On Tue, 8 Nov 2022 18:31:33 GMT, Andy Goryachev <ango...@openjdk.org> wrote:
>> Not 100% sure about this. It sounds somewhat logical. If we set the left >> anchor to be 5. Do we expect it to be 10 with a render scale of 2 -> 200% >> resolution? > > Not sure about render scale - it should be done elsewhere. > > As for snapping, I think this code might be incorrect - it should honor > Region.snapToPixel setting. All the intermediary computations need to use > unsnapped (original) values, then, at the last moment, call proper > Region.snap*() which either rounds or not, depending on the `snapToPixel`. > > The only exception for this rule (of using snapped values in the intermediary > computations) is to ensure some kind of specific alignment, which I believe > is not the case here. I'm not sure about this. Other places do the same, and there are lot of places actually which do something like `snappedLeftInsets` + `snappedRightInsets` or `final double w = snapSizeX(getWidth()) - x - snappedRightInset();` or similar. I think this was for a reason. And I have only oriented myself to the existing code in this case. If you are right, then this was done wrong everywhere. Btw. I had a look and other container do snap their constraints, e.g. `H/VBox` its spacing, `GridPane` his `H/VGap`. So pretty sure we also should snap the anchor values. ------------- PR: https://git.openjdk.org/jfx/pull/910
