QPainter does not like to draw in an empty QImage.
And the rendering code does not like to render it.
So we would get this on the console:
```
QPainter::begin: Paint device returned engine == 0, type: 3
QPainter::save: Painter not active
QPainter::save: Painter not active
QPainter::setRenderHint: Painter must be active to set rendering hints
QPainter::save: Painter not active
QPainter::setClipRegion: Painter not active
QPainter::setPen: Painter not active
QPainter::setBrush: Painter not active
QPainter::restore: Unbalanced save/restore
QPainter::setPen: Painter not active
QPainter::setBrush: Painter not active
QPainter::restore: Unbalanced save/restore
QPainter::save: Painter not active
QPainter::restore: Unbalanced save/restore
QPainter::restore: Unbalanced save/restore
thread 'main' panicked at 'attempt to subtract with overflow', sixtyfps_runtime/rendering_backends/gl/texture.rs:285:42
```
Make the text color a rendering variable, so that it can be passed
through as uniform to the glyph shader and applied to the gray
map of the glyphs. This avoids re-creating the glyph runs when
merely the color changes.
This already "worked" for the glyph cache based text rendering,
but it wasn't used because of the wasm canvas code path. This
patch changes that to render the text into a text using simply black
and then render that texture using our existing glyph shader,
which merely uses the alpha channel anyway.
This reduces also #cfg's.
Currently the Qt backend still redirect everything to the GL backend,
but the goal is to use QPainter and QWindow
This also adds a "default" backend, whose goal is to select the proper
backend at compile time
Borrow the rendering and cache individually with different mutability.
We might later, for destruction handling, need to borrow the rendering
cache mutably. That in turn *may* happen even during rendering itself,
as the models are updated lazily during visitor traversal and that may
result in item destruction.
Instead, pass a reference to the root item when mapping the window,
at which point we can downcast to the new Window item. If we have one,
then we'll read its width/height (for initial values) and install
bindings to keep them up-to-date.
This adds horizontal_alignment/vertical_alignment properties, along with
width/height to Text.
This still uses a hard-coded enumeration in the compiler, which is meant
to go away in favor of general enum support.
Replace the three (ref)cells in the window with one refcell to an enum that
determines that the window is either unmapped (with a factory available)
or mapped (we have a graphics backend and rendering cache).
The interior mutability was visible to the outside via new() returning a
Rc<RefCell<..>>.
When doing deep recursions during rendering for example, that window is
mutably borrowed. That in turn prevents us from adding further members
such as window scale properties, etc. that may be read *during* that
traversal as they'd require an immutable refernce to a already mutably
borrowed refcell contents.
So instead, make the individual fields refcells.
Map it to 800x600 logical pixels for a better initial look.
This implements respecting the initial values for width/height and tries
to apply them to the window begin created.
The PinnedOptionalProp wrapper is needed because while cbindgen mapped
the previous Option<&...> to a raw pointer, the new Option<Pin<&...>> is
not detected as a pointer.
Move fields out of the HighLevelRenderingPrimitive that are suitable to
represent as uniform variables. That reduces fields like Image or
Rectangle to their image or dimension, and only if *they* change, then
we re-do the work of building geometry and uploading it to the GPU.
Instead, the typically animated properties such as the position or the
color are passed separately with each draw call. This avoids re-decoding
PNG files and uploading them again when moving an image element around
on the screen.
