."""
        n_clusters = 4
        labels = np.random.choice(
            [-1, 0, 1, 2, 3], (len(epochs), epochs.times.size)
        )
    
        plot_epoch_segmentation(labels, epochs, n_clusters)
        plt.close("all")
    
        # provide ax
        f, ax = plt.subplots(1, 1)
        plot_epoch_segmentation(labels, epochs, n_clusters, axes=ax)
        plt.close("all")
    
        # provide cbar_ax
>>      f, cbar_ax = plt.subplots(1, 1)            >> Failed here.
3

Do you need an interactive backend for your tests? If not, would it not be better to use agg?

4

No, there is probably no need for the interactive aspect. So you are suggesting setting the env variable MPLBACKEND to agg? Does this looks correct?

job: pytest
    variables:
      MPLBACKEND: agg
   ...
5

Actually no, I’m wrong, I do need interactiveness because I’m using plt.show() often to show a figure for the user.

6

Using Agg and disabling the UserWarning for matplotlib in pytest is a good enough fix for me at the moment. Thanks for the suggestion!