On Tue, 2 May 2023 15:38:06 GMT, Andy Goryachev <[email protected]> wrote:
> > Okay, I will investigate, I have an idea what this can be, just very
> > surprised that the rounding fixes would cause any issues in this code.
>
> Thank you. I wonder if other places might also exhibit a similar failure
> mode. `scaledCeil` is called by `snapSizeX/snapSizeY` as well as
> `layoutInArea()` which are ... everywhere.
The problem is that the code you linked (L517 in HBox) is doing a few things
that are incorrect (and which I fixed in #1111). First, it has this line:
while (Math.abs(available) > 1 && adjustingNumber > 0) {
Here it assumes that pixels have size == 1; the correct check should be
`Math.abs(available) > 1 / renderScale`.
Secondly, it uses `snapPortionX`, which are package private methods (luckily)
which have an interesting side effect. They don't just round "down" (floor for
positive, ceiling for negative) but they also have a minimum value (1 or -1) --
this is incorrect for two reasons: a round down function can also round to 0
(which I corrected), and secondly, using the constant 1 is incorrect because
that should be `1 / renderScale`.
>From what it looks like, these `snapPortion` functions are only used by
>HBox/VBox, and I think it would be best to correct the callers here to not
>rely on the behavior that the smallest values that can be returned are 1 or -1
>(also because those are simply incorrect).
-------------
PR Comment: https://git.openjdk.org/jfx/pull/1118#issuecomment-1531696750
