Evoked.plot_topomap() behaves in strange ways when axes is passed

Describe the bug

Evoked.plot_topomap() produces unexpected behavior for me when I pass an Axes object via the axes keyword argument.

Steps to reproduce

import matplotlib
matplotlib.use('Qt5Agg')

import matplotlib.pyplot as plt
from mne.datasets import sample
from mne import read_evokeds

path = sample.data_path()
fname = path + '/MEG/sample/sample_audvis-ave.fif'

condition = 'Left Auditory'
evoked = read_evokeds(fname, condition=condition, baseline=(None, 0))

times = [0.05]

for colorbar_0, colorbar_1 in zip([False, True, True],
                                  [False, True, False]):
    fig, ax = plt.subplots(2, 2)
    evoked.plot_topomap(times, ch_type='mag', time_unit='s',
                        axes=ax[0, 0], show=True, colorbar=colorbar_0)
    evoked.plot_topomap(times, ch_type='mag', time_unit='s',
                        axes=ax[1, 0], show=True, colorbar=colorbar_1)

Expected results

Not to sure, actually! But certainly not what I do get 😃

Actual results

colorbar_0 = colorbar_1 = False

Two figures appear:

colorbar_0 = colorbar_1 = True

Two figures appear:

colorbar_0 =True; colorbar_1 = False

Two figures appear:

Additional information

Platform:      macOS-10.15.4-x86_64-i386-64bit
Python:        3.8.2 | packaged by conda-forge | (default, Mar 23 2020, 17:55:48)  [Clang 9.0.1 ]
Executable:    /Users/hoechenberger/miniconda3/envs/mne-dev-new/bin/python
CPU:           i386: 8 cores
Memory:        16.0 GB

mne:           0.21.dev0
numpy:         1.18.1 {blas=blas, lapack=blas}
scipy:         1.4.1
matplotlib:    3.2.1 {backend=Qt5Agg}

sklearn:       0.22.2.post1
numba:         Not found
nibabel:       3.0.2
cupy:          Not found
pandas:        1.0.3
dipy:          1.1.0
mayavi:        4.7.1 {qt_api=pyqt5, PyQt5=5.12.3}
pyvista:       0.24.0
vtk:           8.2.0

[mmagnuski]

@agramfort I don’t get missing topos (or duplicated figures) on windows in interactive backend. I was asking about inline because show=True shows multi-topo figure too early, so some topos are missing then.

[agramfort]

to me the pb here is really about colorbar axes that is too magical now