image trace type

[antoinerg]

This PR introduces a new image trace type to display images on cartesian axes and closes #3898

Codepen showing the different mocks: https://codepen.io/antoinerg/pen/OJJPxJK

Quick overview:

• Support different colormodels (rgb, rgba, hsl and hsla).
• Can rescale data via zmin and zmax
• Display data and color information on hover

Todo:

• handle z array with missing values (ie. columns with shorter length) (done in 15c06fa)
• support categorial axes
• support log axis

To discuss:

• the trace will be name image ~~decide the trace's name. ~~ Scroll down below to read different points of view on the matter. Here's a compiled a list of candidates:
  1. image
  2. raster
  3. rasterimage
  4. pixmap
  5. imshow
• decide on the default hover label. Display the color or not?

[antoinerg]

@etpinard It's ready for a first round of review !

[etpinard]

So no x and y for this iteration? Is that intended?

[antoinerg]

As far as I know, this wasn't a requirement for that iteration so I did not include it. For your information, most other libraries do not support this either. It is usually assumed that an image is made of pixels of identical size arranged in a regular fashion as in a square crystal lattice. I'd be surprised if users ask for x and y but we'll see 😄

[nicolaskruchten]

x0 and dx seem enough for a v1... any obvious cases where that wouldn't be enough? @jonmmease @emmanuelle @alexcjohnson ?

[etpinard]

FWIW I'm also very much ok with not implementing x and y in this iteration.

[etpinard]

Very nicely done @antoinerg 🎉 - I think you found the perfect way to get this new trace type out fast with solid performance and not much brittle logic. Thank you very much for that!

I made a few comments - none of them should be difficult to resolve. I'll add one more comment here:

• I'd still vote for naming this trace raster. I'm not a fan of having a the term "image" refer to things in the layout AND the data.

[etpinard]

Oh and one more thing @antoinerg: I didn't notice any funky CSS tricks like you had in one of your proof-of-concepts. I guess you decide to not go down that route, can you confirm?

[antoinerg]

Oh and one more thing @antoinerg: I didn't notice any funky CSS tricks like you had in one of your proof-of-concepts. I guess you decide to not go down that route, can you confirm?

I confirm that I ⚡ the CSS trick I had in my proof of concept. If you want me to, I could add a few comments/references about CSS image-rendering: pixelated which could potentially be used in the future to speed up rendering on browsers that support it (almost all of them except Safari and potentially SVG editors like Illustrator/Inkscape).

I decided not to go that route because it's a bit unsafe and would require manually testing all the different browsers we want to support...

[etpinard]

Great, all my comments have been resolved. Awesome work @antoinerg !!

PSA: we're picking image (not raster, not pixmap, not anything else). Turns out I was the only one that didn't like image.

💃 💃 💃

[antoinerg]

PSA: we're picking image (not raster, not pixmap, not anything else). Turns out I was the only one that didn't like image.

I am with you @etpinard: I'm not a big fan of image either and would have preferred pixmap. It seems like image is more popular so let's give the people what they want!

[archmoj]

PSA: we're picking image (not raster, not pixmap, not anything else). Turns out I was the only one that didn't like image.

Me too!

[pfeatherstone]

Fantastic stuff. Any plans of updating the docs? At the moment i'm using the demo for guidance