Error when color value is None

[kyleam]

Commit d58a84a changed validate_colors so that value in rcParams is None rather than "None". However, in some cases, it seems that a string is still expected downstream.

For example,

import matplotlib.pyplot as plt
import matplotlib as mpl

mpl.rcParams['patch.edgecolor'] = None

plt.plot(list(range(5)))

results in

Traceback (most recent call last):
  File "[...]/matplotlib/lib/matplotlib/colors.py", line 322, in to_rgb
    'cannot convert argument to rgb sequence')
ValueError: cannot convert argument to rgb sequence

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "[...]/matplotlib/lib/matplotlib/colors.py", line 370, in to_rgba
    r, g, b = self.to_rgb(arg)
  File "[...]/matplotlib/lib/matplotlib/colors.py", line 328, in to_rgb
    'to_rgb: Invalid rgb arg "%s"\n%s' % (str(arg), exc))
ValueError: to_rgb: Invalid rgb arg "None"
cannot convert argument to rgb sequence

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "./scratch.py", line 9, in <module>
    plt.plot(list(range(5)))
  File "[...]/matplotlib/lib/matplotlib/pyplot.py", line 3092, in plot
    ax = gca()
  File "[...]/matplotlib/lib/matplotlib/pyplot.py", line 827, in gca
    ax =  gcf().gca(**kwargs)
  File "[...]/matplotlib/lib/matplotlib/figure.py", line 1288, in gca
    return self.add_subplot(1, 1, 1, **kwargs)
  File "[...]/matplotlib/lib/matplotlib/figure.py", line 973, in add_subplot
    a = subplot_class_factory(projection_class)(self, *args, **kwargs)
  File "[...]/matplotlib/lib/matplotlib/axes/_subplots.py", line 73, in __init__
    self._axes_class.__init__(self, fig, self.figbox, **kwargs)
  File "[...]/matplotlib/lib/matplotlib/axes/_base.py", line 422, in __init__
    self.spines = self._gen_axes_spines()
  File "[...]/matplotlib/lib/matplotlib/axes/_base.py", line 812, in _gen_axes_spines
    'left': mspines.Spine.linear_spine(self, 'left'),
  File "[...]/matplotlib/lib/matplotlib/spines.py", line 454, in linear_spine
    result = cls(axes, spine_type, path, **kwargs)
  File "[...]/matplotlib/lib/matplotlib/spines.py", line 54, in __init__
    super(Spine, self).__init__(**kwargs)
  File "[...]/matplotlib/lib/matplotlib/patches.py", line 111, in __init__
    self.set_edgecolor(edgecolor)
  File "[...]/matplotlib/lib/matplotlib/patches.py", line 274, in set_edgecolor
    self._edgecolor = colors.colorConverter.to_rgba(color, self._alpha)
  File "[...]/matplotlib/lib/matplotlib/colors.py", line 376, in to_rgba
    'to_rgba: Invalid rgba arg "%s"\n%s' % (str(arg), exc))
ValueError: to_rgba: Invalid rgba arg "None"
to_rgb: Invalid rgb arg "None"
cannot convert argument to rgb sequence

This particular error can be fixed with

diff --git a/lib/matplotlib/colors.py b/lib/matplotlib/colors.py
index 772da0f..810a528 100644
--- a/lib/matplotlib/colors.py
+++ b/lib/matplotlib/colors.py
@@ -342,7 +342,7 @@ def to_rgba(self, arg, alpha=None):
         *alpha* will replace the original *A*.
         """
         try:
-            if arg.lower() == 'none':
+            if arg is None or arg.lower() == 'none':
                 return (0.0, 0.0, 0.0, 0.0)
         except AttributeError:
             pass

A few other places also seem to expect a string.

lib/matplotlib/collections.py:572:            if c.lower() == 'none':
lib/matplotlib/collections.py:611:            if c.lower() == 'none':
lib/matplotlib/colors.py:393:            if nc == 0 or c.lower() == 'none':
lib/matplotlib/lines.py:725:            if is_string_like(edgecolor) and edgecolor.lower() == 'none':
lib/matplotlib/lines.py:1155:        if is_string_like(facecolor) and facecolor.lower() == 'none':
lib/matplotlib/lines.py:1163:        if is_string_like(facecolor) and facecolor.lower() == 'none':

Should these cases be updated to support None? I'd be happy to submit a PR, but I'm not clear on what the intended behavior is.

[efiring]

The intended behavior is that the None object is not supported; it should always be a string.

[kyleam]

Then should validate_colors return a string instead of None?

[efiring]

Yes. The None object can be used as a kwarg in a function call to indicate that the default should be used. In that case, it is the function's responsibility to find the default in rcParams and use it to replace the None object with a valid color specification (string or tuple) before it goes any farther.

[efiring]

@tacaswell The relevant PR is #3792, which you merged as 92e608d. It looks like this needs to be redesigned. The None object should not be a valid rcParams entry, and this PR relies on it; in the process of doing so, it wrecks the ability to use 'none' as a color spec in rcParams.

[tacaswell]

I see the problem. How do we want to deal with optional rcparams? If we don't let those two new rcparams be None, we break back compatibility as we change what rcparams parts of the legend look at. I think adding a new validator color_or_none for these cases and restore the old color validation.

[efiring]

Things will be broken no matter what we do; the new validator approach would not allow the 'none' color spec to be used in this location. The basic problem is that this was an attempt to introduce a forwarding functionality without putting in a proper mechanism for it. The good thing is that this is only in master, not in color_overhaul, so I think that we can simply revert most of it, and no one will have a problem. We can leave in the new kwargs, but leave out the forwarding (or alias, or whatever it should be called) until we have a suitable mechanism for it. Perhaps it could be handled with a new keyword and a new validator, as an ad-hoc way of managing a small number of such cases.
Or let the rcParams be optional; if legend doesn't find them, it uses the axes values.

[tacaswell]

@efiring I reverted in c90469b

[tacaswell]

Closed by #4193