RFR : 8016446 : (m) Add override forEach/replaceAll to HashMap, Hashtable, IdentityHashMap, WeakHashMap, TreeMap

[Peter Levart]

On 06/19/2013 08:44 AM, Peter Levart wrote:
> On 06/19/2013 01:13 AM, Mike Duigou wrote:
>> On Jun 18 2013, at 05:19 , Doug Lea wrote:
>>
>>> On 06/17/13 19:30, Mike Duigou wrote:
>>>
>>>> I had to add the improved default for ConcurrentMap which was 
>>>> present in the lambda repo in order to have correct behaviour. 
>>>> Since getOrDefault is already in ConcurrentMap I will include this 
>>>> but we have to be careful when we do a jsr 166 syncup to make sure 
>>>> that the defaults don't get lost.
>>>>
>>> Now synched up on my side.
>>>
>>> -Doug
>>>
>>
>>
>> Per a suggestion from Remi I updated the ConcurrentMap.replaceAll 
>> default to use forEach. This trades off the entrySet iterator 
>> overhead for creation of a capturing BiConsumer lambda.
>>
>> http://hg.openjdk.java.net/jdk8/tl/jdk/raw-diff/1f7cbe4829fe/src/share/classes/java/util/concurrent/ConcurrentMap.java 
>>
>>
>> Mike
>
> Hi Mike, Remi,
>
> Since forEach implementation can be taken from default Map.forEach in 
> some implementations of ConcurrentMap, and that implementation is 
> based on entrySet Iterator, isn't it dangerous for this to trigger 
> ConcurrentModificationException in some implementation of 
> ConcurrentMap? I see nothing in the spec. of ConcurrentMap that 
> suggests it's entrySet iterators are never fail-fast. They can be 
> prepared for modifications from other threads (synchronization), but 
> may not tolerate re-entrant calls.
>
> For example some implementation of (Concurrent)Map could be 
> structurally modified as a result of Map.replace(key, old, new) - 
> imagine a ConcurrentWeakHashMap that expunges stale entries on each 
> call - and forEach iteration may not be prepared to handle such 
> situations.
>
> Regards, Peter
>

Hi Mike,

I'd also like to suggest something. You have made lots of tests that 
cover the functionality of new default methods in Map and other 
collections interfaces which prove the correctness of behaviour when 
used with implementations of those interfaces in JDK. Perhaps It would 
be wise to extend those tests to include some JDK-external 
implementations that are known to pass basic tests, taken for example 
from Guava or Apache Commons or even JDK7. The interactions of default 
methods with existing methods in various implementations of collection 
interfaces could reveal "bugs" that may be on the part of default 
methods. I'm sure the maintainers of these external implementations will 
do the same, but may be doing that only after JDK8 release.

Regards, Peter