RFR: 8276058: Some swing test fails on specific CI macos system [v2]

[Sergey Bylokhov]

On Fri, 29 Oct 2021 03:49:04 GMT, Phil Race <prr at openjdk.org> wrote:

> Well .. if you think that's what's needed I suggest you ask the fixer if they'd prefer to do that themselves first as the owner of this issue.

This issue is broader than the cursor, not sure that the test which needs a "tolerance" will be fixed by the hidden cursor. So the cursor is a separate part and could be fixed and especially tested separately. I am still not sure why some of the tests do not work as-is and we need to tweak the scale/tolerance/etc. probably that also some kind of bugs.

> Whoever does it there's a bunch of testing needed to validate it including testing both the cursor size and the screen capture option

Sure, I will run some of the combinations and something can be covered by the mach5.

> As to the suggestion although Robot is 99% used for testing and no tester will care at the cursor blinking off and on again I know there are cases where people have used it for other things and I don't know how long it takes to capture the screen of a 5K retina display. I would want to be very sure it is necessary to sacrifice usability by invoking this on all Robot API calls to compensate for the as yet not understood cases where this is happening even though it should not be. The "spec" issue can be handled by an update there. We can even add new API to request (not guarantee) that the cursor be hidden.

We cannot update the spec in the update releases, and the new behavior could be considered as a regression from x64 to aarch64. But yes the possible change of performance should be considered and measured.

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

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