AbstractDomNodeList access might throw a NPE due to a race condition …

…in the DomHtmlAttributeChangeListenerImpl clearing of cachedElements processing (#882)

src/changes/changes.xml

@@ -12,6 +12,10 @@
  Special handling of the bgsound tag in FF removed. BGSound is now handled as unknown tag in all browsers.
  Looks like this was a leftover from the IE days.
  </action>
+ <action type="fix" dev="rbri" issue="#882">
+ AbstractDomNodeList access might throw a NPE due to a race condition in the DomHtmlAttributeChangeListenerImpl
+ clearing of cachedElements processing.
+ </action>
  <action type="fix" dev="rbri">
  Function initHashChangeEvent() is no longer available in FF_ESR.
  </action>

src/main/java/org/htmlunit/html/AbstractDomNodeList.java

@@ -47,13 +47,13 @@ public abstract class AbstractDomNodeList<E extends DomNode> extends AbstractLis
  */
  public AbstractDomNodeList(final DomNode node) {
  super();
+ node_ = node;
+
  if (node == null) {
- node_ = null;
  cachedElements_ = Collections.EMPTY_LIST;
  return;
  }
 
- node_ = node;
  final DomHtmlAttributeChangeListenerImpl listener = new DomHtmlAttributeChangeListenerImpl(this);
  node_.addDomChangeListener(listener);
  if (node_ instanceof HtmlElement) {
@@ -81,10 +81,20 @@ protected DomNode getDomNode() {
  * @return the nodes in this node list
  */
  private List<E> getNodes() {
- if (cachedElements_ == null) {
- cachedElements_ = provideElements();
+ if (cachedElements_ != null) {
+ return cachedElements_;
  }
- return cachedElements_;
+
+ // a bit of a hack but i like to avoid synchronization
+ // see https:/HtmlUnit/htmlunit/issues/882
+ //
+ // there is a small chance that the cachedElements_ are
+ // set to null after the assignment and before the return
+ // but this is a race condition at all and depending on the
+ // thread state the same overall result might happen also with sync
+ final List<E> providedElements = provideElements();
+ cachedElements_ = providedElements;
+ return providedElements;
  }
 
  /**