The instability came from the way Windows sets window's default position. When it is not predefined on Stage creation (which was the case in these Robot tests) the window is created at some arbitrary position chosen by Windows near top-left corner. It also seems like Windows first picks the position assuming 100% scaling and then multiplies it by UI scale, which can provide X/Y coordinates of Stage with a fractional value.
Due to above behavior, there is a chance that the fractional part of X/Y will be above .5 which fails the test (X/Y values were fetched via casting directly to `int`, which drops the fractional part without rounding). Adding rounding makes the test always pass and the assumption in the comment below consistent. `unstable.test` property check was removed, since this change stabilizes the test and its results. Verified also on macOS to ensure the change did not affect the tests. ------------- Commit messages: - Round stage X/Y coordinates in testPixelCaptureAverage Changes: https://git.openjdk.org/jfx/pull/1242/files Webrev: https://webrevs.openjdk.org/?repo=jfx&pr=1242&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8255079 Stats: 7 lines in 1 file changed: 0 ins; 4 del; 3 mod Patch: https://git.openjdk.org/jfx/pull/1242.diff Fetch: git fetch https://git.openjdk.org/jfx.git pull/1242/head:pull/1242 PR: https://git.openjdk.org/jfx/pull/1242
