"empty" example crashes using LayerRasterSkija in Linux

[chirontt]

When on Linux box, and using the LayerRasterSkija, rather than the default LayerGLSkija class, the "empty" example crashes at runtime.

Here are the relevant code of the "empty" example:

JWM/examples/empty/java/Example.java

Lines 18 to 22 in d5fe7cb

	try {
	window.setLayer(new LayerGLSkija());
	} catch (Throwable e) {
	window.setLayer(new LayerRasterSkija());
	}

The above code is working fine in Linux, because the LayerGLSkija class works fine so the LayerRasterSkija never has a chance to be used.

But when you deliberately force the LayerRasterSkija class to be used, like this change:

//        try {
//            window.setLayer(new LayerGLSkija());
//        } catch (Throwable e) {
            window.setLayer(new LayerRasterSkija());
//        }

then the JVM would crash at runtime when trying to create a new instance of LayerRasterSkija class (?), and produce a crash log file.

The relevant info in the crash's log are here:

...
Stack: [0x00007e67a4f00000,0x00007e67a5000000],  sp=0x00007e67a4ffde58,  free space=1015k
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C  [libX11.so.6+0x282b4]  XInitImage+0x4
Java frames: (J=compiled Java code, j=interpreted, Vv=VM code)
j  io.github.humbleui.jwm.LayerRaster._nResize(II)V+0
j  io.github.humbleui.jwm.LayerRaster.resize(II)V+35
j  io.github.humbleui.jwm.skija.LayerRasterSkija.resize(II)V+22
j  io.github.humbleui.jwm.Window.accept(Lio/github/humbleui/jwm/Event;)V+50
j  io.github.humbleui.jwm.Window.accept(Lio/github/humbleui/jwm/Event;)V+115
j  io.github.humbleui.jwm.Window.setLayer(Lio/github/humbleui/jwm/Layer;)Lio/github/humbleui/jwm/Window;+63
j  io.github.humbleui.jwm.examples.Example.<init>()V+41
j  io.github.humbleui.jwm.examples.Example.lambda$main$0()V+4
j  io.github.humbleui.jwm.examples.Example$$Lambda+0x0000000064041578.run()V+0
j  io.github.humbleui.jwm.App.lambda$start$1(Ljava/lang/ClassLoader;Ljava/lang/Runnable;)V+43
j  io.github.humbleui.jwm.App$$Lambda+0x00000000640491e0.run()V+8
v  ~StubRoutines::call_stub 0x00007e678f85ffa6
j  io.github.humbleui.jwm.App._nStart(Ljava/lang/Runnable;)V+0
j  io.github.humbleui.jwm.App.start(Ljava/lang/Runnable;)V+17
j  io.github.humbleui.jwm.examples.Example.main([Ljava/lang/String;)V+5
v  ~StubRoutines::call_stub 0x00007e678f85ffa6

siginfo: si_signo: 11 (SIGSEGV), si_code: 1 (SEGV_MAPERR), si_addr: 0x0000000000000028
...

Any idea what happens? This crash only happens in Linux; the Windows box works fine for both LayerRasterSkija and LayerGLSkija classes.

(I don't know about the Mac box.)

In fact, the issue happens to all/any subclass of LayerRaster, not just the Skija one. This is how I discovered the issue which happened in my "graphics2d" example in Linux :-(

[dzaima]

Moving the window.setLayer after the window.setWindowSize call seems to fix it. Otherwise it calls void resize(int width, int height) with -1 by -1. Clamping those to 0 makes it work, though perhaps doing that (or skipping the unnecessary 0×0 resize entirely?) should be done at a higher level.

LayerGL also tries to glViewport(0, 0, -1, -1);, but is fine with that.

[chirontt]

Thanks @dzaima for your quick fix. After applying your fix, my "graphics2d" example is now working in Linux!

The fix works fine, but I'm not sure the fix (i.e. filter out the invalid width/height values before resizing the window) would be at the right place. Specifically, where are the -1 value defined for the width/height as default in the first place? I think if we can find where the -1 value is set, it would also be the ideal place to filter out the value before resizing. Just a thought...

[dzaima]

Hmm, perhaps it's not the best place; the -1s are a linux-specific thing apparently. Not sure what consequences there might be from changing those to zeroes. (ideally there'd just not be any layer logic happening until the window is fully initialized, but that's probably rather non-trivial to ensure?)