[Rd] Validity glitch when contains="matrix"

John Chambers jmc at r-project.org
Thu May 6 18:34:48 CEST 2010


Re: the validity bug.  It's just as your example suggests:  the 
inherited initialize() method for "matrix" fails to call validObject(). 
  Should be easy to fix.  (Although it points out that the code should 
perhaps be reorganized so the initialize() method is not responsible for 
checking validity.)


Re: your other question:

On 5/6/10 4:09 AM, Martin Maechler wrote:
>>>>>> Daniel Murphy<chiefmurphy at gmail.com>
>>>>>>      on Wed, 5 May 2010 22:08:06 -0700 writes:
<snip>
>      >  Should I
>      >  1) always put "matrix" into the setClass representation argument instead of
>      >  the contains argument, or
>      >  2) use contains="numeric", put the matrix's dims and dimnames attributes
>      >  into slots, and rely on a constructor to populate the instance?
>

Either is possible, but if you really want your new objects to inherit 
the properties of a matrix, your initial idea of contains="matrix" is 
the natural choice for 2) (once the bug is fixed, but even before for 
most purposes). The choice is to inherit from matrix or have matrix as a 
slot (what Smalltalk called "is a" versus "has a").

The choice is as always whether you want to inherit all the methods and 
then override the ones that DON'T make sense, or put the matrix in a 
slot and write all the methods that DO make sense.

Neither choice fits all examples but "matrix" is special, because R 
treats them in a special (perhaps "weird" would apply) way:  they are 
not a class (not even an S3 class) but much code recognizes them 
internally, meaning you inherit a great deal of stuff.

If your new class of objects are supposed to act like matrices most of 
the time, the contains= version may require a lot less programming. On 
the other hand, if you planned to store the data in a non-standard way 
(as the Matrix package does) then you really don't want to inherit the 
standard methods because any you failed to override could be disastrous.


> Well, one can go the very long and "stable" way as we did in the
> Matrix package...
>
> I'm not sure I would recommend that for you in your situation.
> ...
> not the least because you *could* use the Matrix package if you
> want to use such formal matrices with its thousands of methods.
>
> Martin Maechler,
> ETH Zurich
>
>      >  Option 2 seems most "stable".
>
>      >  Thanks,
>      >  Dan Murphy
>
>      >  Windows Vista, R version 2.11.0 (2010-04-22)
>
>      >  [[alternative HTML version deleted]]
>
>      >  ______________________________________________
>      >  R-devel at r-project.org mailing list
>      >  https://stat.ethz.ch/mailman/listinfo/r-devel
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel



More information about the R-devel mailing list