On Thu, 3 Aug 2023 15:13:09 GMT, Marius Hanl <mh...@openjdk.org> wrote:
>> An IOOBE was thrown when scrolling up via the trough (-> >> `VirtualScrollBar#adjustValue`). >> This happened only when it has bigger cells than the viewport. >> If the the uppermost cell with the index 0 is only visible (although not >> completely scrolled to the top) and then an attempt is made to scroll up >> again, the `VirtualFlow` will try to scroll to the next cell, subtracting >> index 0 by 1, resulting in -1 -> IOOBE. >> >> The code now guards against any under or overflow. >> >> This is technically a regression from >> https://bugs.openjdk.org/browse/JDK-8173321 >> >> Note: While testing with very big cells, I found out that scrolling via the >> trough may not work after the first time. >> This is because the `VirtualFlow` still creates 2 cells, although only one >> can be visible at a time (at least when scrolling to the next cell) (and >> `VirtualScrollBar` does this check, which will never be true then: >> `firstVisibleCell == lastVisibleCell`). This is unrelated to this fix. I can >> create a ticket when I have more information. > > Marius Hanl has updated the pull request incrementally with three additional > commits since the last revision: > > - JDK-8311983: remove JUnit5 import > - JDK-8311983: improve exception handling > - JDK-8311983: More tests The change looks good, tested with the test app in the ticket and the Monkey Tester. I wonder if we ought to consider some kind of `dirty` flag in the virtual layout to ensure that the layout is always performed when needed. I understand this might require a larger change and it's definitely out of the scope for this PR. modules/javafx.controls/src/test/java/test/javafx/scene/control/skin/VirtualFlowTest.java line 1790: > 1788: // Scroll down. > 1789: flow.shim_getVbar().adjustValue(0.9605263157894737); > 1790: // Scroll up. FYI: adding `pulse()` here prevents the exception. We seem to have a similar situation on line 739 (this particular test is ignored, but there quite a few `pulse()`s scattered in the test code: // NOTE Run this test without the pulse and it fails! pulse(); I wonder if there is something in the virtual flow that can detect when `flow.layout()` is needed and do it automatically. modules/javafx.controls/src/test/java/test/javafx/scene/control/skin/VirtualFlowTest.java line 1799: > 1797: @Test > 1798: public void testScrollBarValueAdjustmentShouldScrollOneDown() { > 1799: flow = new VirtualFlowShim<>(); FYI: this test does not fail without the fix. should it? ------------- Marked as reviewed by angorya (Reviewer). PR Review: https://git.openjdk.org/jfx/pull/1194#pullrequestreview-1579016585 PR Review Comment: https://git.openjdk.org/jfx/pull/1194#discussion_r1294877456 PR Review Comment: https://git.openjdk.org/jfx/pull/1194#discussion_r1294877019
