There are 3 possible components to draw at each x-axis position: the waveform "line", the zero line and an outline/clip indicator.
We have to decide which of the 3 to draw at each position, pixel by pixel. This makes the rendering less efficient but it is
the only way I can see to do this correctly.
Items no longer need a parent group (they require a Canvas pointer instead), so all constructors have been rationalized
and have two variants, one with a parent and one with a canvas.
All Items now inherit from Fill and Outline, to banish diagonal inheritance and virtual base classes and all that.
There were zero changes to the Ardour GUI arising from these changes.
previous one, but usually ends up using three of them at the current
CACHE_HIGH_WATER setting.
Should result in a smaller memory footprint for sessions with
multiple copies of nearby audio segments (electronic style).
The downside is the larger memory footprint for linear recording
sessions if CACHE_HIGH_WATER > 1 (1 giving a max of two
half - sized cache entries per audio stream).
The idea now is that a scroll group item can be added to the canvas which will causes its children to scroll in either or both
directions (horizontal or vertical). There are few complications: the position() of the ScrollGroup is ambiguous depending
on whether you want it with scroll taken into account or not, so Item::canvas_position() was added, which defaults to
the same value as Item::position() but is overridden by ScrollGroup to return the position independent of scrolling. This
method is used when translating between item/canvas/window coordinate systems.
Note that the basic idea is that we MOVE the scroll group when a scroll happens. This mirrors what happens in the GnomeCanvas,
where Nick Mainsbridge came up with a great idea that allowed unification of the time bar and track canvases.
This makes continued merges with ardour codebase easier, at the
small cost of removing the XML-based constructors. They were not
in use anyway and their use doesn't really fit into the
conception of the canvas as a dynamic UI element. Could reintroduce
them at a later date.
Mostly this just involved removing the private CacheEntry class that really served no further purpose once
the design reverted to a single cached image
Rather than maintain a set of images in a cache, when we no longer have the required waveform data, create a new image that is appropriately centered and extends to roughly twice the screen width (or the limits of the region's source file(s), as necessary)
