Make text_input_byte_offset_for_position and text_input_cursor_rect_for_byte_offset
low-level in the API, where the implementation becomes easier. Two
helper functions in TextInput in turn collect the necessary parameters
and thus also reduce the amount of duplicated code.
This also decouples the renderer further from the window.
Don't require passing a dyn WindowAdapter. This was only used for
calling request_redraw(), which a proper window adapter should do anyway
if has_active_animations() is true.
So that we don't need to query the geometry multiple time, and this pave
the way to not have the geometry in the items
Part of #1932
As a drive by, fix the clipping of the Path element which incorrectly
offseted the clip by (x,y).
Similar fixes happen in the Clip element in some renderer, but that
didn't have effect because x and y are always 0 for the Clip element
When for example only a maximum height is set, but no maximum width,
avoid passing -1 or 0 to winit for the width. Instead use the same large
value was for Qt.
cc #2562
Return the path events for rendering as an Option, so that if the path is empty,
we simply return.
This avoids a double property dependency in the (likely) event the path is not empty.
Don't rely on the pen not being set when drawing the background of a rectangle
with transparent border. On my machine with the test case the pen has some other
color from previous drawing (or Qt internal?).
As suggested by @wengxt, we should always clear the preedit when instructed by the IME (via
an empty preedit string). We do that with winit, too.
Relates to #1925
Instead of the ClippedImage. Because we use the colorize property more
often than the source property, and these are soon going to be even more
expensive
The default impl of unregister_component in WindowAdapterSealed would call free_graphics_resources
on the rendere. Any re-implementation would have to do the same,
which the winit backend forwarded through
the winit backend specific internal renderer interface.
Instead, the original call site for unregister_component now calls
free_graphics_resources and unregister_component (for accessibility tree dirty notification).
Replace register_root_component in the WindowAdapterSealed altogether with
a default font size getter on the renderer.
This is overall simpler and allows removing the doc-hidden getter in the software renderer.
Permit the style metrics to provide a `default-font-size` and bind that
to the `Window` if set. If not provided, then the backend can set a
`default-font-size`. By ensuring that the value is non-zero at run-time,
we can later introduce a rem unit that can act as factor relative to
this non-zero font size.
Olivier found the following steps that reproduce the crash with Skia:
1. Open the gallery and select the "Text Edit" tab.
2. In the first text edit widget, select the first three letters.
3. Click into the second text edit (while the first one shows the selected text)
4. At the beginning of the text, insert the letter "ù" a few times.
This sequence would result in the cursor_position property get out of
sync. Since it represents a byte offset in the utf-8 text, it would
point straight into a utf-8 sequence, resulting in a panic when the
renderer would try to create a slice of the selected text.
As a remedy, this patch renames the cursor_position and anchor_position
properties and provides wrapper getter accessors for use within the
TextInput to safely bound the stored offset.
Eventually we need to intercept changes to the text property and adjust
the cursor and anchor position properties accordingly, so that they
remain valid. However this requires that the compiler is aware that a
property is not accessible directly but has a setter that needs to be
called instead.
This likely/hopefully also fixes#1707.
This removes the special code for the generated property getters and
ensures type safety in the run-time library for property value setting.
In the Rust generated code we continue to do arithmetic on the scalar
values, that means we immediately extract the scalar, do arithmetic and
rely on the compiler to only allow compatible units.
Danger zone alert: In the interpreter Value::Number can now be converted
to LogicalLength as-is.
- On the web, to return the image size, we need to use the natural size
of the image, and not its dom size, as the later get modified since
commit b727aba4a0
- The target size did not take in account the image fit, that's because
former version of resvg could only render by respecting the aspect
ratio. But since the web don't have this limitation, we now need to
take it into account. And new version of resvg can also scale with
any aspect ratio
This avoids accidental use of logical pixels, esp. for SVG, elimiates
one call to `to_untyped()` as well as an untyped scale factor in the
femtovg renderer.
