The calculation of framebuffer size is about to get more complicated
with the introduction of fractional-scale-v1, so only do it in one
place.
This is also a tiny preparation for eventually behaving better towards
wl_egl_window_resize.
We are about to introduce another scaling method (fractional-scale-v1),
so it will become more important to be specific about what scales are
used where and what their units are.
This removes 'content scale' from window and monitor structs. A monitor
(output) now has just a 'scale', which becomes a 'buffer scale' when
applied to a window. A window now has a list of 'output scales' to
select its buffer scale from. Content scales are calculated from the
respective monitor or window when queried, even if the calculation right
now is to just return the same value as before.
When using the fallback decorations, GLFW_HOVERED was true also when the
cursor was over one of the window decorations.
This also disentangles the GLFW_HOVERED state from the fallback
decorations. Because GLFW_HOVERED should have had the same behavior as
GLFW_MAIN_WINDOW, the latter has been removed.
We now have three window decoration paths in the Wayland backend:
libdecor decorations, XDG decorations and fallback decorations.
This makes it clearer when code relates to the fallback decoration path,
which should increasingly (hopefully) not get used.
This implements window focus requests via the xdg-activation-v1
protocol. These requests will likely only work when another window of
the same application already has input focus, but that isn't unlike the
behavior of other platforms.
The GLFW_FEATURE_UNAVAILABLE error has been removed from this function
for now.
Related to #2284
Related to #2306
Related to #2439
- Button label reflects window mode
- Window attention request button
- Slightly less bad button layout possibly
- Applicable button labels list expected delay
- Less telegraphic message for positions being unsupported
- Window size more likely to fit contents
In a Parallels VM wglGetPixelFormatAttribivARB returns fewer pixel
formats than DescribePixelFormat. This broke context creation on
Windows in Parallels since the changes in
2c0f34b60f. The previous version of the
code worked accidentally.
This adds a workaround by iterating through the minimum of both counts.
It should have no effect when running on conforming implementations.
Tested on Parallels by @ dougbinks.
Closes#2191Fixes#2406Fixes#2467
This was adapted to 3.3-stable from
00e86d4b73.
The list of compile-time dependencies on FreeBSD lacked evdev-proto.
Unlike on Linux, the input-event-codes.h header file was not implicitly
included on FreeBSD.
Fixes#2445
This was adapted to 3.3-stable from
51920ede68.
This mainly updates the version of NodeJS used internally to keep
up with what's going on at GitHub Actions.
Closes#2447
This was adapted to 3.3-stable from
becf1dc14b.
glfwGetKeyName emitted GLFW_INVALID_VALUE when passed GLFW_KEY_UNKNOWN
and any scancode not associated with a key token on that platform.
This causes physical keys with no associated key token to emit
GLFW_INVALID_VALUE when the key and scancode are passed directly from
the key event to glfwGetKeyName. This breaks the promise made in the
reference documentation for glfwGetKeyName.
This commit removes that error for the whole range of valid scancodes.
Fixes#1785Fixes#2214
This was adapted to 3.3-stable from
86bf5698ec.
Reshuffled line to group key-related items together. Omitted key code
output for unknown keys to put focus on the (lack of) name.
(cherry picked from commit d81fcb93f1)
In a Parallels VM wglGetPixelFormatAttribivARB returns fewer pixel
formats than DescribePixelFormat. This broke context creation on
Windows in Parallels since the changes in
2c0f34b60f. The previous version of the
code worked accidentally.
This adds a workaround by iterating through the minimum of both counts.
It should have no effect when running on conforming implementations.
Tested on Parallels by @ dougbinks.
Closes#2191Fixes#2406Fixes#2467
The native access functions for monitor objects did not check whether
the correct platform was initialized and would return invalid handles if
it was not.