RFR : 8016446 : (m) Add override forEach/replaceAll to HashMap, Hashtable, IdentityHashMap, WeakHashMap, TreeMap
peter.levart at gmail.com
Thu Jun 13 06:42:43 UTC 2013
On 06/13/2013 01:14 AM, Mike Duigou wrote:
> On Jun 12 2013, at 15:49 , Remi Forax wrote:
>> On 06/12/2013 11:23 PM, Mike Duigou wrote:
>>> Hello all;
>>> This patch adds optimized implementations of Map.forEach and Map.replaceAll to HashMap, Hashtable, IdentityHashMap, WeakHashMap, TreeMap
>>> The implementations traverse the map more efficiently than the default iterator implementation and sometimes avoid creation of transient Map.Entry instances. The fast-fail behaviour of the default implementation is retained.
>> Hi Mike,
>> funnily I was writing HashMap.forEach/LinkedHashMap.forEach at the same time.
>> (you need also to override forEach in LinkedHashMap because the one you inherits from HashMap doesn't use the linked list of entries).
> I don't think we need to offer a guarantee of ordering for the forEach but using the linked list is probably more efficient.
The javadoc for Map.forEach says differently:
default void forEach(BiConsumer <http://download.java.net/jdk8/docs/api/java/util/function/BiConsumer.html><? superK <http://download.java.net/jdk8/docs/api/java/util/Map.html>,? superV <http://download.java.net/jdk8/docs/api/java/util/Map.html>> action)
Performs the given action on each entry in this map, in the order
entries are returned by an entry set iterator (which may be unspecified)
>> My code is slightly different from yours because I've moved the cases where the item is a red/black tree out of the fast path
>> (the tree is created either if you are unlucky, if hashCode is badly written or if someone forge keys to have collisions)
> Yes, this is probably worthwhile
>> and does the modCount check for each element because a call to the consumer can change the underlying map
>> so you can not do a modCount check only at the end.
> Yes, mine wasn't fast fail enough. I am always torn between satisfying those who will moan at the per-element cost and those who will moan about the lack of fast-fail behaviour.
> Thank you for the improvements!
More information about the core-libs-dev