Learn R Programming

mvbutils (version 2.8.232)

mlocal: Macro-like functions

Description

mlocal lets you write a function whose statements are executed in its caller's frame, rather than in its own frame.

Usage

# Use only as wrapper of function body, like this:
# my.fun <- function(..., nlocal=sys.parent()) mlocal( expr)
# ... should be replaced by the arguments of "my.fun"
# expr should be replaced by the code of "my.fun"
# nlocal should always be included as shown
mlocal( expr) # Don't use it like this!

Arguments

expr

the function code, normally a braced expression

Value

As per your function; also see local.return.

Details

Sometimes it's useful to write a "child" function that can create and modify variables in its parent directly, without using assign or <<- (note that <<- will only work on variables that exist already). This can make for clearer, more modular programming; for example, tedious initializations of many variables can be hidden inside an initialize() statement. The definition of an mlocal function does not have to occur within its caller; the mlocal function can exist as a completely separate R object.

mlocal functions can have arguments just like normal functions. These arguments will temporarily hide any objects of the same name in the nlocal frame (i.e. the calling frame). When the mlocal function exits, its arguments will be deleted from the calling frame and the hidden objects (if any) will be restored. Sometimes it's desirable to avoid cluttering the calling frame with variables that only matter to the mlocal function. A useful convention is to "declare" such temporary variables in your function definition, as defaultless arguments after the nlocal argument.

The nlocal argument of an mlocal function-- which must ALWAYS be included in the definition, with the default specified as sys.parent()-- can normally be omitted when invoking your mlocal function. However, you will need to set it explicitly when your function is to be called by another, e.g. lapply; see the third example. A more daring usage is to call e.g. fun.mlocal(nlocal=another.frame.number) so that the statements in fun.mlocal get executed in a completely different frame. A convoluted example can be found in the (internal) function find.debug.HQ in the debug package, which creates a frame and then defines a large number of variables in it by calling setup.debug.admin(nlocal=new.frame.number). As of 2016, you can also set nlocal to be an environment.

mlocal functions can be nested, though this gets confusing. By default, all evaluation will happen in the same frame, that of the original caller.

Note that (at least at present) all arguments are evaluated as soon as your mlocal function is invoked, rather than by the usual lazy evaluation mechanism. Missing arguments are still OK, though.

If you call return in an mlocal function, you must call local.return too.

on.exit doesn't work properly. If you want to have exit code in the mlocal function itself, use local.on.exit. I can't find any way to set the exit code in the calling function from within an mlocal function. (Not checked for some years)

Frame-dependent functions (sys.parent()) etc. will not do what you expect inside an mlocal function. For R versions between at least 1.8 and 2.15, calling the mvb... versions will return information about the caller of the current mlocal() function caller (or the original caller, if there is a chain of mlocals). For example, mvb.sys.function() returns the definition of the caller, and mvb.sys.parent() the frame of the caller's parent. Note that sys.frame( mvb.sys.nframe()) gives the current environment (i.e. where all the variables live), because this is shared between the caller and the mlocal function. Other behaviour seems to depend on the version of R, and in R 2.15 I don't know how to access the definition of the mlocal function itself. This means, for example, that you can't reliably access attributes of the mlocal function itself, though you can access those of its caller via e.g. attr( mvb.sys.function(), "thing").

See Also

local.return, local.on.exit, do.in.envir, localfuncs, and R-news 1/3 2001 for a related approach to "macros"

Examples

Run this code
# NOT RUN {
# Tidiness and variable creation
init <- function( nlocal=sys.parent()) mlocal( sqr.a <- a*a)
ffout <- function( a) { init(); sqr.a }
ffout( 5) # 25
# Parameters and temporary variables
ffin <- function( n, nlocal=sys.parent(), a, i) mlocal({
    # this "n" and "a" will temporarily replace caller's "n" and "a"
    print( n)
    a <- 1
    for( i in 1:n)
      a <- a*x
    a
  })
x.to.the.n.plus.1 <- function( x, n) {
    print( ffin( n+1))
    print( n)
    print( ls())
  }
x.to.the.n.plus.1( 3, 2) # prints as follows:
# [1] 3 (in "ffin")
# [1] 27 (result of "ffin")
# [1] 2 (original n)
# [1] "n" "x" (vars in "x.to.the..."-- NB no a or i)
# Use of "nlocal"
ffin <- function( i, nlocal=sys.parent()) mlocal( a <- a+i )
ffout <- function( ivec) { a <- 0; sapply( ivec, ffin, nlocal=sys.nframe()) }
ffout( 1:3) # 1 3 6
# }

Run the code above in your browser using DataLab