TableCell自定义Skin运行14小时后触发NPE问题排查求助

我在应用中实现了TableView，为TableCell创建了自定义Skin，并在TableCell的createDefaultSkin()方法中初始化该自定义Skin，前期运行正常。但应用连续运行14小时后崩溃，抛出如下NullPointerException：

java.lang.NullPointerException: Cannot invoke "javafx.scene.control.Skin.dispose()" because the return value of "javafx.scene.control.IndexedCell.getSkin()" is null
    at javafx.scene.control.skin.TableRowSkinBase.recreateCells(TableRowSkinBase.java:713)
    at javafx.scene.control.skin.TableRowSkinBase.updateCells(TableRowSkinBase.java:496)
    at javafx.scene.control.skin.TableRowSkinBase.checkState(TableRowSkinBase.java:640)
    at javafx.scene.control.skin.TableRowSkinBase.layoutChildren(TableRowSkinBase.java:247)
    at javafx.scene.control.Control.layoutChildren(Control.java:601)
    at javafx.scene.control.Cell.layoutChildren(Cell.java:636)
    at javafx.scene.Parent.layout(Parent.java:1207)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Parent.layout(Parent.java:1214)
    at javafx.scene.Scene.doLayoutPass(Scene.java:579)
    at javafx.scene.Scene$ScenePulseListener.pulse(Scene.java:2499)
    at com.sun.javafx.tk.Toolkit.lambda$runPulse$2(Toolkit.java:405)
    at java.base/java.security.AccessController.doPrivileged(AccessController.java:399)
    at com.sun.javafx.tk.Toolkit.runPulse(Toolkit.java:404)
    at com.sun.javafx.tk.Toolkit.firePulse(Toolkit.java:434)
    at com.sun.javafx.tk.quantum.QuantumToolkit.pulse(QuantumToolkit.java:575)
    at com.sun.javafx.tk.quantum.QuantumToolkit.pulse(QuantumToolkit.java:555)
    at com.sun.javafx.tk.quantum.QuantumToolkit.pulseFromQueue(QuantumToolkit.java:548)
    at com.sun.javafx.tk.quantum.QuantumToolkit.lambda$runToolkit$11(QuantumToolkit.java:352)
    at com.sun.glass.ui.InvokeLaterDispatcher$Future.run(InvokeLaterDispatcher.java:96)
    at com.sun.glass.ui.gtk.GtkApplication._runLoop(Native Method)
    at com.sun.glass.ui.gtk.GtkApplication.lambda$runLoop$11(GtkApplication.java:316)
    at java.base/java.lang.Thread.run(Thread.java:833)

TableRowSkinBase中recreateCells()方法的相关代码如下：

/*
     * A map that maps from TableColumn to TableCell (i.e. model to view).
     * This is recreated whenever the leaf columns change, however to increase
     * efficiency we create cells for all columns, even if they aren't visible,
     * and we only create new cells if we don't already have it cached in this
     * map.
     *
     * Note that this means that it is possible for this map to therefore be
     * a memory leak if an application uses TableView and is creating and removing
     * a large number of tableColumns. This is mitigated in the recreateCells()
     * function below - refer to that to learn more.
     */
    WeakHashMap<TableColumnBase, Reference<R>> cellsMap;


    private void recreateCells() {
            if (cellsMap != null) {
                Collection<Reference<R>> cells = cellsMap.values();
                Iterator<Reference<R>> cellsIter = cells.iterator();
                while (cellsIter.hasNext()) {
                    Reference<R> cellRef = cellsIter.next();
                    R cell = cellRef.get();
                    if (cell != null) {
                        cell.updateIndex(-1);
                        cell.getSkin().dispose(); // THIS IS WHERE THE ISSUE OCCURS
                        cell.setSkin(null);
                    }
                }
                cellsMap.clear();
            }

            ObservableList<? extends TableColumnBase/*<T,?>*/> columns = getVisibleLeafColumns();

            cellsMap = new WeakHashMap<>(columns.size());
            fullRefreshCounter = DEFAULT_FULL_REFRESH_COUNTER;
            getChildren().clear();

            for (TableColumnBase col : columns) {
                if (cellsMap.containsKey(col)) {
                    continue;
                }

                // create a TableCell for this column and store it in the cellsMap
                // for future use
                createCellAndCache(col);
            }
        }

自定义Skin逻辑简单，仅包含自定义行为实现：

import javafx.beans.property.ReadOnlyObjectProperty;
import javafx.scene.control.IndexedCell;
import javafx.scene.control.TableCell;
import javafx.scene.control.TableColumnBase;
import javafx.scene.control.skin.TableCellSkinBase;

/**
 * Custom table cell skin.
 *
 * @param <S> The type of the TableView generic type (i.e. S == TableView&lt;S&gt;)
 * @param <T> The type of the content in all cells in this TableColumn
 * @param <C> the cell type
 */
public final class CustomTableCellSkin<S, T, C extends IndexedCell<T>> extends TableCellSkinBase<S, T, C> {

    /** Cell behaviour. */
    private final CustomTableCellBehavior<S, T> behaviour;

    /**
     * Constructor.
     *
     * @param cell skinnable cell
     */
    @SuppressWarnings("unchecked")
    public CustomTableCellSkin(final TableCell<S, T> cell) {
        super((C) cell);
        behaviour = new CustomTableCellBehavior<>(cell);
    }

    /**
     * Return the cell behaviour.
     *
     * @return behaviour
     */
    public final CustomTableCellBehavior<S, T> getBehaviour() {
        return behaviour;
    }

    @Override
    public final ReadOnlyObjectProperty<? extends TableColumnBase<S, T>> tableColumnProperty() {
        @SuppressWarnings("unchecked")
        final TableCell<S, T> cell = (TableCell<S, T>) getSkinnable();
        return cell.tableColumnProperty();
    }
}

排查线索 #

1. 单元格Skin会在哪些情况下被清空？ #

• 手动调用cell.setSkin(null)：业务代码或第三方逻辑中若存在该调用，会直接清空Skin
• Skin被垃圾回收：若Skin仅被弱引用持有，且无其他强引用关联，GC时可能被回收，导致getSkin()返回null
• JavaFX内部Skin替换：控件皮肤被重新创建（如主题切换、控件重新初始化）时，旧Skin会被清理，极端场景下可能出现状态不一致
• 单元格复用异常：TableView的单元格复用机制在极端场景（大量数据滚动、列频繁调整）下，可能出现Skin状态异常，导致Skin被意外清空

2. 为何recreateCells()方法中未在调用dispose()前做空检查？ #

这是JavaFX控件库的潜在bug：

• 开发人员默认认为cell != null时Skin必然存在，但长时间运行的复杂场景中，可能出现Skin被提前释放的情况
• cellsMap使用WeakHashMap弱引用存储单元格，单元格对象被GC前，Skin可能已被意外清空
• 该场景属于边缘case，常规测试难以复现，因此官方代码未添加空判断

临时修复建议 #

• 在自定义TableCell中重写setSkin()方法，确保Skin被正确持有
• 自定义TableRowSkinBase子类，在调用dispose()前添加空检查（需注意JavaFX内部API的访问限制）
• 避免在业务代码中直接操作单元格的Skin对象

内容的提问来源于stack exchange，提问作者Sai Dandem