[infinispan-dev] Fwd: [jgroups-dev] Incompatible change in 3.1.0

[Bela Ban]

On 11/29/11 11:43 AM, Tristan Tarrant wrote:
> On 11/29/2011 11:28 AM, Bela Ban wrote:
>> Yes, I thought about this too, but then trashed the idea because enums
>> take up too much memory. Since an enum is a class, and an enum instance
>> is an instance (although a singleton / classloader instance), we have
> Why, you would just use the Enum when setting and getting, but otherwise
> just store your internal representation.


Hmm, not a bad idea !  Although this way, you couldn't do 
Message.setFlags((short)Message.OOB | Message.NO_FC), bitwise 
operations, if OOB and NO_FC are enum instances...

Hmm, Message.setFlags() would look like this:

public void setFlags(Flag flag) {
      flags |= (1 << flag.ordinal());
}


I do like the idea, although no using enums all the way and converting 
them to/from shorts internally kind of defeats the purpose of enums...

Actually, I don't think this breaks any API, unless people use the byte 
or short values directly.

Hmm, what do you do whith Message.getFlags() ? It returns a short. 
Although this can be represented with an enum, multiple values can't, 
e.g. 'OOB | NO_FC' cannot be represented as an enum... Maybe getFlags() 
should be deprecated then ?

WDYT ? I'm willing to do this in 3.1, it's probably not worse than the 
change from byte to short...

-- 
Bela Ban
Lead JGroups (http://www.jgroups.org)
JBoss / Red Hat