This is going to be really awesome for our notebook users. I'll be sure to feature it in the workshops I teach.

For the PNG, I guess this is okay. However, colormaps have some other options like bad/over/under colors, which I think we could also insert into the HTML repr.

... good point - extend='both' would be a nice addition....

I am happy to bank this improvement and then sort out how to handle the over / under / bad.

@story645 suggested something like f'{name} (over {html rect of over color} | under {html rect of under color} | bad {html rect of bad color}'

Spoke with @bdice on audio. Because we don't currently have public api to get the under/over/bad values I am going to merge this PR as-is and @bdice we open a new PR adding get_* methods and adding them to the html repr.

I guess I'm somewhat leery about this. Is there some reason this doesn't just create a colorbar and then fill the png with that? You won't get the over under w/o the colorbar logic anyhow. It would also be nice if the repr had an axes either going from 0-1 or from 0-N where N is the number of colours.

I suggested to go down this route of direct PNG generation to get the simplest possible code to get a visual representation of the colormap. We already have a hard dependency on pillow so this seems way simpler than going through making a figure -> colorbar -> save it. I also think the split here between the png repr (just the color ramp) and the html (colorramp + metadata) so it will play nicely with some lower-fidelity UI (like terminals that can display png).

I do not think that the tick labels are particularly useful (we don't show them in https://matplotlib.org/3.2.2/tutorials/colors/colormaps.html) because we only want to show the gamut.

As for adding information to the html repr we should move that discussion to #17900

I'm going to go ahead and merge this (despite @jklymak 's concerns) because this is definitely an improvement and all of the user feedback I have seen on this has been super positive.

Congratulations on your first Matplotlib PR @bdice 🎉

Really looking forward to this feature!

Question: Does this already deal with transparent colormaps? I recently had troubles establishing a custom colormap with transparency, because whenever I plotted it, the background in my notebook was white, so I couldn't tell if my transparency settings actually worked.

I ended up with this snippet, but it would be great if the _repr_html/_repr_png could just show a grey-checkered background.

x = numpy.arange(256)
fig, ax = pyplot.subplots(figsize=(12,1))
ax.scatter(x, numpy.ones_like(x) - 0.01, s=100, c=[
    cmap_without_transparency(v)
    for v in x
])
ax.scatter(x, numpy.ones_like(x) + 0.01, s=100, c=[
    cmap_with_transparency(v)
    for v in x
])
ax.set_ylim(0.9, 1.1)
pyplot.show()

This is already available in 3.4.0. It does not do anything special for transparency.