You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
f0e6841ca8
The EntityOutlinerListModel was violating the QAbstractItemModel contract in a few cases, as reported by `QAbstractItemModelTester`. The important ones causing issues were: - Entry order was not guaranteed, leading to model indices pointing at invalid data - Parent/child relationships could be temporarily invalid due to a change I made in EditorEntityModel::RemoveEntity to try to avoid an unnecessary reparent operation - as it turned out, the parent/child data was being cached even for recreated entities and not clearing child data could cause issues - `EntityOutlinerListModel::ProcessEntityUpdates` was emitting data changed between two indices that didn't necessarily share a parent, which is [undefined behavior](https://doc.qt.io/qt-5/qabstractitemmodel.html#dataChanged) The other reported issues (that weren't really causing issues with `QTreeView`) were: - The root index had flags other than `Qt::ItemIsDropEnabled` - `rowCount` showed all columns as having children - `parent` showed indices as being parented to a non-0 column This change introduces fixes for the above issues, namely: - Reverts my change to `EditorEntityModel::RemoveEntity` to ensure we don't have invalid parent/child references sitting in the cache - Ensures `EditorEntityModelEntry` child ordering is guaranteed sorted by EntityId, to prevent the `EntityOutlinerListModel` from having indices pointed at invalid data*. - Fixes various model sanity issues, such as `rowCount` being 0 for indices with a non-0 column Two unit tests were added to reproduce the invalid behavior and validate the fix: TestCreateFlatHierarchyUndoAndRedoWorks and TestCreateNestedHierarchyUndoAndRedoWorks This change focuses on correctness over performance. My subjective in-Editor outliner experience is about the same, but it may be worthwhile to expand the test coverage with a benchmarking suite to look into areas for optimization. *As a rough illustration of the previous child ordering behavior, consider the following entity hierarchy: ``` Root (EID 9999) |_ Child1 (EID 2) |_ Child2 (EID 3) |_ Child3 (EID 4) ``` With an representations like the following pseudocode: ``` // EditorEntityModel representation EditorEntityModelEntry root; root.children[0] = 2; root.children[1] = 3; root.children[2] = 4; // EditorOutlinerListModel representation // row, column, user data (64 bit uint) child1 = QModelIndex(0, 0, 2) child2 = QModelIndex(1, 0, 3) child3 = QModelIndex(2, 0, 4) ``` When removing a child, the `EditorEntityModel` used to do roughly the following: ``` // Swap and pop the last child int indexToRemove = 0; swap(root.children[indexToRemove], root.children[root.children.size() - 1]); root.children.resize(root.children.size() - 1); model.notifyRemoved(root, indexToRemove); // model removes the row indicated // Leading to this EditorEntityModel state root.children[0] = 4; root.children[1] = 3; // And this EntityOutlinerListModel state, note that the row indices are swapped from the indices in the backing storage child2 = QModelIndex(0, 0, 3) child3 = QModelIndex(1, 0, 4) ``` A QModelIndex having a row that doesn't match its underlying data is undefined behavior, and was the source of an intermittent crash in our `QSortFilterProxyModel` as subsequent updates to the wrong row led to an invalid proxy state. Signed-off-by: nvsickle <nvsickle@amazon.com> |
4 years ago | |
|---|---|---|
| .. | ||
| EntityIdQLineEditTests.cpp | 5 years ago | |
| EntityOutlinerTests.cpp | 4 years ago | |
| EntityPropertyEditorTests.cpp | 5 years ago | |