RFR: 8276058: Some swing test fails on specific CI macos system

[Sergey Bylokhov]

On Thu, 28 Oct 2021 03:13:21 GMT, Phil Race <prr at openjdk.org> wrote:

> Well .. the cursor issue is a suggestion for root cause [*] Do you have any other ideas ? I mean other than ...

As noted above our spec mention that the cursor should not be present on the screenshot, and we use the appropriate native API to achieve this. It is hard to say why it does not work, but since the problem is consistent which is great, then the best way to investigate it is by connecting to the system and debugging the problem.

 * Since in some cases the getPixelColor() method is used it could be related to the https://github.com/openjdk/jdk/pull/5864, where I have proved that this method may return the garbage instead of correct data. Without that fix the usage of that method is UB. But as mentioned in some other JBS, it may not be enough, probably some other bugs exist.
 * Maybe it is a timing issue like this https://github.com/openjdk/jdk/pull/5373 but in OGL this time and we draw a garbage for a moment when we did a screenshot.
 * it could also be related to the video card, so on some systems, we use one code path and the other on another system. By using metal we can prove that the bug is in the robot. And by using the native app we can prove is a bug in the robot implementation or not.
 * It also could be related to the system DPI/screen resolution, how it will work if the test will run using scale=1?

> The awt tests that are failing seems to have different root cause and will be handled separately.

In general, it does not look like a test issue, more like a product bug. So the root cause of this particular issue should be investigated.

-------------

PR: https://git.openjdk.java.net/jdk/pull/6140