This is what I wondered about in the first PR. over/under are usually in the extended portion of the colorbar which we represent via a triangle on either side:

https://matplotlib.org/2.0.2/_images/colorbar_only.png

I also wonder if the colorbar needs to be so wide, but that is just an aesthetic choice

@jklymak I think that the use of color blocks for over, under, and bad is a good idea. It is explicit and complete. Using the actual colorbar with extensions would show under and over, but not bad.

ThTs fine. I'm certainly not going to block on this at all. I just think it's a little inconsistent.

For the ordering, I am wondering if there is a way to put the "under" box left-aligned and the "over" box right-aligned to put them next to the values they are "near"? Right now, the yellow "over" is a bit far away from the far right edge. (That might mean putting the title on a row above the squares to squeeze it in) I missed the sprint unfortunately, so missed if you already tried this or not.

@greglucas I discussed with@tacaswell and we decided it would be less obtrusive in a notebook if we kept the color map name and under/over/bad boxes on the same line (above the image). Are you suggesting one of layouts B or C? (represented in ASCII style)

Layout A (current):

viridis  under: [█]  over: [█]  bad: [█][████████████████████████████████████████]

Layout B: left/right aligned with text labels

viridis                           bad: [█][████████████████████████████████████████]under: [█]                       over: [█]

Layout C: left/right aligned with color swatches

viridis                            bad [█][████████████████████████████████████████][█] under                         over [█]

Nice, I think it is OK the way it is and I haven't messed with anything else to see how they look. My thought was something along the lines of your third option. Maybe like this?

                viridis[████████████████████████████████████████][█] under       bad [█]           over [█]

I agree though, this would add more height to the image, so maybe not as ideal as having them all on the same line.

Or you could have over/under beside the colorbar (perhaps as triangles) and the bad swatch beside the title. I'd put the title on the left and the bad swatch on the right.

viridis                            bad [█]< [████████████████████████████████████████] >

Something like this adds a triangle (if you just want to use html/css:https://stackoverflow.com/questions/44109039/how-to-put-a-css-triangle-before-a-span

@jklymak@greglucas I like the idea, but a triangle implemented with CSSborder properties will not work with the thin grey line that denotes the edge of the color map image (which also usesborder). I don't think you could put the thin grey line around the triangle.@tacaswell suggested adding that border because some color maps may fade to white (or transparent) on one end, and showing the boundary of the image helps with that.

@greglucas's suggestion of

                viridis[████████████████████████████████████████][█] under       bad [█]           over [█]

is a viable design, if there is consensus on this being the best option. I am happy to make another revision if we get a stronger consensus on the design.

I feel like we're treading a tad close to "design by committee" and bike-shedding to death here. I'm 👍 to@greglucas's suggestion.

@dopplershift@jklymak@greglucas@timhoffm I spent some time trying to make the design mentioned above and got as close as I could. The alignment of the labels below the image is very tricky. Is the image below acceptable (or is someone else with CSS expertise able to help)?

Revised appearance

I would also appreciate help understanding why CircleCIdocs-python3* is failing. I don't see any error messages in the CI log...?

We fail on sphinx warnings as well

/home/circleci/project/doc/users/next_whats_new/colormap_get_under_over_bad.rst:4: WARNING: py:obj reference target not found: colors.Colormap.get_under/home/circleci/project/doc/users/next_whats_new/colormap_get_under_over_bad.rst:4: WARNING: py:obj reference target not found: colors.Colormap.get_over/home/circleci/project/doc/users/next_whats_new/colormap_get_under_over_bad.rst:4: WARNING: py:obj reference target not found: colors.Colormap.get_bad

Personally, I would go with either your first version (with the boxes on the same line as the name), or with the version below. I think the thing you were missing was to flex the div and justify the contents of the container.'display:flex; justify-content:space-between;

With that I get this image:

defcolor_block(color):hex_color=to_hex(color,keep_alpha=True)return ('<div title="'+hex_color+'" '+'style="display: inline-block; '+'width: 1em; height: 1em; '+'margin: 0; '+'vertical-align: middle; '+'border: 1px solid #555; '+'background-color: '+hex_color+';">'+'</div>')return ('<div>'+'<strong>'+self.name+'</strong> '+'</div>'+'<div><img '+'alt="'+self.name+' color map" '+'title="'+self.name+'"'+'style="border: 1px solid #555;" '+'src="data:image/png;base64,'+png_base64+'"></div>'+'<div>'+'<div>'+color_block(self.get_under())+' under'+'</div>'+'<div>'+'bad '+color_block(self.get_bad())+'</div>'+'<div>'+'over '+color_block(self.get_over())+'</div>')

Thanks@greglucas and@jklymak, that helped a lot. I also got the content to be responsive in narrow windows (max-width instead ofwidth). This PR should be ready for final review! 😄

Do you want to add a test for the_repr_html_ function? Assert that it is returning the html expected in a specific case or something similar.

@greglucas I improved the HTML tests. Checking against a specific output is probably not ideal because the PNG is directly embedded as base64 and the HTML output for viridis is ~2,104 characters. It wouldn't fit well into a test file.

<divstyle="vertical-align: middle;"><strong>viridis</strong></div><divclass="cmap"><imgalt="viridis color map"title="viridis"style="border: 1px solid #555;"src="data:image/png;base64,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"></div><divstyle="vertical-align: middle; max-width: 514px; display: flex; justify-content: space-between;"><divstyle="float: left;"><divtitle="#440154ff"style="display: inline-block; width: 1em; height: 1em; margin: 0; vertical-align: middle; border: 1px solid #555; background-color: #440154ff;"></div> under</div><divstyle="margin: 0 auto; display: inline-block;">bad<divtitle="#00000000"style="display: inline-block; width: 1em; height: 1em; margin: 0; vertical-align: middle; border: 1px solid #555; background-color: #00000000;"></div></div><divstyle="float: right;">over<divtitle="#fde725ff"style="display: inline-block; width: 1em; height: 1em; margin: 0; vertical-align: middle; border: 1px solid #555; background-color: #fde725ff;"></div></div></div>

@QuLogic is there a reason you did not merge?

Should be squashed to one or a few commits.

I restarted tests, I believe. Probably the macOS/TkAgg issue.

@bdice Thanks for your contribution and the patience to work though the various review iterations! We hope to see you back some time.