Scatterplot (and possibly others?) fail when specifying a color list manually

Matplotlib’s scatterplot allows to manually specify the color for each dot. From the documentation of seaborn.scatterplot, I see that additional kwargs should be passed on to matplotlib’s scatterplot, i.e., I should be able to set that color list.

However, here is a minimal example that leads to a crash:

import pandas as pd
import seaborn as sns

data = [{"x": x, "y": y} for x in range (0,5) for y in range(0,5)]
df = pd.DataFrame(data)
colors = [ "#000000" ] * len(data)

sns.scatterplot(x="x", y="y", color=colors, data=df)

The problem, as far as I can tell, is the magic with which seaborn determines colors in line 740 of relational.py. Here, an empty plot is attempted with the kwargs passed, which leads to a mismatch in the length of the c, x and y fields.

I’m not sure what the best solution would be. If the empty plot in line 740 is only needed for the colors, one might want to skip it when colors are specified manually?

[mwaskom]

Unfortunately I guess the answer is that this is broken on newer matplotlib versions, and I see no value in trying to convince upstream otherwise; sorry about that.

[mwaskom]

For some reason the matplotlib developers are constantly changing the c/color functionality. I think this is an irritating example where they’ve made an uneccessary (imo) change and broken seaborn.