Datashader image resampling sometimes fails if not contained in viewport

[philippjfr]

There appear to be various issues when datashading image not contained in current view port. A good way to reproduce this is by drawing multiple non-overlapping images:

import numpy as np
import holoviews as hv
hv.extension('bokeh')

from holoviews.operation.datashader import regrid

ls = np.linspace(0, 10, 200)
xx, yy = np.meshgrid(ls, ls)
bounds_list=[(-1,-1,1,1),(0,0,2,2),(0.5,0.5,3,3)]
imgs = [regrid(hv.Image(np.sin(xx)*np.cos(yy), bounds=bounds)) for i,bounds in enumerate(bounds_list)]

hv.Overlay(imgs)

This raises errors originating both in datashader and holoviews.

[stevievb]

If I change the first example a bit I am seeing a similar error. @philippjfr I assume this is related?

import numpy as np
import holoviews as hv
hv.extension('bokeh')

from holoviews.operation.datashader import regrid, datashade

ls = np.linspace(0, 10, 200)
xx, yy = np.meshgrid(ls, ls)
bounds_list=[(-1,-1,1,1),(0,0,2,2),(0.5,0.5,3,3)]
imgs = [hv.Image(np.sin(xx)*np.cos(yy), bounds=bounds) for i,bounds in enumerate(bounds_list)]

datashade(hv.Overlay(imgs))

[philippjfr]

Yes, I can reproduce it.

[philippjfr]

Something going wrong in the shade operation, seems like the densities aren't being passed through. Thought I'd fixed that but apparently not.

[philippjfr]

This has been fixed in the PR referenced above.