Bug in pyplot.plot() with zorder/solid_capstyle kwarg combinations

[megies]

There is a strange bug that changes the color used in the plot, happening to me for certain combinations of zorder and solid_capstyle kwargs. Happens to me on matplotlib 1.4.3 installed through anaconda/conda package (tried it with TKAGG and QT4AGG).

import matplotlib.pyplot as plt

plt.subplot(321)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20, zorder=None)
plt.title("zorder=None")
plt.subplot(322)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20, zorder=None, solid_capstyle="round")
plt.title('zorder=None, solid_capstyle="round"')
plt.subplot(323)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20, zorder=10)
plt.title('zorder=10')
plt.subplot(324)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20, zorder=10, solid_capstyle="round")
plt.title('zorder=10, solid_capstyle="round"')
plt.subplot(325)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20)
plt.title('')
plt.subplot(326)
plt.plot([0, 1], [0, 1], c="black", ls="-", lw=20, solid_capstyle="round")
plt.title('solid_capstyle="round"')

plt.tight_layout()
plt.show()

[tacaswell]

well, that is a fun bug....

I can't reproduce this on master and the zorder=None examples raise an exception that None is not orderable (on py3k).

With 1.4.3 + py3k I still get exceptions for zorder=None, and get different results (all blue, and sometimes the middle right one is black.

I also simplified the code a bit:

import matplotlib.pyplot as plt

fig, axs = plt.subplots(3, 2)

dicts = [[dict(c="black", ls="-", lw=20),
          dict(c="black", ls="-", lw=20, solid_capstyle="round")],
          [dict(c="black", ls="-", lw=20, zorder=10),
           dict(c="black", ls="-", lw=20, zorder=10, solid_capstyle="round")],
          [dict(c="black", ls="-", lw=20),
           dict(c="black", ls="-", lw=20, solid_capstyle="round")]
          ]

for row, kwargs in zip(axs, dicts):
    for ax, style in zip(row, kwargs):
        ax.plot([0, 1], [0, 1], **style)
        ax.set_title('{!r}'.format(style))

plt.tight_layout()
plt.show()

Just to be sure there was no funny state-machine related shenanigans going on.

[tacaswell]

And on 2.7 I get almost the same result as @megies.

This is probably related to the use of self.update(**kwargs) is the __init__ of Line2D which ends up being dependent on the order of iterations through the dictionary.

[tacaswell]

Changing c -> color works in all cases on 2.7 and 3.4.

[efiring]

This was fixed in master with #4198.

[megies]

This was fixed in master with #4198.

Sorry for not checking against master before reporting. I was on the road and in a bit of a hurry. Thanks for clarifying.

[tacaswell]

It reminded us that this should be back ported so not a waste.