Until the parameters are actually removed, they are valid and rcParams.copy() must copy them if they are in the config. At least in the present master branch they are not included anymore.

Can you please check your local config file (path can be obtained using matplotlib.matplotlib_fname()) to see if these parameters are set there?

@timhoffm Thanks for your comment!

There is an empty matplotlibrc in my project root.

>>> import matplotlib
>>> print(open(matplotlib.matplotlib_fname()).read())

>>> rc_params = matplotlib.rcParams.copy()
>>> rc_params['examples.directory']
''
>>> rc_params['text.latex.unicode']
True

I think examples.directory and text.latex.unicode are included in defaultParams and it is used on setup.

You are right. However, I don't have a clear view how to prevent this.

We cannot remove the parameter unconditionally because someone might be relying that it's there.
One option would be to add a skip_deprecated parameter to copy(). However, that seems a bit of an overkill and would also not help in other use-cases like iterating over rcParams.

I also don't have any solutions…
However, I think it is inappropriate to display the warning when applying parameters which are included in the default rcParams :(

The most troubling thing on this is that these warnings are displayed when using seaborn and seaborn.reset_orig(). Now, I think this can be controlled in seaborn. So I will create an issue in the seaborn repository.

I thought we had some machinery to avoid this (or at least a discussion somewhere already).

I thought we had some machinery to avoid this (or at least a discussion somewhere already).

I think we catch and discard the warnings internally, I suspect seaborn should do the same.

I agree that this is annoying behavior, but I don't see a way around it.

• not returning deprecated rcParams in copy (and not including them in iteration?) is not great (as things don't round-trip in weird ways). However, if we are sure that the rcparam has no effect (that is we completely ignore its value), this may be a reasonable path, but would be a pretty big API change and would break people who are looking for the deprecated value in the copy with no warning (which will happen next release anyway, but they are at least getting some warning about the key now).
• not warning if the value is the default value would help this case, but would fail to warn someone doing rcParams['old'] = dflt_value in their code which would start to raise in the future
• warn based on the type of the object passed into .update, but that seems a bit too magical

I see. Thank you for your responses!

I hit upon two solutions, but I don't think they are good.

1. Deprecated rcParams should be removed from default value.
   This will solve the problem, but it will be a big change.

2. Preparing a function which returns a instance of matplotlib.RcParams
   matplotlib.rcParams.update(matplotlib.rcParamsDefault) don't show warnings. This is because matplotlib.rcParamsDefault is not a dictonary but a instance of matplotlib.RcParams. So, It is also a way to prepare the following method.

class RcParams(MutableMapping, dict):
    [...]
    def wrapped_copy(self):
	   with warnings.catch_warnings():
	       warnings.simplefilter("ignore", MatplotlibDeprecationWarning)
	       return RcParams(self.copy())