Actually do not think connected to that PR, not sure why I thought it was.

The cause is straight forward when looking at the code (set_ticks calls update_ticks, which sets minor ticks on for LogNorm). Post #13234, if it were accepted in current form, the turning on of minor ticks for LogNorm could be moved to new _reset_locator_formatter_scale, though not sure if that introduces any new issues.

Can you give it a try and run the tests, and submit a PR if that is a reasonable solution? Seems you found something that works, so feel free to give it a try...

OTOH, did you test if the same thing happens for log-axes on a normal axes instead of a colorbar? The point of the colorbar changes recently were to make the colorbar behave more like a normal axis, so, anything we can do to further that end is welcome.

If the following is sufficient to test normal axes, then the same issue does not appear for them:

t = np.arange(0.01, 20.0, 0.01)

ax = plt.subplot(111)
plt.loglog(t, 20*np.exp(-t/10.0), basex=2)

ax.minorticks_off()
ax.set_yticks([3, 4, 6, 10, 20])

plt.show()

I am actually not sure how to submit a PR when the per-requisite PR has not been merged.

I think the problem is in master w/o #13234 right? Or would your proposed change stomp on #13234?

Problem is in master w/o #13234. However the proposed fix needs behavior/function introduced by #13234.

Right - may as well wait for #13234 to go in, though you can make a PR against that branch and then rebas when its merged, its a bit fragile to do that.

@LevN0 #13234 is merged, so if you want to talk a crack at fixing this, that'd be great. If you can't get to it, ping me and I can probably do it. Thanks!