> Hi, could you explain what's the motivation behind this change?
> Is it an optimization (i.e. with hopefully no visible effect), or does
> it fix an actual problem?  If it fixes an actual problem, could you try
> and add a corresponding regression test for it?

I made a dumb mistake while fixing Bug#34027.  When enabling a theme we
need to change the theme-value or theme-face property of the symbols
the theme customizes, and also save a "changed" value, in case there is
any.  Since custom-push-theme did that, I thought it was fine to call
custom-push-theme here and in disable-theme (I haven't had the time yet
to test if something bad happens in disable-theme).

As a result, theme settings, saved under the theme-settings property of
THEME, got duplicated (you can see the code that adds each setting, at
the end of custom-push-theme).  So, this change fixes that bug I
introduced.  Since the test for Bug#34027, custom--test-theme-variables,
still pass after the change, I didn't consider adding a test for this.
But sure, it should be easy to add one.