To avoid this problem, you can write abstract classes that implement
the interfaces. If third party implementations then extend the
abstract class there's no backward compatibility issue.
Of course, if you add functionality to the interface that needs to
have a specific implementation (not just the no-op in the abstract
class) then yes, you have broken backward compatibility. But usually
this kind of feature addition happens at major releases and it's
presumed to be ok.
Craig
On Mar 27, 2010, at 8:42 AM, Mark Phippard wrote:
I'd suggest taking a look at guidelines that other Java projects have
established to get an idea of consequences.
Here is the one for Eclipse:
http://wiki.eclipse.org/Evolving_Java-based_APIs
There are probably some Apache projects that publish something
similar.
Mark
On Fri, Mar 26, 2010 at 7:27 PM, Hyrum K. Wright
<hyrum_wri...@mail.utexas.edu> wrote:
Mark,
A couple of months ago, you mentioned the avenue of interface-izing
the various objects that are part of the public API, such as the
callbacks and the client interface. In thinking a bit more about
this, and given my general lack of Java background (in spite of the
recent commits to the contrary) I've got a question about backward
compat in this scenario.
In creating an interface (which we have) for our client API, we
could also create interfaces for the objects that interface
references. The primary reason for doing so is enabling third-
party implentors to do so without having to actually use any of our
objects, giving them (and us, ultimately) more flexibility for
compatibility.
This does add an additional question: if third-parties are
implementing these interfaces, and we chose to add more
functionality them (and hence more APIs) to the interfaces, does
this break backward compat? Code that formerly compiled against an
interface which now includes another function definition would no
longer compile, yes? If so, does that consistute a broken backward
compat? What if somebody just dropped in the new jar, and didn't
recompile?
If adding interfaces to the JavaHL API means that we'll need to rev
them (not just extend them) in the event we add more data to those
interfaces/objects, I don't know that it's worth the work.
Thoughts?
-Hyrum
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
Craig L Russell
Architect, Oracle
http://db.apache.org/jdo
408 276-5638 mailto:craig.russ...@oracle.com
P.S. A good JDO? O, Gasp!
