Skip to content

Pb with .findInheritedMethods

7 messages · Hervé Pagès, John Chambers, Seth Falcon

#
Hi again,


Here is a very simplified version of a class hierarchy
defined in the Biobase package (Bioconductor). I post
here because this seems to be an S4 related problem:

    setClass("A", representation(name="character"))
    setMethod("initialize", "A", function(.Object) {.Object at name <- "I'm
    an A"; .Object})

    setClass("Ab", contains="A")
    setMethod("initialize", "Ab", function(.Object) callNextMethod(.Object))

    setClass("Abc", contains="Ab")

    setClass("Abcd", contains = c("Abc"))

Now if I do:

    tmp1 <- new("Abc")
    tmp2 <- new("Abcd")

I get the following warning:

    Warning message:
    Ambiguous method selection for "initialize", target "Abcd" (the
    first of the signatures shown will be used)
        Abc
        Ab
     in: .findInheritedMethods(classes, fdef, mtable)

I don't really understand why .findInheritedMethods is
complaining here...
And if I don't do 'tmp1 <- new("Abc")' before I
do 'tmp2 <- new("Abcd")', then I don't get the warning
anymore!

Does anybody have an explanation for this?


Thanks,
H.
#
Hi again,

This happens with R-2.4.0 and R-devel.

Cheers,
H.
Herve Pages wrote:
#
Hi John,
John Chambers wrote:
Thanks for looking at this!

Yes it is a simplified version of a real case and
here .Object is all what callNextMethod() needs because
the initialize method for an "A" object takes no argument
other than .Object

More generally I don't see what's wrong with not passing
to callNextMethod all the arguments coming from the call
to new:

    setClass("A", representation(toto="integer"))
    setMethod("initialize", "A", function(.Object, toto0) {.Object at toto
    <- as.integer(toto0); .Object})
    new("A", 45.1)

    setClass("Ab", contains="A")
    setMethod("initialize", "Ab", function(.Object, x, y)
    callNextMethod(.Object, x*y+1))
    new("Ab", 5, 2)


Regards,

H.
1 day later
#
Herve Pages wrote:
As I mentioned, this relates to writing methods for initialize().   
Imagine someone else extends the class "Ab", for which you wrote a 
method.  If they add slots to their class and you do not pass down ... 
to callNextMethod(), then you have blocked users from setting values for 
those slots in calls to new(), since the ... argument is thrown away by 
your method.

So in your example, it's more socially responsible to add "..." as a 
formal argument to your method, and then to pass it on to callNextMethod():

   setMethod("initialize", "Ab", function(.Object, x, y, ...)
    callNextMethod(.Object, x*y+1, ...))

The other aspect to this is that the last specialized method in your chain of class definitions should end up with:
   callNextMethod(.Object, ...)
Then the default initialize() method will set values for named slots.  Again, the point is to allow others to extend your class definitions.
#
John Chambers <jmc at r-project.org> writes:
If you have written an initialize method, then it is likely because
you want to do something other than just fill slots.  A subclass will
most likely need to define its own initialize method and in this case,
I'm not sure passing ... will matter.
Unless that isn't the behavior one desires (and I would claim this is
a rather common situation).  As part of the user interface to the
class, the developer may want to decouple the intitialization
interface from specific slot names.


+ seth