Update modifiers before propagating key press events

[chrisduerr]

Currently, when receiving WindowEvent::KeyboardInput events, it is not possible to detect when all shift keys have been released unless manually tracking the shift key state. This is the case for all modifiers.

The cause of this is that the modifiers get updated after the events are sent, which means that shift press key events do not have the modifiers shift set, but shift release events always do have them set. As a result, it's impossible to know if RShift is still held when the released event for LShift comes in.

My proposal to solve this would be to just update the shift state before sending the key events. This would mean that when LShift comes in and the shift modifier is still set, RShift must still be held.

Of course this is a breaking change and I'm not sure what inspired the initial design, which is why I haven't looked into the code changes myself yet. But are there any specific reasons why this would be a bad idea?

[elinorbgr]

First of all, I think we could try to asses the exact current situation, because it is likely that not all platforms behave the same way regarding that... Which platforms have you observed this on?

[chrisduerr]

This was tested on Linux/X11 (i3), but I also checked Linux/Wayland (sway) and it behaves exactly the same way. So I kinda just assumed that this would be consistent across all platforms.

It seems like there's mostly macOS/Windows left only, I can take a quick look at macOS if that's of interest.

[chrisduerr]

Okay, testing macOS was a good idea. Turns out macOS behaves exactly the way I have proposed here,s it seems like this is a platform inconsistency issue.

Is there a process for standardizing things across platforms if there are inconsistencies detected between them? I'd selfishly say that if it hasn't been reported as a bug, the macOS way doesn't seem to cause any trouble and there's obviously benefits to it for me.

The current behaviour on Linux doesn't sound right to me. The proposed behaviour sounds more reasonable. I would suggest checking Windows as well. Without putting too much thought into it, I'm for standardising the macOS behaviour across all backends.

[Osspial]

I can confirm that the proposed behavior is what happens on Windows.

[goddessfreya]

cc @murarth