Tooltips stops working after writing code

[Giganzo]

Tested versions

Godot v4.4.dev7

Doesn't look like it happens on v4.4.dev6

System information

Fedora Linux 40 (KDE Plasma) on Wayland - Wayland display driver, Single-window

Issue description

Tooltips stops working after writing some code
Have only seen it happen with prefer Wayland on, so not sure if it can happen with other settings. It doesn't happen every time you write code, so might need to write more or less code until the bug triggers.

Screencast_20241220_043013.webm

Steps to reproduce

Write code and then try to hover over something that triggers a tooltip with documentation help.

Minimal reproduction project (MRP)

N/A

[Giganzo]

Is this something that should be added to release blockers?
This bug is always triggered after working in a project and writing some amount of code. So if you need help from a tooltip you have to restart the editor to get them back.

[akien-mga]

Is this something that should be added to release blockers?

Only if it's a regression, and even then, Wayland support is still experimental so I don't think known issues with opting into it would be considered release blocking for 4.4.

[Giganzo]

Only if it's a regression,

Haven't encountered it before 4.4.dev7

[Riteo]

Hi, thank you for your report! I'm currently looking into it. It looks like some keys are getting "stuck" for some reason.

[Riteo]

Welp that was quick. It looks like every single key got stuck in a phantom state that apparently only specific things cared about if the window got unfocused while said keys were pressed. I assumed that the compositor would release them at the right time but I was wrong.

The fix is, like, a single line lol. PR incoming!

Also, FTR, I don't think this is a regression, I always assumed this behavior :P

[Riteo]

The PR is in, please confirm whether that fixes the issue for you :D

#101819

[Giganzo]

Also, FTR, I don't think this is a regression, I always assumed this behavior :P

I see
But it's strange that I cannot trigger it from 4.3 to 4.4.dev6 then. Something must have changed in dev7 to make it have this effect on tooltips now.

The PR is in, please confirm whether that fixes the issue for you :D

No still same problem.

With fix:

Screencast_20250119_182303.webm

4.4.dev6:

Screencast_20250119_182835.webm

There is a difference with the fix, if I tab to another window and back the tooltips will be back. In beta1 I have to restart the editor.

[Riteo]

@Giganzo I see, thank you for testing the change further. The fact that things work again when you focus/unfocus implies that some buttons still get stuck; not sure how though.

[Riteo]

All right, I think I found the final culprit and it's actually quite crazy.

This diff will tell us any stuck keys:

diff --git a/core/input/input.cpp b/core/input/input.cpp
index 4413c426cf..27dfeb12a5 100644
--- a/core/input/input.cpp
+++ b/core/input/input.cpp
@@ -296,6 +296,22 @@ bool Input::is_anything_pressed() const {
 	}
 
 	if (!keys_pressed.is_empty() || !joy_buttons_pressed.is_empty() || !mouse_button_mask.is_empty()) {
+
+		if(!keys_pressed.is_empty()) {
+			print_line("KEYS PRESSED:");
+			for (const Key &key : keys_pressed) {
+				print_line(key);
+			}
+		}
+
+		if(!joy_buttons_pressed.is_empty()) {
+			print_line("JOY BUTTONS PRESSED");
+		}
+
+		if(!mouse_button_mask.is_empty()) {
+			print_line("MOUSE BUTTON MASK");
+		}
+
 		return true;
 	}

After some keyboard mashing it reported only 95, which meant an underscore (_)

I could consistently replicate the issue by:

1. Pressing Shift and -, resulting in _, and let the key repeat do its job for a second or so;
2. While keeping the - key pressed, releasing Shift, and let they key repeat do its thing a bit more;
3. Releasing everything.

I'm quite sure this is related to #99633. That PR correctly makes keys fully Unicode aware but some code somewhere still assumes that the physical key matters, which is not always true.

Update: I think it's the key repetition code that does not handle keyboard state changes properly.

[Riteo]

Hi, I pushed a new update which should fix the issue once and for all: #101848

See also issue #100879.