| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| |
|
|
| |
This type is used by the Request::ShowCursor enum variant.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The previous version of the library passed events to an implementation
of a WindowController trait by calling the trait method associated with
each event, and received requests by calling different trait methods and
reading the returned values. This had the downside of requiring that any
data received from one event had to be stored in the type so that it
could be passed back to Phosphor when a request method was called.
The new library structure uses a single handle_event method on a trait,
which is passed data representing any single event when it is called.
Data is returned via a passed mutable reference to an EventQueue,
meaning that any number of responses for any event can be immediately
returned to Phosphor without having to wait in storage.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
When another window was dragged over a Phosphor window, the Phosphor
window would have streaks drawn over it from the dragged window and
would need to be redrawn. This was handled by an unrequested render
request, where a render request is received that was sent not by the
program itself but by Phosphor.
These unrequested requests were being ignored due to commit 30c61fd,
because of an issue on Windows where programs were receiving upwards of
50 unrequested requests per second for no reason, which was impacting
program performance. They have been re-enabled, pending an investigation
into the cause of these requests on Windows.
|
| | |
|
| |
|
|
| |
It makes sense to handle all cursor icon state at once.
|
| |
|
|
|
|
| |
This commit reverts commit 3d44c40. Calling window.title() was causing
the X11 display server to go from 1% to 20% constant CPU usage when
testing on Linux.
|
| | |
|
| |
|
|
| |
We don't need to keep track of the previous value of the window title.
|
| |
|
|
| |
Program windows can choose to show as fullscreen or windowed.
|
| |
|
|
|
|
| |
Setting the window title every frame was causing Windows Explorer to
show a consistent 4% CPU usage when testing under Windows 10. This
change brings it down to 0.2%.
|
| |
|
|
|
|
| |
Rapidly modifying the window properties was causing the cursor to
vanish when attempting to drag-resize the window, and was preventing
the window from correctly maximising when double-clicking the title bar.
|
| |
|
|
|
|
|
|
|
| |
Unrequested renders are detected and ignored to prevent unnecessary
rendering work. This was accomplished by keeping track of not just the
current render hint, but also keeping track of whether a render has
been requested at all.
Unrequested renders have been observed while testing on Windows 10.
|
| | |
|
| |
|
|
|
| |
I can't remember exactly what these changes were added to fix, it's
been a long time since I wrote them
|
| | |
|
| |
|
|
|
| |
Previously, scroll values would increase in the upwards and leftwards
directions. They now increase in the downward and rightward directions.
|
| | |
|
| |
|
|
|
| |
Dependencies are now fetched from the benbridle.com git server, instead
of Ben's local filesystem.
|
| |
|
|
|
|
| |
Previously, this crate was defining its own set of aliases for types
from the geometry crate, which were clashing with the aliases already
defined by the buffer crate. The buffer aliases are now used instead.
|
| |
|
|
|
|
|
| |
This commit forwards the winit display server feature flags as phosphor
feature flags, so that downstream software can choose which display
servers to enable or disable support for. Disabling support for a
display server has the effect of reducing compile times and binary size.
|
| |
|
|
|
| |
Libraries are never supposed to track the Cargo.lock file, it was tracked
in this repository a while ago by mistake.
|
| |
|
|
|
|
|
|
|
| |
If an application attempted to change either of the window dimensions to
zero, the surface would resize to the new dimensions and cause the window
definition in the display server to enter an invalid state, causing the
display server to crash the application.
This change ensures that the window dimensions are always greater than zero.
|
| |
|
|
|
|
| |
Previously the window would only be redrawn when the physical dimensions
changed, which meant that the effects of a pixel scale change would not
be visible until the application requested a re-render.
|
| |
|
|
|
|
|
| |
The code to scale the window buffer and draw it to the window surface
has been made more efficient through the use of slice::copy_from_within
and slice::fill, instead of what was slow pixel-by-pixel iteration over
long horizontal strips.
|
| |
|
|
|
|
|
| |
Previously the physical cursor position instead of the logical cursor
position was being passed to WindowController::on_cursor_move, which
meant that the cursor position that the application was receiving would
be incorrect when the window scale was greater than 1.
|
| |
|
|
|
|
|
|
| |
When resizing a window, the window width and height will be made to be a
multiple of the current pixel scale. This is to prevent thin black strips
from rendering along the bottom edge and the right edge of the window
when the dimensions of the window exceed the dimensions of the scaled
screen content.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
A window now can declare a scale factor to be used when rendering logical
pixels to a physical window. Each logical pixel will be drawn as an NxN
block of physical pixels, where N is the scale factor.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
This has been a long-awaited task, the code has been accumulating small
changes for a while now. This commit consolidates all these changes in
order to make the code more readable and maintainable for the future.
Notable changes:
- Remove the concept of a ProgramController
- Remove all of the dead OpenGL stub code
- Update winit to version 28.1, from 27.4
- Use softbuffer for writing pixels to the native display server
|
| |
|
