SDL3: IME regressions

[bdach]

Should arguably only do the IME part.

Discussed in #28162

^{Originally posted by cdwcgt May 12, 2024}

also for beatmap search

[bdach]

Seems like before ppy/osu-framework#6234 we were able to suppress IME events from emitting SDL key down events and now we're not.

@Susko3 any bright ideas? other than scheduling (please god no)?

[Susko3]

This is a timing bug related to windows raw keyboard. Since the events are very low latency, the keydown event is handled before TextBox has a chance to block it because of an ongoing composition.

Disabling SDL_HINT_WINDOWS_RAW_KEYBOARD works around the issue, but that increases keyboard latency by a tiny bit.

Best hack I can think of is to add back keyProducesCharacter() and unconditionally handle those keys.

With ppy/osu-framework#5790 / ppy/osu-framework@6fb7adc we could do it in a smarter way -- check whether the character produced by the key is handled by the textbox.

diff --git a/osu.Framework/Graphics/UserInterface/TextBox.cs b/osu.Framework/Graphics/UserInterface/TextBox.cs
index 472272b2d..899ca32f7 100644
--- a/osu.Framework/Graphics/UserInterface/TextBox.cs
+++ b/osu.Framework/Graphics/UserInterface/TextBox.cs
@@ -1143,7 +1143,7 @@ protected override bool OnKeyDown(KeyDownEvent e)
             textInputScheduler.Update();
 
             // block on recent text input *after* handling the above keys so those keys can be used during text input.
-            return base.OnKeyDown(e) || textInputBlocking;
+            return base.OnKeyDown(e) || textInputBlocking || CanAddCharacter(e.Character);
         }
 
         /// <summary>

[bdach]

With ppy/osu-framework#5790 / ppy/osu-framework@6fb7adc we could do it in a smarter way

I'm all for this but ppy/osu-framework#6229 is a prerequisite for that pull is it not? That has unresolved reviews still.

[Susko3]

Disabling SDL_HINT_WINDOWS_RAW_KEYBOARD works around the issue, but that increases keyboard latency by a tiny bit.

The hint is now disabled by default. There were other issues with raw keyboard, so the SDL3 folks decided to disable it.

Also, I'm unable to reproduce the original issue, even with envvar SDL_WINDOWS_RAW_KEYBOARD set to 1.