R devel - Aug 2006 - S4 classes and objects -- fixed structure? No...

Hello.

Suppose you define a new S4-class, say
>  setClass("track", representation(x="numeric",
y="numeric"))
Don't worry if you have a deja vu, it's from the help page.
Your new class is said to have a fixed structure: two slots, x, and y,
and that should apply to all objects you construct as members of that
class.
>  tr <- new( "track" )
Now do the following:
>  tr[ "ping" ] <- "pong"
>  tr$bingo <- "bongo"
>  tr[[ "blaa" ]] <- c( 200, 300 )
Of course you can use the well known operators to access these "list
entries
in a S4-class object":
>  tr[ "blaa" ]
>  etc. 
You see what can happen if you decide to do a bit stress testing. The
question is not whether my examples makes sense or not. The question is if
these examples should be possible at all.
I wonder which is true,
-- the theory (Chambers, 1998, p. 279ff; Venables and Ripley, 2000, p. 99ff:
"... All objects in a [S4] class must have the same structure. ...";
etc.)
-- or the actual implementation in R (see my example, which successfully
violates the design principles of the language)

Bug, or feature? Any clarifications are appreciated -- it may be my
half-cooked knowledge that I find this confusing and dangerous.

Best 

Joerg Beyer

P.S.: 
Oh, the specs: PowerMac G4/400 PCI -- 1GB RAM -- Mac OS X 10.4.6 -- R 2.2.1

Under R 2.3.1 these work as you indicate but under R 2.4.0
they all give errors:
>  setClass("track", representation(x="numeric",
y="numeric"))
[1] "track"
>  tr <- new( "track" )
>  tr[ "ping" ] <- "pong"
Error in "[<-"(`*tmp*`, "ping", value = "pong")
:
        object is not subsettable
>  tr$bingo <- "bongo"
Error: cannot coerce type S4 to list vector
In addition: Warning message:
Coercing LHS to a list
>  tr[[ "blaa" ]] <- c( 200, 300 )
Error in "[[<-"(`*tmp*`, "blaa", value = c(200, 300)) :
        object is not subsettable
>  tr[ "blaa" ]
Error in tr["blaa"] : object is not
subsettable
> R.version.string
[1] "R version 2.4.0 Under development (unstable) (2006-08-08 r38825)"


On 8/28/06, J?rg Beyer <Beyerj at students.uni-marburg.de>
wrote:
> Hello.
>
> Suppose you define a new S4-class, say
> >  setClass("track", representation(x="numeric",
y="numeric"))
>
> Don't worry if you have a deja vu, it's from the help page.
> Your new class is said to have a fixed structure: two slots, x, and y,
> and that should apply to all objects you construct as members of that
class.
> >  tr <- new( "track" )
>
> Now do the following:
> >  tr[ "ping" ] <- "pong"
> >  tr$bingo <- "bongo"
> >  tr[[ "blaa" ]] <- c( 200, 300 )
>
> Of course you can use the well known operators to access these "list
entries
> in a S4-class object":
> >  tr[ "blaa" ]
> >  etc.
>
> You see what can happen if you decide to do a bit stress testing. The
> question is not whether my examples makes sense or not. The question is if
> these examples should be possible at all.
> I wonder which is true,
> -- the theory (Chambers, 1998, p. 279ff; Venables and Ripley, 2000, p.
99ff:
> "... All objects in a [S4] class must have the same structure.
..."; etc.)
> -- or the actual implementation in R (see my example, which successfully
> violates the design principles of the language)
>
> Bug, or feature? Any clarifications are appreciated -- it may be my
> half-cooked knowledge that I find this confusing and dangerous.
>
> Best
>
> Joerg Beyer
>
> P.S.:
> Oh, the specs: PowerMac G4/400 PCI -- 1GB RAM -- Mac OS X 10.4.6 -- R 2.2.1
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

I think you're right---this shouldn't happen in theory, but it does
because of
the internal representation of S4 objects in R.

In R devel (to be 2.4.0), this changes and I believe your example will no longer
work.

-roger

J?rg Beyer wrote:
> Hello.
> 
> Suppose you define a new S4-class, say
>>  setClass("track", representation(x="numeric",
y="numeric"))
> 
> Don't worry if you have a deja vu, it's from the help page.
> Your new class is said to have a fixed structure: two slots, x, and y,
> and that should apply to all objects you construct as members of that
class.
>>  tr <- new( "track" )
> 
> Now do the following:
>>  tr[ "ping" ] <- "pong"
>>  tr$bingo <- "bongo"
>>  tr[[ "blaa" ]] <- c( 200, 300 )
> 
> Of course you can use the well known operators to access these "list
entries
> in a S4-class object":
>>  tr[ "blaa" ]
>>  etc. 
> 
> You see what can happen if you decide to do a bit stress testing. The
> question is not whether my examples makes sense or not. The question is if
> these examples should be possible at all.
> I wonder which is true,
> -- the theory (Chambers, 1998, p. 279ff; Venables and Ripley, 2000, p.
99ff:
> "... All objects in a [S4] class must have the same structure.
..."; etc.)
> -- or the actual implementation in R (see my example, which successfully
> violates the design principles of the language)
> 
> Bug, or feature? Any clarifications are appreciated -- it may be my
> half-cooked knowledge that I find this confusing and dangerous.
> 
> Best 
> 
> Joerg Beyer
> 
> P.S.: 
> Oh, the specs: PowerMac G4/400 PCI -- 1GB RAM -- Mac OS X 10.4.6 -- R 2.2.1
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
> 
-- 
Roger D. Peng  |  http://www.biostat.jhsph.edu/~rpeng/