Set the auto scale method explicitly in ImagesWin32Tests

[fedejeanne]

Rename the test class and let it test the "smooth" scale method only, which was the root cause of the problem all along.

Contributes to #1790

[github-actions]

Test Results

   502 files  ±0     502 suites  ±0   10m 38s ⏱️ + 1m 26s
 4 334 tests ±0   4 320 ✅ ±0   14 💤 ±0  0 ❌ ±0 
16 575 runs  ±0  16 466 ✅ ±0  109 💤 ±0  0 ❌ ±0 

Results for commit 7250aa9. ± Comparison against base commit 920fe85.

This pull request removes 1 and adds 1 tests. Note that renamed tests count towards both.

org.eclipse.swt.graphics.ImagesWin32Tests ‑ testImageIconTypeShouldNotChangeAfterCallingGetHandleForDifferentZoom

org.eclipse.swt.graphics.ImageSmoothScalingWin32Tests ‑ testImageIconTypeShouldNotChangeAfterCallingGetHandleForDifferentZoom

♻️ This comment has been updated with latest results.

[HeikoKlare]

Some thoughts on this:

• The test is somehow specific to "smooth" scaling (at least the test was introduced as a regression test related to that scaling mode).
• The test class is annotated with the WithMonitorSpecificScalingExtension to enable monitor-specific scaling, which, in turn, enables "smooth" scaling.

So given that, I would expect one of the following solution:

• Make the test specific to "smooth" scaling, which would include renaming the test and removing the extension
• Make monitor-specific scaling properly enable smooth mode (for which we have a dedicated issue), so that the additional setup logic for "smooth" scaling is not required.

[fedejeanne]

I went for option nr. 1:

* Make the test specific to "smooth" scaling, which would include renaming the test and removing the extension

Why?
Removing @ExtendWith(WithMonitorSpecificScalingExtension.class) makes it work a proper regression test, which means that we were testing the smooth scaling behavior all along without knowing it. I double checked this assumption by switching the VM parameter -Dswt.autoScale.method between nearest and smooth and again: nearest lets the test pass always and smooth doesn't (i.e., it behaves like a regression test).

[HeikoKlare]

Changing the test to focus on smooth scaling is fine for me, but the current solution will not work. SWT_AUTOSCALE_METHOD is read only once in the static initialized of DPIUtil. There is no guarantee that static intializer is executed after the setup method of the test class. And even if you were able to ensure that, the current restoration logic will not apply as the system property that is reset will not be re-evaluated afterwards.

[fedejeanne]

Hm, you're right: as it is, this test passes only by chance and it doesn't restore the previous auto scale method. I guess we will need to wait until the auto scale method can be properly set in runtime (if that's what we want) or we can force the auto scale method somehow, probably with a new method in DPIUtil.

I'll postpone this discussion (and this PR) until we discuss how to proceed with vi-eclipse/Eclipse-Platform#227, which is related.

[akurtakov]

What is the status of this one?

[HeikoKlare]

This still needs to be revised. Tests rely on smooth scaling being activated but smooth scaling cannot be enabeld/disabled explicitly during tests but just implicitly via the monitor-specific scaling option.

[fedejeanne]

<TL; DR>

The current status of this test class on master is OK so there is no need to apply this PR

Longer explanation

The test class is already testing the smooth method by activating the monitor-specific UI scaling via the annotation @ExtendWith(WithMonitorSpecificScalingExtension.class).
The annotation also restores the status prior to the test when finishing the test class so all is good.