I would have used

colors = mpl.colormaps['plasma'](np.linspace(0, 1, n_lines))

for i, color in enumerate(colors):
    ax.plot([0, i], color=color)

in particular

• mpl.colormaps.get_cmap('plasma') is somewhat unconventional. The ColormapRegistry supports item access for colormap names. That feels more natural to me for just looking up a colormap on the registry. get_cmap() is similar but has additional functionality like Colormap-passthrough and default-lookup for None. I would either use mpl.colormaps['plasma'] or just use the pyplot function plt.get_cmap('plasma') to save the import mpl (technically you could also do plt.colormaps['plasma'] because we re-export the registry in pyplot).
• I didn't even know that one can index colormaps by integer, and I find it a bit harder to understand. First, one has to know the integer-indexing (which feels half an implementation detail to me because I regard the colormaps as conceptually continuous). Second, looking up the color is now split beween creating the resampled cmap (l.24) and later looking up the color (l.29). They only make sense together but are quite separted. The vectorized approach of mapping an array of floats to colors is more explicit. Note also that you can easily sample parts of the colormap in the vectorized lookup, e.g. mpl.colormaps['RdYlGn'](np.linspace(0.5, 1, n_lines)) to sample only the upper half of a diverging colormap.

Thanks @timhoffm. It seems you and @jklymak are in agreement about the integer indexing being “suprising” so we should probably change it!

I did start with plt.get_cmap but I thought we were generally supposed to avoid pyplot for most things.

The qualitative colormaps are ListedColormaps and obtaining all their colors can be done easier via:

colors = mpl.colormaps['Dark2'].colors

Note that this is a significantly different task than selecting N equidistant colors from a continuous map. I'm not clear on the focus we want to set here: #27671 was primaily "I have N lines and want differnt colors", which is the first part here. My comment on a context example was the idea one should mention in the context of the above task that you can make a context so that you don't have to carry the colors explicitly. That's a different direction than focussing on various aspects of getting color values. Both ways are possbile, but we should not mix them.

Hmmm, I started by using the colors attribute for both examples (plasma is also a ListedColorMap), but changed it in response to one of @jklymak’s comments.

My thinking for using a qualitative colormap: if someone is changing the color cycle in rcParams then they are using it across several plots which might need different numbers of colors, so getting a specific number of colors out of a continuous map would make less sense there. This is in the context where the user was also asking for an increase in the default cycle.