On Thu, 9 Apr 2026 03:27:24 GMT, Charlie Schlosser <[email protected]> wrote:
>> Fixes https://bugs.openjdk.org/browse/JDK-8181411 and >> https://bugs.openjdk.org/browse/JDK-8380308 >> >> `TreeUtil.getItem` is used pervasively throughout `TreeView` and >> `TreeTableView`, specifically `getTreeItem`. This performs a >> depth-first-search (DFS) of the tree, and caches the target item. This is ok >> for random access, but is slow for sequential access with a cold cache, as >> each subsequent call to `getTreeItem` starts a new DFS from the root of the >> tree. For `selectAll`, this has O(N^2) time complexity. >> >> This PR adds a stateful iterator that resumes the DFS from the previously >> retrieved item, and caches each item during the traversal. This reduces the >> time complexity `selectAll` case to O(N). >> >> This change revealed an otherwise harmless bug in the order of event >> handling in the `invalidated()` method of the root property. >> >> Current behavior: >> 1. `expandedItemCountChangeEvent` fires >> 2. Focus model listener fires >> 3. Focus model calls `getTreeItem`, `expandedItemCountDirty` is false, cache >> miss, correct item returned >> 5. `rootEvent` fires, sets `expandedItemCountDirty = true` >> >> The correct behavior would be to set `expandedItemCountDirty = true` before >> the focus model calls getTreeItem. It works anyway because any item that is >> not in the cache triggers a full DFS from the root, which is always valid. >> >> This PR: >> 1. `expandedItemCountChangeEvent` fires >> 2. `rootEvent` fires, sets `expandedItemCountDirty = true` >> 3. Focus model listener fires >> 4. Focus model calls `getTreeItem`, `expandedItemCountDirty` is true, >> iterator resets, correct item returned >> >> This is fixed by using `expandedItemCountChangeEvent` instead of the super >> type `treeNotificationEvent`. The subclass event types are processed before >> their super counterparts, regardless of registration order. Now that they >> are the same type, the order is correct. >> >> This behavior is already covered (and is motivated) by `TreeTableViewTest > >> test_rt17522_focusShouldMoveWhenItemRemovedBeforeFocusIndex_1()` and >> `TreeTableViewTest > >> test_rt17522_focusShouldMoveWhenItemAddedAtFocusIndex_1()`. >> >> >> >> ┌─────────┬───────────────────┬──────────────────┬────────────────────────┬───────────────────────┐ >> │ Items │ TreeView (before) │ TreeView (after) │ TreeTableView (before) >> │ TreeTableView (after) │ >> >> ├─────────┼───────────────────┼──────────────────┼────────────────────────┼───────────────────────┤ >> │ 64 │ 1ms │ 1ms │ ... > > Charlie Schlosser has updated the pull request incrementally with one > additional commit since the last revision: > > copyright update The testing (on macOS 26.3.1) looks good - a major performance improvement! There seems to be no negative impact selecting and deselecting individual items after select-all. I agree with @Maran23 about adding a few tests. modules/javafx.controls/src/main/java/javafx/scene/control/TreeTableView.java line 766: > 764: // See JDK-8125681 for the justification and performance gains. > 765: private Map<Integer, SoftReference<TreeItem<S>>> treeItemCacheMap = > new HashMap<>(); > 766: // Persistent DFS iterator so that sequential access is O(N) instead > of O(N^2). please expand "DFS" modules/javafx.controls/src/main/java/javafx/scene/control/TreeTableView.java line 807: > 805: // this forces layoutChildren at the next pulse, and therefore > 806: // updates the item count if necessary > 807: expandedItemCountDirty = true; "if necessary" no longer applies. modules/javafx.controls/src/main/java/javafx/scene/control/TreeUtil.java line 171: > 169: } > 170: > 171: @Override public boolean hasNext() { please put `@Override` annotations on its own line in the new code modules/javafx.controls/src/main/java/javafx/scene/control/TreeView.java line 410: > 408: // See JDK-8125681 for the justification and performance gains. > 409: private Map<Integer, SoftReference<TreeItem<T>>> treeItemCacheMap = > new HashMap<>(); > 410: // Persistent DFS iterator so that sequential access is O(N) instead > of O(N^2). please expand "DFS" ------------- PR Review: https://git.openjdk.org/jfx/pull/2142#pullrequestreview-4083392045 PR Review Comment: https://git.openjdk.org/jfx/pull/2142#discussion_r3058819814 PR Review Comment: https://git.openjdk.org/jfx/pull/2142#discussion_r3058822507 PR Review Comment: https://git.openjdk.org/jfx/pull/2142#discussion_r3058859239 PR Review Comment: https://git.openjdk.org/jfx/pull/2142#discussion_r3058867862
