Use JNU_LOG_EVENT(env, msg, ...) to save a message in the internal
JVM ring buffer that gets printed out to the Events section
of the fatal error log if JVM crashed.
- Adding information to WLPointerEvent about wl_pointer::axis* events along the X axis;
- Introducing 'WLComponentPeer#convertPointerEventToMWEParameters' - a routine for converting WLPointerEvent parameters to parameters required for MouseWheelEvent s;
- Handling both X and Y axes within the WLPointerEvent dispatching routine.
(cherry picked from commit 6afbc34d4b)
- Remaking the mapping of wl_pointer::axis events values to MouseWheelEvent rotations to eliminate the touchpad scrolling behavior "the more slowly the fingers move, the more pixels are scrolled";
- Accumulating the fraction parts of wl_pointer::axis events values to improve the accuracy of touchpad scrolling;
- Distinguishing between wheel scrolling and touchpad scrolling to fine-tune MouseWheelEvent parameters for each of these cases.
(cherry picked from commit 874d5698bc)
Swing requires that clicking frame decorations should cause the window
to be ungrabbed. However, if a custom title is used, and that title contains
user-provided components, then clicking such components should not
cause the window to be ungrabbed, otherwise a menu located in a custom
title behaves incorrectly.
Fix by using the same logic as for the native actions, such as moving the window.
If the native actions are allowed, then ungrabbing is allowed as well.
Otherwise, do not ungrab, let the component behave like it's located in the client area.
The fix is supplemented with a new regression test "test/jdk/jb/javax/swing/CustomTitleBar/JMenuClickToCloseTest.java".
AppKit resets the cursor on native mouse entered/exited events. Depending on the order of events, it may end up setting the wrong cursor. So update it forcibly on such events.
To fix missing mouse entered/exited events when
using rounded corners, we keep track of mouse moved events. When a mouse moved event is detected, and the current peer under the cursor belongs to a different window, we send fake mouse entered/exit events to the old and new windows. We also filter late mouse exited events.
The workaround is enabled by default with the VM option "awt.mac.enableMouseEnteredExitedWorkaround" to disable it in case something breaks.
About the test:
Use the robot to find the points when the mouse
entered event is sent to the popup when the mouse
enters through a rounded corner, and the similar
point for entering the outer window when exiting
through such a corner.
Once the points are found, move the mouse back
and forth to that point, but not beyond.
The correct behavior is that when the mouse
enters the popup, a mouse exited event is sent
to the outer frame and vice versa.
Therefore, every mouse entered/exited event
should be received exactly once.
Use reflection to set the rounded corners,
as JBR API isn't available in tests.
- Added proper synchronization and image layout transitions.
- Refactored VKRenderer to hold per-device rendering context. Isolated surface rendering contexts.
- Implemented reusing of command buffers and semaphores
- Fixed surface resize, made surface initialization more robust.
- Added on-demand pipeline creation for actual surface formats.
- Added missing destruction logic.
- Added macros for easy checking of return codes, logging with source code location.
- Moved implementation details out of headers where possible. Stripped dead code.
- Implemented consistent OOM strategy from dynamic arrays and ring buffers.
Certain Wayland compositors (wlroots) invalidate xdg_surface after window with popup loses focus. Subsequent attempts to
commit the popup window fail with protocol error "xdg_surface has never been configured".
Handle popup_done event by hiding the popup window. Also emit WINDOW_CLOSING event as otherwise focus remains on the popup parent.
Co-authored-by: Maxim Kartashёv <maxim@kartashev.spb.ru>
- based on common AccelTexturePool
- new VKTexturePool instance in VKLogicalDevice
- fixed SIGSEGV in VKImage dispose
- store device in TPI
- indentation fixes
- merged with latest changes for JBR-7460
- use (ATexturePoolLock_init)(void)
- fixed logs in lock implementations + fixed indentation
- fixed MTLTexturePool to pre-processor conditions (not runtime) on USE_ACCEL_TEXTURE_POOL
