Public image layer attributes for colormap 'bad' value

[bmorris3]

This PR adds public attributes to the image layer artist, state, and CompositeArray. The goal is to specify how colormaps in image layers use the bad color and alpha.

This PR is a followup to:

• Preserve alpha while compositing with cmap.set_bad #2468
• Allow transparent regions in composite images when out of bounds #2482

and specifically implements the suggestion in #2468 (comment).

Comments for discussion:

• the Colormap bad feature is specific to mpl colormaps, and in mpl docs for get/set_bad, the feature refers to "masked" values rather than "bad" values. I could rename the glue features to cmap_masked, but the reference to bad keeps the implementation specific to matplotlib colormaps.
  • this has been fixed in Public image layer attributes for colormap 'bad' value #2556 (review)

Demo

import numpy as np
import matplotlib.pyplot as plt
from astropy.io import fits
from glue_jupyter import JupyterApplication as App
from glue_jupyter.bqplot.image.viewer import BqplotImageView


# generate two example images with different x/y extents:
fits.HDUList(fits.PrimaryHDU(data=np.random.uniform(size=(20, 20)))).writeto('arr_a.fits', overwrite=True)
fits.HDUList(fits.PrimaryHDU(data=np.random.uniform(size=(10, 10)) + 2)).writeto('arr_b.fits', overwrite=True)

# load two images
app = App()
a = app.load_data('arr_a.fits')
b = app.load_data('arr_b.fits')

# link the x/y coordinates:
app.add_link(a, 'Pixel Axis 0 [y]', b, 'Pixel Axis 0 [y]')
app.add_link(a, 'Pixel Axis 1 [x]', b, 'Pixel Axis 1 [x]')

# create an image viewer with both data entries:
viewer = app.new_data_viewer('image')
viewer.add_data(a)
viewer.add_data(b)

colormaps = [plt.cm.viridis, plt.cm.Grays]

for layer, cmap in zip(viewer.layers, colormaps):
    layer.state.cmap = cmap
    layer.state.alpha = 1

    # the default behavior has the out-of-bounds opacity set to `None`, which
    # makes the layer's out-of-bounds area opaque black:
    layer.state.cmap_bad = None

    # alternatively, set the out-of-bounds opacity to zero 
    # to see the lower, colorful layer:    
    # layer.state.cmap_bad = (0, 0, 0, 0)

    # alternatively, set the out-of-bounds opacity to `1` to 
    # render the `cmap.get_bad()` color (white) in the out-of-bounds areas:
    # layer.state.cmap_bad = (1, 1, 1, 1)

using the three options available in comments in the loop:

cmap_bad = None

This is the default behavior, both before and after this PR. The out-of-bounds area is opaque black for any colormap, not just Grays:

cmap_bad = (0, 0, 0, 0)

This is the use case that motivated this PR. The out-of-bounds area is completely transparent:

cmap_bad_alpha = (0, 0, 0, 1)

The default Grays colormap has bad set to (0, 0, 0, 0) i.e. transparent black, so cmap_bad_alpha=1 gives opaque black:

[bmorris3]

I had expected cmap_bad_alpha==1 to produce the bad RGBA bad_color + (1,), but the actual result is (1, 1, 1, 1). update: fixed by #2556 (comment)

[astrofrog]

Looks good overall, thanks! See suggestion below for a possible fix.

[astrofrog]

Looks good to me! Is it ready from your side?

[bmorris3]

It's good for me, so long as you don't have better name suggestions.

[astrofrog]

Looks good, thanks!