Just to confirm, in general the preferred approach is as.

In __init__(..., parameter=None, ...):

self._parameter = None 
self.set_parameter(parameter)

In set_parameter:

...
     parameter : type, default: :rc:`...`
...
if parameter is None:
    self._parameter = rcParams...
else:
    self._parameter = parameter

?

I noted that this didn't really hold in all files. Main benefit, I guess, is that the default value is handled consistently both in documentation and in the code (so that it is "always" possible to call with None to reset to default) and so that the redundant calls to rcParam are reduced.

Modified the test to use rc_context (which I assume is the better way?).

Modified the test to use rc_context (which I assume is the better way?).

Each test gets fresh rcParams to ensure they are isolated. So it doesn't really matter here.

Just to confirm, in general the preferred approach is as.

We do not have a policy on this. This approach makes the setters to accept None, meaning "set to rcParams". This is a bit of a stretch of the setter concept, but there's precedence, so it's okish.

This is a bit of a stretch of the setter concept.

In a good or a bad way? 😄

Just noted that this was heavily used in the rest of this file (at least).

(Saw on SE (although I cannot find it now) that the self._parameter = None in __init__ was not really encouraged. I seem to recall that some static code analysis tools are complaining, but maybe they realize that it in fact is initialized when __init__ is called.)

Finally got around to updating this. Now with a minimal impact on the code, simply checking on creation.

I think we should have another reviewer because the PR changed significantly after @tacaswell's approval.