On Sat, 10 Oct 2026 18:32:18 GMT, Marius Hanl <[email protected]> wrote:
> The third split PR of the CSS saga™. > > In this episode, I have a cornercase, but still a valid bug, found in > https://github.com/openjdk/jfx/pull/2225. > > In JavaFX, there is this rule where `USER < AUTHOR` style. That is, CSS will > set the property value, e.g. opacity, to the value defined in the `Scene` > stylesheet and ignore if the value was set by a setter. > > It is still possible to overwrite such value by a setter, but it will be > reapplied in the very next CSS pass. > > Now, when we apply a style in a listener which is called during CSS reset, > this style may override the CSS value. > This only happens when you style a property that is defined before the > property you are listening to. > > A CSS style reset is the operation to set the property value to the original > value because it is not styled anymore (e.g. because we removed or added a > `PseudoClass`). > > --- > > Example: `Node`'s `CssMetaData` adds `OPACITY` first and later `TRANSLATE_X`. > > So opacity is reset first, then `translateX`. If we now set the `opacity` in > a `translateX` listener, we can successfully override the `opacity`. > > The very next CSS pass will however recover this (e.g. calling `applyCss` > again). > This is not needed anymore with the fix. > > --- > > I added two tests, they are basically the same, just the order is changed. > - The first test failed before, because it sets the `opacity` in a > `translateX` listener > - The second test succeeds before and after, because it sets the `translateX` > in a `opacity` listener > > Again, shows: `OPACITY` is processed first, later `TRANSLATE_X`, which with > the fix does not matter anymore. > I reused the existing reset property logic. It is now called like before, and > as part of this fix. > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). Found a couple of possible issues (inline). Thank you for adding tests! Only one fails in master - `testStyleOverrideByTranslateListenerOnPseudoClassChange()` is this expected? modules/javafx.graphics/src/main/java/javafx/scene/CssStyleHelper.java line 995: > 993: } > 994: > 995: it.remove(); is it possible for a property to be (temporarily) non-settable? once it's removed the later css pass won't be able to restore its value. should the entry be removed only after the successful reset? modules/javafx.graphics/src/main/java/javafx/scene/CssStyleHelper.java line 1009: > 1007: } > 1008: } catch (Exception e) { > 1009: PlatformLogger logger = Logging.getCSSLogger(); this exception is only logged. should it propagate to `CssParser.errorsProperty`? ------------- PR Review: https://git.openjdk.org/jfx/pull/2343#pullrequestreview-5493899283 PR Review Comment: https://git.openjdk.org/jfx/pull/2343#discussion_r4250546115 PR Review Comment: https://git.openjdk.org/jfx/pull/2343#discussion_r4250552364
