namespace S3 and S4 generic imports cannot both be satisfied:

PkgA wishes to write a method for 'unique' on S4 class 'A'. ?Methods
indicates that one should

   setGeneric("unique")

   setClass("A")
   unique.A <- function(x, incomparables=FALSE, ...) {}
   setMethod(unique, "A", unique.A)

Both S3 and S4 methods need to be exported in the NAMESPACE

   import(methods)
   S3method(unique, A)
   exportMethods(unique)

PkgB introduces a new class and method

   setClass("B")
   unique.B <- function(x, incomparables=FALSE, ...) {}
   setMethod(unique, "B", unique.B)

and in the NAMESPACE has

   import(methods)
   importFrom(PkgA, unique)
   S3method(unique, B)
   exportMethods(unique)

Unfortuantely, R CMD check says that

* checking whether package 'PkgB' can be installed ... WARNING
Found the following significant warnings:
   Warning: found an S4 version of 'unique' so it has not been imported
correctly
See '/home/mtmorgan/tmp/PkgB.Rcheck/00install.out' for details.

This is from (svn r61253) R-devel/src/library/base/R/namespace.R:1339,
where the code finds the S4 generic, but not the S3 generic. Obviously
the namespace cannot have both the S3 and S4 symbols defined, but this
seems to be required? A workaround might extend the check to include
getGeneric(genname)@default.

This scenario is reproducible in the attached tarball

   tar xzf PkgAB.tar.gz
   R CMD INSTALL PkgA
   R CMD check PkgB

Martin Morgan

Yes, you are right.

Mixing S3 and S4 methods for a generic is fine, although in subtle cases one is
safer promoting the S3 method to an S4 method, as you did in your example.

Usually, the default method for the S4 generic is the S3 generic.  But, in
general, it's not possible to check algorithmically whether the S3 methods will
be dispatched.  For example, an S4 method on "vector" could dispatch S3 methods
on subclasses "numeric", etc. (don't ask why ...), for a generic that had no
default method.

That the trouble with this check isn't found right away is likely because one is
typically working with a primitive where no generic function is created.

Should be fixed in rev. 61263.  Please check on your real example; it seems fine
on the test you submitted.

Thanks for the catch.

John

On 12/8/12 3:05 PM, Martin Morgan wrote:

PkgA wishes to write a method for 'unique' on S4 class 'A'. ?Methods
indicates that one should

   setGeneric("unique")

   setClass("A")
   unique.A <- function(x, incomparables=FALSE, ...) {}
   setMethod(unique, "A", unique.A)

Both S3 and S4 methods need to be exported in the NAMESPACE

   import(methods)
   S3method(unique, A)
   exportMethods(unique)

PkgB introduces a new class and method

   setClass("B")
   unique.B <- function(x, incomparables=FALSE, ...) {}
   setMethod(unique, "B", unique.B)

and in the NAMESPACE has

   import(methods)
   importFrom(PkgA, unique)
   S3method(unique, B)
   exportMethods(unique)

Unfortuantely, R CMD check says that

* checking whether package 'PkgB' can be installed ... WARNING
Found the following significant warnings:
   Warning: found an S4 version of 'unique' so it has not been imported
correctly
See '/home/mtmorgan/tmp/PkgB.Rcheck/00install.out' for details.

This is from (svn r61253) R-devel/src/library/base/R/namespace.R:1339,
where the code finds the S4 generic, but not the S3 generic. Obviously
the namespace cannot have both the S3 and S4 symbols defined, but this
seems to be required? A workaround might extend the check to include
getGeneric(genname)@default.

This scenario is reproducible in the attached tarball

   tar xzf PkgAB.tar.gz
   R CMD INSTALL PkgA
   R CMD check PkgB

Martin Morgan