R devel - Jun 2009 - Inheritance before ANY...

Hi all,
I am programming using S4. I define two classes, "B" is inheriting
from
"A". Apparently (at least since version 2.9.0 ?), when the correct 
signature is not find,
R prefers to chose a signature in the ancestor BEFORE a signature in the 
class. This is very strange to me...

If I define plot for ("A",missing) and ("B",ANY), calling 
plot("B",missing), I would expect R to chose plot("B",ANY)
before
plot("A",missing)

Sincerely
Christophe


------------------ (translate from french) --------
Note : method with signature "A#missing" chose for function
"plot",
signature cible "B#missing".
"B#ANY" could also be compatible

>>>>> "CG" == Christophe Genolini <cgenolin at
u-paris10.fr>
>>>>>     on Tue, 09 Jun 2009 22:11:51 +0200 writes:
    CG> Hi all,
    CG> I am programming using S4. I define two classes, "B" is
inheriting from
    CG> "A". Apparently (at least since version 2.9.0 ?), when the
correct
    CG> signature is not find,
    CG> R prefers to chose a signature in the ancestor BEFORE a signature in
the
    CG> class. This is very strange to me...

    CG> If I define plot for ("A",missing) and ("B",ANY),
calling
    CG> plot("B",missing), I would expect R to chose
plot("B",ANY) before
    CG> plot("A",missing)

    CG> ------------------ (translate from french) --------
    CG> Note : method with signature "A#missing" chose for function
"plot",
    CG> signature cible "B#missing".
    CG> "B#ANY" could also be compatible

Yes, method dispatch had indirectly been greatly improved
in 2.9.0, as the elimination of duplicate superclasses and a
consistent (superclass) ordering was introduced.

Look at
   ?Methods
and skip to its references (simply "sr" in ESS),
finding

     Chambers, John M.(2009) _Developments in Class Inheritance and
     Method Selection_ <URL:
     http://stat.stanford.edu/~jmc4/classInheritance.pdf>.

which is very recommended, and, BTW, as I just notice, has been
updated in the very last few days,
judging from my earlier reading; the updates include new
features of class definition / method dispatch in the situation
where S3 and S4 are intermingled.

Regards,
Martin Maechler, ETH Zurich