S3-style Objects and S4-class Objects
Old-style (S3) classes may be registered as S4 classes (by calling
setOldClass), and many have been. These classes can
then be contained in (that is, superclasses of) regular S4 classes,
allowing formal methods and slots to be added to the S3 behavior. The
S3Part extracts or replaces the S3 part of such an object.
S3Class extracts or
replaces the S3-style class.
S3Class also applies to object
from an S4 class with
S3methods=TRUE in the call to
See the details below. Also discussed are S3 <-> S4 coercion; see the section S3 and S4 objects
S3Part(object, strictS3 = FALSE, S3Class)S3Part(object, strictS3 = FALSE, needClass = ) <- valueS3Class(object)S3Class(object) <- valueisXS3Class(classDef)slotsFromS3(object)
- an object from some class that extends a registered
S3 class, usually because the class has as
one of its superclasses an S3 class registered by a call to
setOldClass, or from a class that extends a basic vector, matrix or array object type. See the details.
For most of the functions, an S3 object can also be supplied, with the interpretation that it is its own S3 part.
TRUE, the value returned by
S3Partwill be an S3 object, with all the S4 slots removed. Otherwise, an S4 object will always be returned; for example, from the S4 class created by
setOldClassas a proxy for an S3 class, rather than the underlying S3 object.
charactervector to be stored as the S3 class slot in the object. Usually, and by default, retains the slot from
- Require that the replacement value be this class or a subclass of it.
S3Part<-, the replacement value for the S3 part of the object. This does not need to be an S4 object; in fact, the usual way to create objects from these classes is by giving an S3 object of the right class as an argument to
S3Class<-, the character vector that will be used as a proxy for
class(x)in S3 method dispatch. This replacement function can be used to control S3 per-object method selection.
- a class definition object, as returned by
Classes that register S3 classes by a call to
setOldClass have slot
".S3Class" to hold
the corresponding S3 vector of class strings.
The prototype of such
a class has the value for this slot determined by the argument to
Other S4 classes will have the same slot if the argument
S3methods = TRUE is supplied to
this case, the slot is set to the S4 inheritance of the class.
New S4 classes that extend (contain) such
classes also have the same slot, and by default the prototype has
the value determined by the
contains= argument to
Individual objects from the S4 class may
an S3 class corresponding to the value in the prototype or to an
(S3) subclass of that value. See the examples below.
strictS3 = TRUE constructs the underlying
S3 object by eliminating
all the formally defined slots and turning off the S4 bit of the
strictS3 = FALSE the object returned is from
the corresponding S4 class. For consistency and generality,
S3Part() works also for classes that extend the basic vector,
matrix and array classes. Since R is somewhat arbitrary about what
it treats as an S3 class (
"ts" is, but
S3Part() tries to return an S3 (that is, non-S4) object
whenever the S4 class has a suitable superclass, of either S3 or
basic object type.
One general application that relies on this generality is to use
S3Part() to get a superclass object that is guaranteed not to
be an S4 object. If you are calling some function that checks for
S4 objects, you need to be careful not to end up in a closed loop
fooS3, which checks for an S4 object and
fooS4 again, maybe indirectly). Using
strictS3 = TRUE is a mechanism to avoid such loops.
Because the contents of S3 class objects have no definition or guarantee, the computations involving S3 parts do not check for slot validity. Slots are implemented internally in R as attributes, which are copied when present in the S3 part. Grave problems can occur if an S4 class extending an S3 class uses the name of an S3 attribute as the name of an S4 slot, and S3 code sets the attribute to an object from an invalid class according to the S4 definition.
S3Part can and should be avoided by simply
coercing objects to the desired class; methods are automatically
defined to deal correctly with the slots when
called to extract or replace superclass objects.
slotsFromS3() is a generic function used
internally to access the slots associated with the S3 part of the
object. Methods for this function are created automatically when
setOldClass is called with the
argument. Usually, there is only one S3 slot, containing the S3
class, but the
S4Class argument may provide additional slots,
in the case that the S3 class has some guaranteed attributes that
can be used as formal S4 slots. See the corresponding section in
the documentation of
S3Part: Returns or sets the S3 information (and possibly some S4 slots as well, depending on arguments
keepSlots). See the discussion of argument
strictabove. If it is
TRUEthe value returned is an S3 object.
S3Class: Returns or sets the character vector of S3 class(es) stored in the object, if the class has the corresponding
.S3Classslot. Currently, the function defaults to
FALSEaccording to whether the class defined by
ClassDefextends S3 classes (specifically, whether it has the slot for holding the S3 class).
slotsFromS3: returns a list of the relevant slot classes, or an empty list for any other object.
S3 and S4 Objects: Conversion Mechanisms
Objects in R have an internal bit that indicates whether or not to
treat the object as coming from an S4 class. This bit is tested by
isS4 and can be set on or off by
The latter function, however, does no checking or interpretation;
you should only use it if you are very certain every detail has been
handled correctly. As a friendlier alternative, methods have been defined for coercing
to the virtual classes
"S4". The expressions
as(object, "S3") and
as(object, "S4") return S3
and S4 objects, respectively. In addition, they attempt
to do conversions in a valid way, and also check validity when
coercing to S4. The expression
as(object, "S3") can be used in two ways. For
objects from one of the registered S3 classes, the expression will
ensure that the class attribute is the full multi-string S3 class
class(object). If the registered class has known
attribute/slots, these will also be provided. Another use of
as(object, "S3") is to take an S4 object and
turn it into an S3 object with corresponding attributes. This is
only meaningful with S4 classes that have a data part. If you want
to operate on the object without invoking S4 methods, this
conversion is usually the safest way. The expression
as(object, "S4") will use the attributes in
the object to create an object from the S4 definition of
class(object). This is a general mechanism to create
partially defined version of S4 objects via S3 computations (not
much different from invoking
new with corresponding
arguments, but usable in this form even if the S4 object has an
initialize method with different arguments).
Chambers, John M. (2008) Software for Data Analysis: Programming with R Springer. (For the R version).
Chambers, John M. (1998) Programming with Data Springer (For the original S4 version.)
## two examples extending S3 class "lm", class "xlm" directly ## and "ylm" indirectly setClass("xlm", representation(eps = "numeric"), contains = "lm") setClass("ylm", representation(header = "character"), contains = "xlm") ## lm.D9 is as computed in the example for stats::lm y1 <- new("ylm", lm.D9, header = "test", eps = .1) xx <- new("xlm", lm.D9, eps =.1) y2 <- new("ylm", xx, header = "test") stopifnot(inherits(y2, "lm")) stopifnot(identical(y1, y2)) stopifnot(identical(S3Part(y1, strict = TRUE), lm.D9)) ## note the these classes can insert an S3 subclass of "lm" as the S3 part: myData <- data.frame(time = 1:10, y = (1:10)^.5) myLm <- lm(cbind(y, y^3) ~ time, myData) # S3 class: c("mlm", "lm") ym1 <- new("ylm", myLm, header = "Example", eps = 0.) ##similar classes to "xlm" and "ylm", but extending S3 class c("mlm", "lm") setClass("xmm", representation(eps = "numeric"), contains = "mlm") setClass("ymm", representation(header="character"), contains = "xmm") ym2 <- new("ymm", myLm, header = "Example2", eps = .001) # but for class "ymm", an S3 part of class "lm" is an error: try(new("ymm", lm.D9, header = "Example2", eps = .001)) setClass("dataFrameD", representation(date = "Date"), contains = "data.frame") myDD <- new("dataFrameD", myData, date = Sys.Date()) ## S3Part() applied to classes with a data part (.Data slot) setClass("NumX", contains="numeric", representation(id="character")) nn <- new("NumX", 1:10, id="test") stopifnot(identical(1:10, S3Part(nn, strict = TRUE))) m1 <- cbind(group, weight) setClass("MatX", contains = "matrix", representation(date = "Date")) mx1 <- new("MatX", m1, date = Sys.Date()) stopifnot(identical(m1, S3Part(mx1, strict = TRUE)))