R devel - Sep 2004 - as(object,"list") as(object,"matrix") differences?

Hello!

as(object,"list") and as(object,"matrix") behave quite
differently if it comes to their attributes.
I define two classes. One of them "contains" a "list" the
other a "matrix"
setClass("myclass"
,representation(info="character")
,contains="matrix"
)
setClass("mylist"
,representation(info="character")
,contains="list"
)

#init
dd<-matrix(1:6,nrow=2)
rownames(dd)<-c("a","b")
tt<-new("myclass",dd)

tmp<-vector("list",4)
names(tmp)<-1:4
ll <- new("mylist",tmp,info="foo")

rownames(tt)
[1] "a" "b"
> rownames(as(tt,"matrix"))
[1] "a" "b"
> names(ll)
[1] "1" "2" "3"
"4"
> names(as(ll,"list")) 
NULL
#but
names(ll@.Data)
The difference in behaviour to which i would like to point your attention is
that as(object,"matrix") preserves the "dimnames" but at the
same time as(object,"list") drops the "names" attribute.

Is it recomended not to use "contains" with old style classes?
Are there plans to provide standarized S4 versions/wrappers for list, matrix
etc... classes?


/E

PS.
R : Copyright 2004, The R Foundation for Statistical Computing
Version 1.9.1 Patched (2004-08-30), ISBN 3-900051-00-3

Yes, there are differences, and you should expect some. 

That as(object, "matrix") preserves dimnames is hardly an accident,
since dimnames are part of the definition of a matrix.

as(object, "list") works like all coercion to basic datatypes.  The
method uses the corresponding old-style as.<class> function.  In
addition, if coercion is strict, all attributes are removed. See ?as.

The effect of this is that names would be removed, unless you used
as(object, "list", strict=FALSE).  (The coerce in dispatching methods
is
NOT strict, so a method for class "list" would see the names.)

This is consistent behavior, but yes, the elimination of names may be
surprising.  It's possible that future versions might treat names
differently from other attributes, but there would need to be a strong
argument.

The workaround is to define a new class that trivially extends "list",
and then define classes to extend that instead:

 setClass("listWithNames", "list")

Then if x is an object from a class defined with
contains="listWithNames", rather than contains="list",
  as(x, "listWithNames")
would preserve names.

The problem here is that the definition of class (or dataype) "list"
was
never very clear as to whether names were an attribute or part of the
datatype.  The green book and S-Plus take the view that lists are just
vectors, and the current R implementation follows that interpretation.

(There is also in S-Plus a different class, "named", specifically for
the applications that lists with names typically deal with.  But the
spirit of that class is NOT equivalent to lists with a names attribute.)

As for extending basic datatypes by S4 classes, that's generally OK,
provided you understand the consequences (pages 314-316 of the green
book have a brief discussion).

And as for "wrapper" classes, there are no immediate plans.  The basic
datatypes do have class definitions:

R> getClass("list")

No Slots, prototype of class "list"

Extends: "vector"

The main candidates for wrapper classes are "matrix" and
"array", since
these do not have a fixed set of attributes (they may or may not have
"dimnames").  Class "ts" is already a formal class.

John Chambers

Wolski wrote:
> 
> Hello!
> 
> as(object,"list") and as(object,"matrix") behave quite
differently if it comes to their attributes.
> I define two classes. One of them "contains" a "list"
the other a "matrix"
> setClass("myclass"
> ,representation(info="character")
> ,contains="matrix"
> )
> setClass("mylist"
> ,representation(info="character")
> ,contains="list"
> )
> 
> #init
> dd<-matrix(1:6,nrow=2)
> rownames(dd)<-c("a","b")
> tt<-new("myclass",dd)
> 
> tmp<-vector("list",4)
> names(tmp)<-1:4
> ll <- new("mylist",tmp,info="foo")
> 
> rownames(tt)
> [1] "a" "b"
> > rownames(as(tt,"matrix"))
> [1] "a" "b"
> 
> > names(ll)
> [1] "1" "2" "3" "4"
> > names(as(ll,"list"))
> NULL
> #but
> names(ll@.Data)
> The difference in behaviour to which i would like to point your attention
is that as(object,"matrix") preserves the "dimnames" but at
the same time as(object,"list") drops the "names" attribute.
> 
> Is it recomended not to use "contains" with old style classes?
> Are there plans to provide standarized S4 versions/wrappers for list,
matrix etc... classes?
> 
> /E
> 
> PS.
> R : Copyright 2004, The R Foundation for Statistical Computing
> Version 1.9.1 Patched (2004-08-30), ISBN 3-900051-00-3
> 
> ______________________________________________
> R-devel@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
-- 
John M. Chambers                  jmc@bell-labs.com
Bell Labs, Lucent Technologies    office: (908)582-2681
700 Mountain Avenue, Room 2C-282  fax:    (908)582-3340
Murray Hill, NJ  07974            web: http://www.cs.bell-labs.com/~jmc

Hi!

Forgot to ask: Can you restore the behaviour from version 1.9.1? Please.
I have 2 packages on Bioconductor and one almost ready which rely on it. And as
I just tested the rewrite which changes the inheritance to a contains relation
would have more consequences then I thought and not be as easy.

Please let me know as soon as possible.

Sincerely yours
Eryk.