widgets: fall back to Xwayland #234
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #234 +/- ##
==========================================
- Coverage 93.17% 93.16% -0.02%
==========================================
Files 58 58
Lines 11051 11064 +13
==========================================
+ Hits 10297 10308 +11
- Misses 754 756 +2 ☔ View full report in Codecov by Sentry. |
|
This doesn't seem to work. For some reason it still connects to Wayland anyway. It does it when I set |
|
Defaulting to |
OpenQA test summaryComplete test suite and dependencies: https://openqa.qubes-os.org/tests/overview?distri=qubesos&version=4.3&build=2025022311-4.3&flavor=pull-requests Test run included the following:
New failures, excluding unstableCompared to: https://openqa.qubes-os.org/tests/overview?distri=qubesos&version=4.3&build=2025021804-4.3&flavor=update
Failed tests9 failures
Fixed failuresCompared to: https://openqa.qubes-os.org/tests/129058#dependencies 15 fixed
Unstable testsPerformance TestsPerformance degradation:15 performance degradations
Remaining performance tests:57 tests
|
|
Is it necessary for clicking outside the menu to close the menu, or is it good enough to require clicking again on the widget? The latter might be fixable in the menu itself, by closing the menu when the widget is clicked. |
DemiMarie
force-pushed
the
widgets-xwayland
branch
from
9dee5a2 to
18901a0
Compare
It is necessary, it is a dom0 widget, it should not have the bad behavior of the vm widgets (which also should work better wrt to closing, but that's out of scope here). |
Can you try my most recent commit? That dismisses the menu when it loses focus. |
|
I don't think this addresses the root cause, which is some wayland failure - this works correctly under X and so, from what you said previously, should work correctly under XWayland. I don't think we want to patch the whole world, if it requires additional workarounds like this (which BTW duplicate what the menu generally does on its own) I don't think this is a good solution. |
DemiMarie
force-pushed
the
widgets-xwayland
branch
from
5d08789 to
d48cfae
Compare
DemiMarie
force-pushed
the
widgets-xwayland
branch
from
d48cfae to
b8abcd6
Compare
|
pylint complains a lot, and also rpm fails to build (%files not updated) |
|
No, unless you figure out a way to communicate to the user clearly and without superfluous clutter why the cursor looks like it does. |
This has two parts: 1. Set GDK_BACKEND=x11, so that GDK3 chooses to use X11 as its backend. This must be done while only one thread is running, as otherwise undefined behavior results. It also must be done before gi.overrides.Gdk is imported, as otherwise it will be too late to set the GDK backend. Therefore, this code is run very early, and RuntimeError is raised if the preconditions are violated. 2. Create a fullscreen invisible window for mouse input. This works around Xwayland not passing all pointer input to X11. The menu is dismissed if the user clicks on the fullscreen window. This hack is only used if WAYLAND_DISPLAY is set.
DemiMarie
force-pushed
the
widgets-xwayland
branch
from
e12308e to
77d9920
Compare
|
This version is tested and functional. |
The widgets do not work under Wayland, so just use Xwayland for now.
If GTK somehow still connects to Wayland despite WAYLAND_DISPLAY being unset I will not be happy.
This is broken under KDE because of a bug in KDE (https://bugs.kde.org/show_bug.cgi?id=468085), but that can be fixed.