For users with multiple keyboard layouts configured, glfwGetKeyName
works fine only with the primary layout. Switching layouts results in
changing the group index. This commit querries the current group index
when initializing keyboard input and keeps track of any change to it.
As a result the scancode -> keyname mapping may change while the program
is running (needs to be documented).
Fixes#1462.
Closes#1528.
MSVC 2019 complains that the code at line 1744
(`GetMonitorInfo(window->monitor->win32.handle, &mi);`) can potentially
dereference a null pointer. The compiler is wrong in this case (it has
not spotted that `monitor` and `window->monitor` must be equal), but
I think it makes sense for our non-NULL test to be on the variable we
actually use rather than the one it was set from.
Related to #1491.
The EWMH feature detection atoms are now named and loaded the same way
as other X11 atoms. Detection is now performed after all
non-conditional atoms have been loaded. The EWMH detection now has
hopefully more readable comments.
This protocol is part of the core Wayland, but it is pretty badly
designed and is missing quite a few features, and is in the process of
being phased out in compositors. Its support in GLFW requires
duplicating pretty much every single window management codepath.
This bumps the required compositor versions to the ones which have
implemented xdg-shell, approximately two years ago, which seems sensible
to me.
Window relative mouse locations provided via NSWindow and NSEvent are
based at 0,1 while screen relative locations use 0,0. Incorrect
handling of this had crept into other coordinate transformations. Note
that most of these errors canceled each other out, so the reported
positions of windows, monitors and work areas are unaffected. This
corrects the cursor position for glfwGetCursorPos and glfwSetCursorPos.
Fixes#1461.
DWM composition is always enabled on Windows 8 and later, so there's no
need to call DwmIsCompositionEnabled on those platforms.
Fixes#1452.
Closes#1453.
This solution of one display link per window is far from ideal but is
still better than no solution.
As a side-effect this fixes swap interval breaking being ignored for
occluded windows on earlier versions of macOS.
Fixes#680.
Fixes#1337.
Related to #1417.
Fixes#1435.
This fixes OpenGL swap interval (vsync) on macOS 10.14 Mojave by using
CVDisplayLink to synchronise to the monitor refresh rate rather than
setting NSOpenGLContextParameterSwapInterval.
Solution based on advice provided by @rcgordon.
Closes#1417.
This is another small step towards having GLFW play nice with other
toolkits sharing the same process, including AppKit.
Any macOS platform function that touches Cocoa must now wrap itself in
an autoreleasepool block.
Since GLFW no longer provides an autoreleasepool outside of its
functions, THIS MAY BREAK EXISTING CODE MIXING GLFW AND COCOA. Sorry!
Please add your own autoreleasepool blocks as needed.
Fixes#1107.
Closes#1114.
This intersects the global work area from _NET_WORKAREA with the monitor
viewport. The monitor viewport falls back to the core display
dimensions where working RandR is missing. The _NET_WORKAREA query is
now checked for success. The _NET_WORKAREA extent array is now indexed
by _NET_CURRENT_DESKTOP. The _NET_WORKAREA atom is now checked for
availability.
Related to #1322.
This renames 'raw input' to 'raw mouse motion' as there are other kinds
of raw input. The update path is restructured to avoid reinitializing
all of disabled cursor mode. Modification of shared state is moved out
into shared code. Raw mouse motion is disabled by default for
compatibility.
Related to #1401.
We now keep track of the fullscreen and activated state and only iconify
if we were previously fullscreen and now we are either not fullscreen or
not activated anymore.
This is the proper way to do it, compared to the previous hack where we
didn’t iconify only if it was the first configure event received.
I was getting an error in this cmake file when using 3.10.2 on linux.
Here was the error message I was getting:
[cmake] CMake Error at external/glfw/src/CMakeLists.txt:82 (if):
[cmake] if given arguments:
[cmake]
[cmake] "STREQUAL" "GNU" "OR" "STREQUAL" "Clang"
[cmake]
[cmake] Unknown arguments specified
Adding the quotes around the cmake variables seems to do the trick. That
was also done with the STREQUAL condition earlier on line 66.
Closes#1411.
The application delegate needs to be set at init to receive monitor
events before window creation. Menu creation is moved to
applicationWillFinishLaunching: to play nicer with other toolkits in the
same process.
Related to #1317.
This makes glfwSetGamma generate a gamma ramp of the same size as the
monitor's current ramp, which will avoid failure on non-256 entry
monitors on X11 and avoid ramp interpolation on macOS.
Closes#1387.
Fixes#1388.
Commit 9c513346ad ("Gamma will never be
supported on Wayland") made it clear that it cannot be implemented, so
this removes the TODO markers and rewords the error messages.
Related to #1387.
Older versions did not provide fmin or fmax. This adds internal
versions of fminf and fmaxf that should not be confused with
standards compliant implementations.
This adds reporting of those Caps Lock key events that cause the lock
state to change.
The full fix involving IOHID is being worked on in #1368.
Related to #1368.
Closes#1373.
This removes the GLFW NSApplication subclass as a step towards better
coexistence with other libraries that touch Cocoa.
This moves application object creation to platform init to allow event
processing before window creation.
Related to #1317.
This has the advantage that the user may override e.g. the include
location, and the correct libdir (lib, lib64, lib/something) is
automatically determined.
Closes#1367.
click events would have an incorrect position after changing workspace,
if the mouse didn't move in between.
(Another example where this matters is a new window, if it appears under
the cursor, clicking would lead the application to think the user clicked
at 0,0)
As of the release of Mir 1.0, libmirclient has been deprecated[1] and
its developers recommend clients using it to switch to Wayland. This
patch removes support for libmirclient and instruct users to use the
experimental Wayland backend instead.
[1] https://discourse.ubuntu.com/t/mir-news-28th-september-2018/8184
This allows compositors which prefer to draw the decorations around
clients to do so, rather than letting GLFW draw its own decorations.
The appearance is thus entirely subject to the compositor used, but
should generally be better than the current solid colour decorations we
have, which we continue to use when the compositor doesn’t support this
protocol or tells us to draw the decorations ourselves.
This new protocol has been tested against wlroots’s rootston compositor.
Fixes#1257.
This allows the compositor to avoid having to setup and teardown a
SIGBUS signal handler whenever it needs to read from this surface, as it
knows we won’t be able to shrink the file and so doesn’t have to protect
against that.
This codepath will only be used on Linux ≥ 3.17 with glibc ≥ 2.27, and
possibly other kernels and libc. The former code will continue to be
used as a fallback, either if memfd_create() fails or if it isn’t
available.
This adds the GLFW_SCALE_TO_MONITOR window hint for automatically
resizing the content area of a window to the requested size times the
monitor content scale each time it is placed on a new monitor. This
only applies to windowed mode windows and includes the initial placement
at window creation.
This hint only has an effect on platforms where screen coordinates and
pixels always map 1:1 such as Windows and X11. Platforms like macOS
instead change the resolution of the framebuffer independently of the
window size.
Related to #676.
Related to #1115.
This adds basic support for the Per-Monitor V2 level of DPI awareness
in Windows 10, which allows for automatic DPI scaling of window
decorations.
This commit does not include resizing the window content area to match
the new window content scale.
Related to #1115.
Fixes#1294.
When mouse trails are enabled, the cursor becomes invisible when the
OpenGL ICD detects a full screen window and switches to page flipping.
Mouse trails are now disabled as long as any full screen windows are
visible.
Fixes#1263.
This adds a window hint and attribute for controlling whether
glfwShowWindow gives the specified window input focus in addition to
making it visible.
Fixes#1189.
Closes#1275.
In 32e78aeb2 the definition of DWM_BLURBEHIND in win32_platform.h was
moved behind a WINVER < 0x0600 preprocessor check (< Vista). This broke
the build for WINVER >= 0x0600 since DWM_BLURBEHIND is not defined.
Starting with Vista DWM_BLURBEHIND is available in Dwmapi.h.
So we can just include the header directly on Vista and above.
Closes#1253.
GLFW now checks for the libvulkan.1.dylib loader instead of what is now
the ICD. This removes checking for libMoltenVK.dylib to avoid cryptic
errors. This unfortunately also breaks compatibility with the
standalone MoltenVK SDK.
This also removes support for the static loader library as that is not
present in the LunarG SDK.
Related to #870.
This adds support for the + and - and ~ input modifiers for joystick
axes. It also changes how joystick axes are translated to buttons to
more closely match SDL 2.0.7.
Output modifiers are still not supported but have not yet been seen in
the wild.
Allow window creation despite video mode setting failure.
Video mode setting failure is ignored the rest of the time and the
desired video mode has never been a hard constraint anyway.
OpenGL / OpenGL ES cannot share presentation on a window with Vulkan.
This adds an error to `glfwCreateWindowSurface` when it is called on a
window without the GLFW_CLIENT_API hint set to GLFW_NO_API. This
prevents undefined bahevior and hard to debug crashes.
Fixes#1194.
Closes#1205.
This protocol matches desktops much better than the deprecated
wl_shell, fixing a bunch of race conditions, removing undefined
behaviour, adding missing features, and generally providing a much more
user-friendly experience.
Since most compositors don’t support it yet, the wl_shell_surface role
is kept as fallback for now.
The Wayland protocol is asynchronous, by the time we destroy a surface,
the compositor may have sent a wl_keyboard::enter or wl_pointer::enter
events which now point to no surface, yet we receive it after.
To prevent this race, we can just ignore any enter event targetting a
NULL surface.
Fixes#1150.
