The as_any() method returning a `&mut dyn Any` meant that the type has to be
'static. The class solution of returning &()
doesn't work because it has to be a mutable ref. Therefore just return an Option.
Now that we store the layer textures as Rc<Texture>, before rendering a
layer we can take a clone of a possibly existing layer texture and if if
the layer's tracker is dirty, we can re-use the existing layer texture
if it's the same size.
This avoids unnecessary texture allocations.
Fix a logic bug in the function to determine if we need to fall back to
other fonts or if the font we selected provides complete coverage.
In the common case where coverage is given, we start with some unknown
coverage (in old_uncovered_*) and end up with empty
remaining_required_*_coverage. That also happens to be less than
old_uncovered_*. Since that check came first, we always returned
"Improved", which meant we still built up the font fallback list, even
though we didn't need to. Fix the order of checks to avoid that.
I want to track component structure changes in the window without
generating more code. So use a more generic name for the init_*_items
functions, so that I can add the functionality I need in there.
Also add a register_component to PlatformWindow and call that.
I want a more generic name as I want to do to track component structure
changes in addition to resource freeing and I do not want to add another
call into the generated code.
Clean up texture caching in the GL renderer
In preparation for moving the image decoding and caching of decoded
image data into the core library, change the texture cache to be really
a cache of textures. Previously it cached Rc<CachedImage>, which could
have also been just a CPU side image. However this was "run-time"
asserted. Instead, with this patch it's an `Rc<Texture>` and that will
always be GPU side.
The previous approach of calling send_event on a timer has the disadvantage
or re-entering the event loop for every queued event. Any requested redraw
after one of those events will end up actually drawing, even if that frame could be
replaced by the next event's redraw without being shown to the user.
winit also doesn't expose publicly its web Runner send_events method that would
allow us to queue the events outside and pass them all together.
We can however queue the events inside winit by putting the event loop in
ControlFlow::Poll mode so that winit batches them itself.
This has the side effect of processing and painting those events using
requestAnimationFrame.
To achieve this we take advantage of winit processing send_event calls
synchronously, possibly while on a native event handler, by entering the
event loop just to send WakeEventLoopWorkaround, set the event loop in
Poll mode, exit, and call send_events again with our event which then
ends up being queue in the web event loop's Runner until the next animation
frame where all queued events are processed and redrawn together.
Otherwise the timing code might think at the time of an event (eg key press)
that the starting point of an animation that starts in this event is
earlier in time than it really is. Causing the animation to appear as it
had started earlier or literaly be finished before it starts.
Fixes#1255
Enable fontdb's fontconfig feature to parse fontconfig files to locate
the directories where truetype fonts are located. This helps in system
setups that differ from the defaults that fontdb uses otherwise.
Fixes#1240
Changing the constraint doesn't work on non-rezsizable window.
So first set the window as resizeable, then change the constraints, then
maybe remove the resizable flag
The situation differs depending if the widget show the virtual keyboard or not:
For widget that don't show the virtual keyboard, we rely on the winit events,
but for some reason we don't recieve WindowEvent::ModifiersChanged events with
wasm. Since the event handling in winit is about to be rewriten, I did not
bother reporting the bug upstream, but just work around by using the deprecated
API. So that way shift + tab will no longer be understood as just tab.
For widget using the virtual keyboard, then the event handling is in
wasm_input_helper.rs. There is a couple of issues:
- We must map shift+tab to backtab.
- We need to prevent the default events to trigger, so that tab and other
shortcuts don't take effect on the browser. Winit already inhibit these
events so we must do the same otherwise tab and shift+tab would change the
html focus.
- By luck, tab used to give the focus back to the canvas before (see previous
point) and that's why it worked. But now that we don't do that anymore,
hiding the virtual keyboard should actually re-focus the canvas
- That will cause the focus event to be intercepted by winit, and will cause
recursions and borrow error, so we make sure that we do not recurse when
getting the focus event
Commit c85e1b6d25 added a workaround for a
winit issue, which has been fixed upstream. Until a new release is
available, let's patch in winit from a branch that has the fix
cherry-picked.
This way we don't have to remember to remove the workaround with the
next update and this has been verified on the device.
If the global opacity is zero, we don't need to paint rectangles, etc.
This shortcut is a compromise between opting out much later on femtovg
level and too early before querying properties of the rectangle. We
still want to do the latter as somebody might depend on that.
IDWriteFontFallback::MapCharacters only returns one single font that tries to cover
as much of the input string as possible. We need to continue processing the remaining parts of the text,
in order to get a complete list for all fonts we need.
Fixes#1139
