methods (version 3.5.3)

LocalReferenceClasses: Localized Objects based on Reference Classes

Description

Local reference classes are modified ReferenceClasses that isolate the objects to the local frame. Therefore, they do not propagate changes back to the calling environment. At the same time, they use the reference field semantics locally, avoiding the automatic duplication applied to standard R objects.

The current implementation has no special construction. To create a local reference class, call setRefClass() with a contains= argument that includes "localRefClass". See the example below.

Local reference classes operate essentially as do regular, functional classes in R; that is, changes are made by assignment and take place in the local frame. The essential difference is that replacement operations (like the change to the twiddle field in the example) do not cause duplication of the entire object, as would be the case for a formal class or for data with attributes or in a named list. The purpose is to allow large objects in some fields that are not changed along with potentially frequent changes to other fields, but without copying the large fields.

Usage

setRefClass(Class, fields = , contains = c("localRefClass",....),
     methods =, where =, ...)

Details

Localization of objects is only partially automated in the current implementation. Replacement expressions using the $<- operator are safe.

However, if reference methods for the class themselves modify fields, using <<-, for example, then one must ensure that the object is local to the relevant frame before any such method is called. Otherwise, standard reference class behavior still prevails.

There are two ways to ensure locality. The direct way is to invoke the special method x$ensureLocal() on the object. The other way is to modify a field explicitly by x$field <- ... It's only necessary that one or the other of these happens once for each object, in order to trigger the shallow copy that provides locality for the references. In the example below, we show both mechanisms.

However it's done, localization must occur before any methods make changes. (Eventually, some use of code tools should at least largely automate this process, although it may be difficult to guarantee success under arbitrary circumstances.)

Examples

# NOT RUN {
## class "myIter" has a BigData field for the real (big) data
## and a "twiddle" field for some parameters that it twiddles
## ( for some reason)

myIter <- setRefClass("myIter", contains = "localRefClass",
  fields = list(BigData = "numeric", twiddle = "numeric"))

tw <- rnorm(3)
x1 <- myIter(BigData = rnorm(1000), twiddle = tw) # OK, not REALLY big

twiddler <- function(x, n) {
  x$ensureLocal() # see the Details.  Not really needed in this example
  for(i in seq(length = n)) {
      x$twiddle <- x$twiddle + rnorm(length(x$twiddle))
      ## then do something ....
      ## Snooping in gdb, etc will show that x$BigData is not copied
  }
  return(x)
}

x2 <- twiddler(x1, 10)

stopifnot(identical(x1$twiddle, tw), !identical(x1$twiddle, x2$twiddle))

# }