WIP, MAINT: Fix CircleCI

[larsoner]

CircleCI is failing on main. I can replicate locally easily with:

$ make -C doc clean && PATTERN=intro make -C doc html_dev-pattern
...
generating gallery for auto_tutorials/intro... [ 14%] 10_overview.py
generating gallery for auto_tutorials/intro... [ 28%] 15_inplace.py                                                                                                                 
...
AssertionError: 1 Brain:
dict: len=35, 9 referrers: dict/frame/list/module
method: <bound method Brain.save_image of <mne.viz._brain._brain.Brain object at 0x7f732744f1c0>>
cell: <cell at 0x7f7325b8c8b0: Brain object at 0x7f732744f1c0>
cell: <cell at 0x7f7325ae1c10: Brain object at 0x7f732744f1c0>
dict: len=2, 4 referrers: UpdateLUT/dict/list

In other words, something in memory can no longer be cleaned up.

According to git bisect this actually starts for me on #9742, so I reverted #9807, then #9802, then #9742. Let's see if [circle full] actually works here. If it does, we can drill into why it's actually breaking things. I looked through pooch's code a bit and I don't see any fancy mechanics that could be causing this problem (?).

I also have no idea why this is just happening now (after #9754 was merged), since the pooch PRs have has been merged for a while and builds came back green.

[agramfort]

Outch you propose to revert all the changes to pooch?

[larsoner]

No, I just want to see first if it does indeed fix the problem in order to isolate it. If the problem is somewhere in those PRs I reverted, then we'll have to figure out where/why and make a more targeted fix rather than reverting everything.

[drammock]

FWIW, on current main (8cb7838) I cannot replicate the error locally using your make line in the issue description. Here is my sysinfo:

Platform:      Linux-5.4.0-80-generic-x86_64-with-glibc2.31
Python:        3.9.6 | packaged by conda-forge | (default, Jul  6 2021, 08:53:59)  [GCC 9.3.0]
Executable:    /opt/miniforge3/envs/mnedev/bin/python
CPU:           x86_64: 16 cores
Memory:        125.7 GB

mne:           0.24.dev0
numpy:         1.21.2 {blas=NO_ATLAS_INFO, lapack=lapack}
scipy:         1.7.1
matplotlib:    3.4.3 {backend=Qt5Agg}

sklearn:       0.24.2
numba:         0.53.1
nibabel:       3.2.1
nilearn:       0.8.0
dipy:          1.4.1
cupy:          9.3.0
pandas:        1.3.2
mayavi:        4.7.2
pyvista:       0.31.3 {pyvistaqt=0.5.0, OpenGL 4.5.0 NVIDIA 460.91.03 via GeForce RTX 2060/PCIe/SSE2}
vtk:           9.0.1
PyQt5:         5.12.3
pooch:         v1.4.0

[larsoner]

Looks like #9814 is already on the way to working, so we'll go with that...