Thank you Doug, this addresses my concerns. On 05/28/2013 11:29 AM, Doug Lea wrote:
On 05/27/13 10:30, Chris Hegarty wrote:1) CHM no longer extends AbstractMap. I guess this should not be a problem in the real world, and I guess users would not be too surprised by instanceof checks. Just worth highlighting the change for compatibility.Yes, thanks. It was a bad idea to extend AbstractMap in the first place, since all of its methods are or should be overridden. But if there is any reason to retain this, it is harmless to do so,
OK, this can be the backup plan if this issue arises during CCC.
at the expense of adding two unused fields to footprint. (The AbstractMap keySet and Values fields cannot be used by CHM because they are package-protected, so it needs to declare its own versions anyway.)4) Does is make sense for KeySetView to be Serializable? It looks a little odd with value as its only field.(Plus the backing ConcurrentHashMap.)
Yes, of course.
Serialization is needed (or at least desired) because this class is now used for CHM.newKeySet, which finally adds a way to get the equivalent of a ConcurrentHashSet. (You could get one via Collections.newSetFromMap, but then lose the advertised concurrency/atomicity properties.)
Thanks, -Chris.
-Doug
