Re: classpath ./AUTHORS ./ChangeLog java/util/HashM...

[Bryce McKinlay]

Tom Tromey wrote:

> > > > > > "Eric" == Eric Blake <address@hidden> writes:
>
> Eric> Modified files:
> Eric> 	.              : AUTHORS ChangeLog 
> Eric> 	java/util      : HashMap.java Hashtable.java 
>
> Eric>        * java/util/Hashtable.java (contains): check for null
> Eric>        (Hashtable(Map)): more efficient
> Eric>        (clear): more efficient
> Eric>        (clone): more efficient, by adding Entry.copy
>
> I think it is disputable whether these are really more efficient.
>
> I'm not too concerned with `clear'.  In that case the new code may
> very well be more efficient.  It is hard to say in a
> platform-independent way.
I don't agree with the change to 'clear'.

@@ -389,10 +400,7 @@
  public synchronized void clear()
  {
    modCount++;
-    for (int i=0; i < buckets.length; i++)
-      {
-        buckets[i] = null;
-      }
+    buckets = new Entry[buckets.length];
    size = 0;
  }

Creating a whole new array is an expensive operation. In addition to the 
cost of the allocation itself (typically this requires synchronization 
on a global heap lock), it creates more garbage for the collector to 
deal with (increasing GC frequency), and potentially uses more memory by 
causing the heap to expand. In contrast, a small, tight loop like the 
one above should execute very fast on any modern cpu (assuming a JIT and 
taking cache behaviour into account), so unless buckets.length is very 
large, its hard to imagine that it could be less efficient.

Any contradictory opinions?

regards

Bryce.