[Bug]: rcParams['legend.loc'] can't use float-tuple like kwarg legend(loc...)

[cphlewis]

Bug summary

We can now use a (x,y) argument for plt.legend(loc=(x,y)),
but using the tuple as a value for rcParams['legend.loc'] fails.

Code for reproduction

import matplotlib.pyplot as plt
import matplotlib as mpl
xs, ys = [1,2,3], [2,3,1]

fig, ax = plt.subplots(3)

ax[0].scatter(xs, ys, label='loc-tuple-arg')
ax[0].legend(loc=(1.1, .5)) # works

ax[1].scatter(xs, ys, label='loc-rcparam-tuple')
mpl.rcParams['legend.loc'] = (0.9, .7)
ax[1].legend() # fails w/o documentation

ax[2].scatter(xs, ys, label='loc-rcparam-str(tuple)')
mpl.rcParams['legend.loc'] = '(.8, .3)'
ax[2].legend() # fails with documentation

Actual outcome

Setting rcParams['legend.loc'] to a tuple fails with

AttributeError: 'tuple' object has no attribute 'lower'

setting rcParams['legend.loc'] to the string representation of a tuple fails with a list of supported values.

Expected outcome

Ideally rcParams['legend.loc'] would apply a tuple the same way the kwarg does.

Second best would be a bit more documentation; 1) reporting the list of supported strings for any un-interpretable value, and 2) mentioning in the default matplotlibrc that only a set of strings are supported.

Additional information

I believe the loc kwarg originally only took string values, but that seems to have been at least a major version ago.

Operating system

Ubuntu

Matplotlib Version

3.3.4

Matplotlib Backend

TkAgg

Python version

Python 3.9.7

Jupyter version

No response

Installation

pip

[alexlenail]

Also having this issue

[tacaswell]

As a general comment, including the full traceback, not just the final line makes these much easier to read

In [2]: plt.rcParams['legend.loc'] = (0.9, .7)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In [2], line 1
----> 1 plt.rcParams['legend.loc'] = (0.9, .7)

File /usr/lib/python3.10/site-packages/matplotlib/__init__.py:649, in RcParams.__setitem__(self, key, val)
    647             return
    648 try:
--> 649     cval = self.validate[key](val)
    650 except ValueError as ve:
    651     raise ValueError(f"Key {key}: {ve}") from None

File /usr/lib/python3.10/site-packages/matplotlib/rcsetup.py:72, in ValidateInStrings.__call__(self, s)
     69     _api.warn_deprecated(
     70         self._deprecated_since, name=name, obj_type="function")
     71 if self.ignorecase:
---> 72     s = s.lower()
     73 if s in self.valid:
     74     return self.valid[s]

AttributeError: 'tuple' object has no attribute 'lower'

The error is coming out of the validators in rcsetup, not the legend call.

I agree there are two things to do here. The first should be done no matter what, but I am a bit wary of the second.

1. the string validator should fail gracefully if value is not a string an tell you what strings are allowed (easy difficulty)
2. the validator used for 'legend.loc' could be extended to allow sequences of floats as well as strings. Pro is users expect it. The slight risk of this is that if any other clients who are expecting a string out of this value they will be broken, but I think that is a small risk and can be handled with documentation. (medium difficulty)

They probably should be done in two PRs. The first only requries the changes at a test, the second will require the changes, a test, an API change note, and a bit of lobbying as to why we really should make this change.

[NoyHanan]

Hi all, is it still open?

[oscargus]

Yes, the original issue is still not solved. Be aware that the test added in #24600 should be modified as part of it (as that test is to make sure that a somewhat sensible error message is fed with a tuple).

[NoyHanan]

Thanks, I'll start working on it.