Accept a PIL Image as input to draw_image()

[jwiggins]

This adds Image from the Python Imaging Library as an acceptable argument for draw_image on all relevant kiva backends. Further, it uses Image internally to handle conversion of color modes to something which is compatible with the color mode of the graphics context.

Fixes #574

[jwiggins]

Damn. Looks like PIL.ImageQt doesn't work the same on Pyside2... Saved by the new pillow 6.2.2 eggs on EDM!

[corranwebster]

One thing to be careful about - I think Enable uses the agg backend for rendering backbuffers no matter what the Kiva library being used is and then (I suspect) uses draw_image to do the blitting. Removing the ability of the draw_image to accept agg images may break this.

This is the relevant code, I think:

enable/enable/component.py

Lines 766 to 805 in 8bc13b1

	if not self.draw_valid:
	# get a reference to the GraphicsContext class from the object
	GraphicsContext = gc.__class__
	if hasattr(GraphicsContext, "create_from_gc"):
	# For some backends, such as the mac, a much more efficient
	# backbuffer can be created from the window gc.
	bb = GraphicsContext.create_from_gc(
	gc, (int(width), int(height))
	)
	else:
	bb = GraphicsContext((int(width), int(height)))
	# if not fill_padding, then we have to fill the backbuffer
	# with the window color. This is the only way I've found that
	# it works- perhaps if we had better blend support we could set
	# the alpha to 0, but for now doing so causes the backbuffer's
	# background to be white
	if not self.fill_padding:
	with bb:
	bb.set_antialias(False)
	bb.set_fill_color(self.window.bgcolor_)
	bb.draw_rect((x, y, width, height), FILL)
	# Fixme: should there be a +1 here?
	bb.translate_ctm(-x + 0.5, -y + 0.5)
	# There are a couple of strategies we could use here, but we
	# have to do something about view_bounds. This is because
	# if we only partially render the object into the backbuffer,
	# we will have problems if we then render with different view
	# bounds.
	for layer in self.draw_order:
	if layer != "overlay":
	self._dispatch_draw(layer, bb, view_bounds, mode)
	self._backbuffer = bb
	self.draw_valid = True
	# Blit the backbuffer and then draw the overlay on top
	gc.draw_image(self._backbuffer, (x, y, width, height))

Edit: we might be OK - it looks like it tries to create a backbuffer of the same GC type.

[jwiggins]

Removing the ability of the draw_image to accept agg images may break this.

The hasattr(img, "bmp_array"): checks are basically equivalent to checking isinstance(img, agg.GraphicsContextArray). Additionally, most backends can accept an instance of themselves as the image argument.

[jwiggins]

As discussed in #589

[rahulporuri]

The code changes make sense to me and IIUC, the code is implicitly tested. The code definitely looks cleaner - and mostly the same between the various kiva backends.

Not sure if adding unittests make the code easier to understand.

[rahulporuri]

Just to confirm, we don't need to copy and pad the array data - because the use of PIL.Image.Image means that this is implicitly handled. Correct?

[jwiggins]

We do, but PIL handles it for us now via PIL.ImageQt.ImageQt

[jwiggins]

@corranwebster Do you have any more feedback or questions?

[corranwebster]

@corranwebster Do you have any more feedback or questions?

I think I'm good with this.

[jwiggins]

Thanks for the feedback