[Rd] how to properly extend s3 data.frames with s4 classes?

[Ulf Martin]

Dear R Programmers!

After some time of using R I decided to work through John Chambers book 
"Programming with Data" to learn what these S4 classes are all about and 
how they work in R. (I regret not having picked up this rather fine book 
earlier!)

I know from the documentation and the mailing archives that S4 in R is 
not 100% the book and that there are issues especially with dataframes, 
but to my knowledge the following has not been reported yet.


Summary
-------
(a) When extending a S3 data.frame with a S4 class adding a slot, it 
seems to be impossible to initialize objects of these 
"ExtendedDataframes" (XDF) with S3 data.frames.

(b) Extending data.frames with an S4 class without a slot, i.e. creating 
a "WrappedDataframe" (WDF), seems to allow initialization with a 
data.frame, but the behaviour appears to be somewhat inconsistent.

(c) Trying to be "smart" by extending the WrappedDataframe from (b) by 
adding a slot, yields a similar behaviour than (a), i.e. initialization 
with a WDF object fails although WDF is an instance of an S4 class.

It is actually (c) that surprises me most.


Code
----
# (Should be pastable into an R session)
# R version is 2.4.1
#
# === Preliminaries ===
# (">" indicates output)
#
library("methods")
setOldClass("data.frame")
tdf <- data.frame(x=c(1,2), y=c(TRUE,FALSE)) # For testing purposes
#
# === (a) Exdended Dataframe Case ===
#
XDF <- "ExtendedDataframe" # Convenient shortcut
setClass(XDF, representation("data.frame", info="character"))
getClass(XDF)
#
# > Slots:
# >
# > Name:       info
# > Class: character
# >
# > Extends:
# > Class "data.frame", directly
# > Class "oldClass", by class "data.frame", distance 2
#
# So far everything looks good.
# But now,
#
new(XDF)                                 # a1)
new(XDF, data.frame())                   # a2)
new(XDF, tdf, info="Where is the data?") # a3)
#
# all yield:
#
# > An object of class "ExtendedDataframe"
# > NULL
# > <0 rows> (or 0-length row.names)
#
# Only (a3) additionally has
#
# > Slot "info":
# > [1] "Where is the data?"
#
# === (b) Wrapped Dataframe ===
#
WDF <- "WrappedDataframe"
setClass(WDF, representation("data.frame"))
getClass(WDF)
#
# > No Slots, prototype of class "S4"  # N.B.!
# >
# > Extends:
# > Class "data.frame", directly
# > Class "oldClass", by class "data.frame", distance 2
#
new(WDF)
#
# > <S4 Type Object>
# > attr(,"class")
# > [1] "WrappedDataframe"
# > attr(,"class")attr(,"package")
# > [1] ".GlobalEnv"
#
# Now we have attributes -- there wheren't any with XDF.
# Thus, not supplying a slot adds attributes -- confusing.
#
# Now: Initialization with an empty data.frame instead of nothing:
#
new(WDF, data.frame())
#
# > An object of class "WrappedDataframe"
# > Slot "row.names":
# > character(0)
# > Warning message:
# > missing package slot (.GlobalEnv) in object of class
# > "WrappedDataframe" (package info added) in: initialize(value, ...)
#
# OBS! Now there is
#  (i) a slot "row.names" -- which is wrong
#      since WDFs aren't suposed to have any slots;
# (ii) an odd warning about another missing slot
#      (presumably called "package" but the message is
#      somewhat ambigous).
#
# But at least
#
new(WDF, tdf)
#
# yields:
#
# > $x
# > [1] 1 2
# >
# > $y
# > [1]  TRUE FALSE
# >
# > attr(,"row.names")
# > [1] 1 2
# > attr(,"class")
# > [1] "WrappedDataframe"
# > attr(,"class")attr(,"package")
# > [1] ".GlobalEnv"
# > Warning message:
# > missing package slot (.GlobalEnv) in object of class
# > "WrappedDataframe" (package info added) in: initialize(value, ...)
#
# So, at least the data seems to be there. Let's use this one.
#
wdf <- new(WDF, tdf)
#
# === (c) "Smart" Dataframes ===
#
SDF <- "SmartDataframe"
setClass(SDF, representation(WDF, info="character"))
getClass(SDF)
#
# > Slots:
# >
# > Name:       info
# > Class: character
# >
# > Extends:
# > Class "WrappedDataframe", directly
# > Class "data.frame", by class "WrappedDataframe", distance 2
# > Class "oldClass", by class "WrappedDataframe", distance 3
#
# Now I would expect this:
#
new(SDF,wdf)
#
# to show the data in wdf, but in fact I get:
#
# > An object of class "SmartDataframe"
# > NULL
# > <0 rows> (or 0-length row.names)
# > Slot "info":
# > character(0)
#
# which is the same as:
#
new(SDF)
#
# or
#
new(SDF, data.frame())
#
# The slot does get initialized, though
#
new(SDF,wdf,info="Where is the data?")
new(SDF,tdf,info="Where is the data?")
#
# END OF CODE


Further Remarks
---------------
The rationale behind being able to extend S3 data.frames with S4 classes 
is that
(a) there is so much legacy code for data.frames (they are the 
foundation of the data part in "programming with data");
(b) S4 classes allow for validation, multiple dispatch, etc.

I also wonder why the R developers chose this "setOldClass" way of 
making use of S3 classes rather than adding a clean set of wrapper 
classes that delegate calls to them cleanly down to their resp. S3 
companions (i.e. a "Methods" package (capital "M") with "Character", 
"Numeric", "List", "Dataframe", etc.). The present situation appears to 
be somewhat messy.


Anyway -- a great tool and great work!
Cheers!
Ulf Martin