[jira] Commented: (HARMONY-3883) [classlib]WeakHashMap.keySet().toArray() intermittently throws NoSuchElementException in java.lang.ThreadTest of DRLVM kernel test

["Li-Gang Wang (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HARMONY-3883?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12496483 ] 

Li-Gang Wang commented on HARMONY-3883:
---------------------------------------

Running the modified WeakHashmapTest, the problem can appear whether using GCv5 or GCv4.1.

> [classlib]WeakHashMap.keySet().toArray() intermittently throws NoSuchElementException in java.lang.ThreadTest of DRLVM kernel test
> ----------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: HARMONY-3883
>                 URL: https://issues.apache.org/jira/browse/HARMONY-3883
>             Project: Harmony
>          Issue Type: Bug
>          Components: Classlib
>         Environment: win/linux 32/64
>            Reporter: Li-Gang Wang
>         Attachments: WeakHashmapTest.java
>
>
> I found there is a design bug in WeakHashMap.keySet().toArray() .
> WeakHashMap.keySet().toArray() inherits AbstractCollection.toArray(). Its implementation is as follows:
> 	public Object[] toArray() {
> 		int size = size(), index = 0;
> 		Iterator<?> it = iterator();
> 		Object[] array = new Object[size];
> 		while (index < size) {
>             array[index++] = it.next();
>         }
> 		return array;
> 	}
> After assigning 'size()' to 'size', the 'while' loop expects 'size' elements in the iterator 'it'. But actually GC might happen and clear some weak keys of the WeakHashMap instance in the meantime. The iterator will skip the cleared weak keys in 'it.hasNext()' called by 'it.next()', so the actual number of the elements in 'it' is smaller than 'size'. When 'it' runs out its elements, 'it.next()' will throw a NoSuchElementException. Parts of HashIterator implementation in WeakHashMap is as follows:
>         public boolean hasNext() {
>             if (nextEntry != null) {
>                 return true;
>             }
>             while (true) {
>                 if (nextEntry == null) {
>                     while (position < elementData.length) {
>                         if ((nextEntry = elementData[position++]) != null) {
>                             break;
>                         }
>                     }
>                     if (nextEntry == null) {
>                         return false;
>                     }
>                 }
>                 // ensure key of next entry is not gc'ed
>                 nextKey = nextEntry.get();
>                 if (nextKey != null || nextEntry.isNull) {
>                     return true;
>                 }
>                 nextEntry = nextEntry.next;
>             }
>         }
>         public R next() {
>             if (expectedModCount == modCount) {
>                 if (hasNext()) {
>                     currentEntry = nextEntry;
>                     nextEntry = currentEntry.next;
>                     R result = type.get(currentEntry);
>                     // free the key
>                     nextKey = null;
>                     return result;
>                 }
>                 throw new NoSuchElementException();
>             }
>             throw new ConcurrentModificationException();
>         }
> I suspect the intermittent failure in java.lang.ThreadTest may disappear if this bug is fixed.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.