Bug report

Bug summary

As I updated Matplolib from version 3.0.3 to version 3.3.0 I noticed that Colour unit test time skyrocketed: our plotting sub-package test suite went from 120 seconds to around 3000 seconds!

I quickly noticed that it was related to definitions using the spectrum plot:

This definition is using plt.bar and as I suspected that polygon clipping might be the culprit, I ran some tests on one of the offending definitions while keeping and removing the polygon clipping:

With Polygon Clipping

>>> import timeit
>>> import colour
>>> timeit.timeit(lambda : colour.plotting.plot_single_illuminant_sd(filename='test.png'), number=1)
59.81594165299998

Without Polygon Clipping

>>> import timeit
>>> import colour
>>> timeit.timeit(lambda : colour.plotting.plot_single_illuminant_sd(filename='test.png'), number=1)
1.3159556400000056

Code for reproduction

I haven't created a reproducible case but my assumption is that the clipping code is orders of magnitude slower than before, the definition I used is available here: https://github.com/colour-science/colour/blob/d5f68005f62fc86ba59745bd4c8bb8a01bb5dcb4/colour/plotting/colorimetry.py#L183 and the meat is roughly as follows:

    polygon = Polygon(
        np.vstack([
            (x_min, 0),
            tstack([wavelengths, values]),
            (x_max, 0),
        ]),
        facecolor='none',
        edgecolor='none')
    axes.add_patch(polygon)

    padding = 0.1
    axes.bar(
        x=wavelengths - padding,
        height=max(values),
        width=1 + padding,
        color=colours,
        align='edge',
        clip_path=polygon)

There are roughly 450 wavelengths and each one of them generates a coloured bar.

Matplotlib version

• Operating system: macOS 10.15.5
• Matplotlib version: 3.3.0
• Matplotlib backend (print(matplotlib.get_backend())): MacOSX
• Python version: 3.8
• Jupyter version (if applicable):
• Other libraries:

Matplotlib was installed with Pip.