Regressed in d6bf4fd90d.
One very visible instance of this regression is the login form.
https://github.com/user-attachments/assets/5ba10ac5-4cb1-49af-b55c-89cf58ca0b44
The `CommentEditor` usage was discovered with one of my favourite tricks
which is doing
```diff
diff --git a/osu.Game/Graphics/UserInterface/OsuTextBox.cs b/osu.Game/Graphics/UserInterface/OsuTextBox.cs
index fefe776b01..c17cca726b 100644
--- a/osu.Game/Graphics/UserInterface/OsuTextBox.cs
+++ b/osu.Game/Graphics/UserInterface/OsuTextBox.cs
@@ -42,6 +42,12 @@ public partial class OsuTextBox : BasicTextBox
Margin = new MarginPadding { Left = 2 },
};
+ public new bool Masking
+ {
+ get => base.Masking;
+ set => base.Masking = value;
+ }
+
protected bool DrawBorder { get; init; } = true;
private OsuCaret? caret;
```
and then looking for usages of the setter. That's all due diligence
you're getting here, I'm not auditing every single text box in the game.
And yes, the `CommentEditor` usage is OMEGA dodgy but the change applied
here is the only one that preserves its visual appearance. I'm not
putting in time to fix it.
This removes the disabled sound, but I think that's fine. If we want
that, it should be handled by `HoverClickSounds` (which I'm currently
intentionally not using because it can be a bit noisy).
Closes#36503.
This was attempted to be fixed by frenzibyte using some hack workaround
logic, but this is the true fix.
Things were never matching due to `UpdateSize` spamming `Resize`
transforms every frame, causing the fade out to complete before
transforms have reached a final state.
See inline commentary. I don't really have any energy left to provide
anything else, other than maybe a demonstration of how this dies in
framework:
diff --git a/osu.Framework.Tests/Visual/UserInterface/TestSceneScreenStackUnbindOnExit.cs b/osu.Framework.Tests/Visual/UserInterface/TestSceneScreenStackUnbindOnExit.cs
new file mode 100644
index 000000000..c74ce6636
--- /dev/null
+++ b/osu.Framework.Tests/Visual/UserInterface/TestSceneScreenStackUnbindOnExit.cs
@@ -0,0 +1,59 @@
+// Copyright (c) ppy Pty Ltd <contact@ppy.sh>. Licensed under the MIT Licence.
+// See the LICENCE file in the repository root for full licence text.
+
+using NUnit.Framework;
+using osu.Framework.Allocation;
+using osu.Framework.Graphics.UserInterface;
+using osu.Framework.Screens;
+
+namespace osu.Framework.Tests.Visual.UserInterface
+{
+ public partial class TestSceneScreenStackUnbindOnExit : FrameworkTestScene
+ {
+ [Cached]
+ private ScreenStack screenStack = new ScreenStack();
+
+ [Test]
+ public void TestScreenExitUnbindDoesNotInterruptLoadComplete()
+ {
+ AddStep("set up the scenario", () =>
+ {
+ Child = screenStack;
+ screenStack.Push(new Screen());
+ screenStack.Push(new BrokenScreen());
+ });
+ AddUntilStep("wait to get to target screen", () => screenStack.CurrentScreen, Is.InstanceOf<Screen>);
+ }
+
+ private partial class BrokenSlider : BasicSliderBar<float>
+ {
+ [Resolved]
+ private ScreenStack screenStack { get; set; } = null!;
+
+ protected override void LoadComplete()
+ {
+ // exiting the current screen provokes the behaviour of unbinding all bindables in the screen's subtree
+ screenStack.CurrentScreen.Exit();
+
+ // ...but the following calls should still take correct effect inside `SliderBar`
+ // (namely one consisting of propagating `{Min,Max}Value` into `currentNumberInstantaneous`)
+ // so that it doesn't have its internal invariants violated
+ CurrentNumber.MinValue = -10;
+ CurrentNumber.MaxValue = 10;
+
+ // this notably calls `Scheduler.AddOnce(updateValue)` inside, which will happen *in the imminent future, in the same frame as `LoadComplete()` here.
+ // if the above mutations of `{Min,Max}Value` don't correctly propagate inside the slider bar due to an overly eager unbind, this will cause a crash.
+ base.LoadComplete();
+ }
+ }
+
+ private partial class BrokenScreen : Screen
+ {
+ [BackgroundDependencyLoader]
+ private void load()
+ {
+ InternalChild = new BrokenSlider();
+ }
+ }
+ }
+}
I attempted to address what I perceive to be the root issue here which
is that `ScreenStack` is allowed to arbitrarily unbind bindables under
drawables which are by all means still in the scene graph. The attempt
consisted of scheduling the unbind until after children of the screen
stack, but that caused 150 game-side tests to fail, seemingly on
something relevant to bindable leases, so I give up.
- Implement IProvideCursor in SongSelect to hide cursor when background is revealed
- Cursor reappears on mouse movement and hides again after 1 second of inactivity
- Fix MenuCursorContainer to preserve drag rotation state during hide/show cycles