[Rd] R2.6 bug in arithmetics of S4 classes inherited from 'array', or intended behavior?

Oleg Sklyar osklyar at ebi.ac.uk
Sat Jul 28 18:03:52 CEST 2007


John, thank you for clarification. I just wonder that some packages
could also see that their behaviour has changed. Anyway, I redefined
Arith as the rest seems to work fine and also it is pretty straight
forward in my case what the result of Arith should be. Thanks again.
Oleg

On Sat, 2007-07-28 at 09:38 -0400, John Chambers wrote:
> Yes, this is the intended behavior.  Nothing to do with implicit
> generics, but with some ongoing changes to make class "structure" and
> related classes such as "array" give trustworthy results consistent
> with the S language idea of vector structures.  There's a short note
> in NEWS about the changes, added for the revision where the change was
> introduced.
> 
> As for programming your class, using showMethods("+") as below will
> tell you that the method is inherited from the group generic function
> "Ops", so a method for that function is needed, if you _really_ know
> what the results should be.
> 
> The behavior of the base package code for operations is unacceptable
> for trustworthy computations with classes.  Roughly, it takes the
> attributes of the left-hand operand and inserts those into the result
> of operating on the vector or array part.  So in the example below we
> have two  unrelated classes that both extend "array".  A mixed
> operation gives a result from class "A1" if done in one order and from
> "A2" if done in the other order.  Both are likely wrong and quite
> possibly seriously misleading.
> 
> Even if the two objects came from the same class, copying one set of
> attributes makes sense only if the two sets are identical (an
> expensive test) and not necessarily even then, if any of the
> attributes would be invalidated by the operation.
> 
> Two acceptable results are either to reduce to the common inheritance
> (array, here) or to call the result undefined and throw an error.  The
> current implementation in 2.6.0  is the first, as the example shows:
> 
> > setClass("A1", contains="array", representation(flag = "character"))
> [1] "A1"
> > setClass("A2", contains="array", representation(positive =
> "logical"))
> [1] "A2"
> > a1 = new("A1", array(1:8, rep(2,3)), flag = "test")
> > a2 = new("A2", array(8:1, rep(2,3)), positive = rep(TRUE,8))
> > a1+a2
> , , 1
> 
>      [,1] [,2]
> [1,]    9    9
> [2,]    9    9
> 
> , , 2
> 
>      [,1] [,2]
> [1,]    9    9
> [2,]    9    9
> 
> > showMethods("+")
> Function: + (package base)
> e1="A1", e2="A2"
>     (inherited from: e1="array", e2="array")
>     (definition from function "Ops")
> 
> 
> Oleg Sklyar wrote: 
> > Narrowed down to Rev. 42246:
> > 
> > * ~/R/Rsvn: svn up -r42245
> > * ~/R/Rsvn: make
> > * ~/R/Rsvn: Rsvn CMD INSTALL ~/tmp/EBImage_2.1.13.tar.gz
> > * ~/R/Rsvn: Rsvn
> >   
> > > version$version.string
> > >     
> > [1] "R version 2.6.0 Under development (unstable) (2007-07-16 r42245)"
> >   
> > > library(EBImage)
> > > a <- Image(0,c(2,2))
> > > class(a+a)
> > >     
> > [1] "Image"
> > attr(,"package")
> > [1] "EBImage"
> > 
> > * ~/R/Rsvn: svn up -r42246
> > * ~/R/Rsvn: make
> > * ~/R/Rsvn: Rsvn CMD INSTALL ~/tmp/EBImage_2.1.13.tar.gz
> > * ~/R/Rsvn: Rsvn
> >   
> > > version$version.string
> > >     
> > [1] "R version 2.6.0 Under development (unstable) (2007-07-16 r42246)"
> >   
> > > library(EBImage)
> > > a <- Image(0,c(2,2))
> > > class(a+a)
> > >     
> > [1] "array"
> > 
> > * ~/R/Rsvn: svn log -r42246
> > ------------------------------------------------------------------------
> > r42246 | jmc | 2007-07-16 14:32:16 +0100 (Mon, 16 Jul 2007) | 1 line
> > 
> > implicitGeneric() and structure class
> > ------------------------------------------------------------------------
> > 
> >



More information about the R-devel mailing list