Good suggestion, thank you.
Mike
On May 7 2013, at 06:36 , Jason Mehrens wrote:
> Mike,
>
> EmptyNavigableMap and EmptyNavigableSet should contain the static final
> references themselves and not the Collections class. That way they load on
> demand and not with the Collections class. Use default visibility inside the
> inner class so the compiler doesn't generate method to promote access. See
> the EmptyIterator class.
>
> Jason
>
>
> ----------------------------------------
>> From: mike.dui...@oracle.com
>> Subject: RFR : 7129185 : Add
>> Collections.{checked|empty|unmodifiable}Navigable{Map|Set}
>> Date: Mon, 29 Apr 2013 18:54:19 -0700
>> To: core-libs-dev@openjdk.java.net
>>
>> Hello all;
>>
>> This is a non-integration code review. I am picking this patch up after
>> ignoring it for most of the last year. I've recently expanded the regression
>> tests to, I believe, handle almost all of the new code paths added by this
>> patch.
>>
>> http://cr.openjdk.java.net/~mduigou/JDK-7129185/0/webrev/
>>
>> This issue is a follow-on to JDK-4533691 which added emptySortedSet(). In
>> addition to adding support for NavigableSet/Map this patch also corrects
>> differences between the behaviour of:
>>
>> Set<Integer> uts = Collections.unmodifiableNavigableSet(new TreeSet())
>>
>> and
>>
>> Set<Integer> es = Collections.emptyNavigableSet()
>>
>> involving bounded sub-ranges. At this point I believe that "uts" and "es"
>> will be operationally indistinguishable. emptyNavigableSet() will still be
>> more efficient though as it is a singleton and doesn't generally(*) consume
>> additional resources for each instance. The asterisk next to generally comes
>> from the bounded sub-ranges functionality. Sub ranges of empty SortedSet and
>> NavigableSet will no longer be the singleton. They are instead instances
>> which capture the range.
>>
>> Because so much time has passed since this issue originally surfaced I'm
>> concerned that I might be forgetting something. I do know that I still need
>> to create an EmptyNavigableMap unit test and add serialversionid to all the
>> new classes but does anything else seem to be missing either in terms of the
>> implementation or the tests?
>>
>> Mike
