incorrect Nifty GUI layout in full-screen LWJGL v3 context

[stephengold]

I believe I found a regression in JME 3.6. It occurs when Nifty GUI is running in full-screen LWJGL v3 context. Nifty seems to believe the mouse cursor is about 100 pixels above where it appears onscreen.

I've observed this phenomenon in several apps. It is particularly noticeable in the TestIssue1013 app of jme3-examples. Clicking on the "Restart Context" button has no effect. In order to highlight the button, the cursor must be placed about 100 px above the button. Clicking in that area of the GUI will trigger a restart.

JME 3.5.2 doesn't have this issue, but I've found it in "master" branch as far back as May 2022. I'm attempting to determine which commit introduced the regression.

[stephengold]

The regression occurred with PR #1750 .

[Ali-RS]

I can not reproduce it.

[stephengold]

I'll attempt to characterize it further and determine the root cause.

[stephengold]

Something in PR 1750 altered how the TestIssue1013 GUI gets laid out. Prior to 1750 it looked like this:

After 1750 it looked like this:

[stephengold]

Since PR 1750, there's a situation during initialization when the height of the viewport changes from 1002 to 1080. My suspicion is that the GUI gets laid out for a display 1002 pixels tall.

[stephengold]

Yes, in NiftyJmeDisplay.initialize() line 279, this.height gets set to 1002.

[Ali-RS]

After many attempts, I randomly was able to reproduce this issue one time!

[stephengold]

I suspect this is related to #1793. If the display size changes, JME is supposed to notify Nifty so it can re-do the layout.

[stephengold]

I changed the title because I now believe the mouse positions are correct; mouse clicks miss the button because the button is displayed in the wrong location.

[Sailsman63]

Just to add a datapoint, now that I understand how to get lwjgl3 running for the tests:

I thought that I wasn't seeing this, but it is happening for me. It's just harder to make out because I'm getting the full effect of #1013 itself. The blank-screen effect takes over, shifted down by about the height of the button. Still need to target the original, correct location of the button to get additional context restarts. (Confirmed restarts are happening from terminal output logging.)

• Java 11 & 16, Arch Linux
• Running on an ancient Nvidia GPU, Nouveau drivers.

[stephengold]

The workaround for #1793 was integrated to "master" branch on 5 April 2022. Yet even after that commit, this issue still occurs. Here's why:

• NiftyJmeDisplay has two initialize() methods, which are invoked at different times.
• The 4-argument method gets invoked first. It sets the h field (cached window height) to 1002.
• The 2-argument (SceneProcessor) method gets invoked later, during LwjglWindow.updateSizes(). It configures InputSystemJme for a height of 1080, but doesn't update the h field.
• The cached height is used to lay out the GUI.

I believe the simplest fix will be to update h in the 2-argument initialize() method. I'll submit a PR.