RFO: GUI tests for wakefield

[Phil Race]

setVisible isn't required to be synchronous like that.

So tests that expect something to be actually on the screen need to wait 
and/or check.

Over the years, we've stabilised many tests in this regard. It is an 
ongoing process.

Having said that, I have to ask why Wayland is a special case.

Can you compare the X[wayland] vs wayland steps for window display ?

-phil.


On 12/10/2025 5:09, Maxim Kartashev wrote:
> Hi, All!
>
> Many tests in test/jdk/javax/swing/ or awt/ that we run turned out to 
> be not very useful: they create some UI and then immediately return 
> from main(). If such a test is executed under jtreg, jtreg will almost 
> immediately call System.exit() and terminate the whole test. One 
> concrete example 
> is test/jdk/javax/swing/PopupFactory/8048506/bug8048506.java, but 
> there are many more like this.
>
> Currently, running such tests on Linux/Xorg may cause a window to 
> flash briefly, which could be enough time for the test to actually 
> complete (but maybe not). Under Wayland, however, showing a window on 
> the screen requires an involved, asynchronous process of negotiation 
> with the server; this happens on EDT after setVisible(true) has 
> returned already. Consequently, such tests under Wayland usually don't 
> run long enough for a single pixel to be shown on the screen. The 
> process exits at around the time the Wayland server agrees about the 
> size and the role of the window.
>
> The specification of Component.setVisible() is vague, so I defer to 
> the expertise of people on this list to determine which part is at 
> fault and needs to be "fixed". Should such tests wait before exiting 
> main()? Or should the Wayland toolkit guarantee that the window is 
> indeed on the screen before setVisible(true) returns?
>
> Thanks in advance.
>
> Maxim.