quotedargs: Facility for using quoted arguments
In quotedargs: A Way of Writing Functions that Quote their Arguments

Description Usage Arguments Details Value See Also Examples

This package assists with writing functions that automatically quote their arguments, but that may also wish to evalute them, in their original environment. These quoted arguments can be passed to other functions that quote their arguments, with proper passing of their quoting environment. It is also possible to set up a variable so that it looks just like a quoted argument.

In the simplest applications of this package, one can think of calling quoted_arg(x) (see below) as altering the default meaning of a reference to an argument x from the value of the actual argument, with the expression passed for x accessible via substitute(x), so that instead the default is the expression passed for x, with the value accessible via quoted_eval(x), as described below.

However, in more complex applications, the facilities provided by this package are more than just a convenient change of defaults, as they allow functions that quote arguments to be combined in ways that would otherwise be difficult.

quoted_arg (...)
quoted_eval (arg)
quoted_environment (arg)
notquoted (x)

quoted_assign (name, value, eval.env, assign.env = parent.frame())

`...`	names (unquoted) of function arguments that should be quoted
`arg`	the name of a quoted function argument (unquoted)
`x`	any expression.
`name`	the name (as a character string or symbol) of a variable to assign to
`value`	a value to assign to the variable `name`
`eval.env`	the environment in which `value` may be evaluated; may be missing, with default as described below
`assign.env`	the environment in which to assign to `name`

The quoted_arg function should be called at the start of a function that uses quoted arguments, with arguments that are the (unquoted) names of the arguments that should be quoted. After the call of quoted_arg, simple references to these arguments will give the expressions passed as arguments, rather than the values of these expressions. Currently, ..., ..1, ..2, etc. are not allowed as arguments of quoted_arg.

The caller of a function can disable any quoting with quoted_arg by passing notquoted(x) instead of x, in which case x will be evaluated when quoted_arg is called, and references to x will deliver this value, not the expression.

To obtain the value of a quoted argument, quoted_eval can be used. The evaluation will be done in the environment of the quoted expression. If quoted_eval is called more than once for the same argument, the argument will be evaluated that many times (possibly with different results). If the actual argument used notquoted, quoted_eval will simply return the already-evaluated argument.

The environment used by quoted_eval can be obtained with quoted_environment, which will be NULL if the actual argument used notquoted, and may be emptyenv() if the expression is self-evaluating, and hence its evaluation would not reference an environment.

When a quoted argument is passed as an argument to another function that quotes that argument, the quoted argument received will be the argument originally passed, not a quoting of the name of the quoted argument.

A variable can be set up so that it looks like a quoted argument using quoted_assign.

The name of the variable to set is specified by the name argument of quoted_assign, which must evaluate to a single character string or a symbol. The environment in which this variable is assigned is specified by the assign.env argument, which defaults to the current environment (the parent frame of quoted_assign).

The value argument to quoted_assign is evaluated to obtain an expression analogous to an actual argument, which is stored in the variable specified by name. The environment eval.env is stored with the assigned expression (in a “promise”), and will be used when evaluating this expression if quoted_eval is called for the assigned variable. If eval.env is missing, it defaults to the current environment, unless value is itself a quoted argument, in which case the default is quoted_environment(value). If the eval.env argument of quoted_assign is NULL, what is stored in name will look like a quoted argument in which the actual argument used notquoted, and evaluated to value, with the expression stored in the promise being the unevaluated form of value.

quoted_eval and quoted_environment return values as described above.

notquoted returns its argument.

quoted_arg and quoted_assign always return NULL.

substitute, for how to get at the expression passed when an argument is not quoted.

delayedAssign, for another function that is somewhat analogous to quoted_assign.

# A simple example in which both the expression passed and its value
# are used.

showmean <- function (v) {
    quoted_arg(v)
    cat ("Mean of", deparse(v), "is", mean(quoted_eval(v)), "\n")
}

showmean(100+(1:3))                   # Will print 100 + (1:3)
showmean(notquoted(100+(1:3)))        # Will print c(101, 102, 103)

# A function that uses the function above, passing along its quoted 
# argument.

showmeansummary <- function (u) {
    quoted_arg(u)
    cat("Summary: ")
    showmean(u)
}

showmeansummary(100+(1:3))            # Will print 100 + (1:3), not u!
showmeansummary(notquoted(100+(1:3))) # Will print c(101, 102, 103)

u <- v <- 100+(1:3)   # Evaluation of showmeansummary's argument
showmeansummary(u)    #   is done in the environment of the caller,
showmeansummary(v)    #   not that of showmeansummary or showmean


# An illustration of quoted arguments being evaluated many times.

prsim <- function (a,b,n) {
    quoted_arg(a,b)
    cat ("Running simulation to find probability that all\n")
    cat (deparse(a), "are greater than all", deparse(b), "\n")
    count <- 0
    for (i in 1:n) {
        if (min(quoted_eval(a)) > max(quoted_eval(b)))
            count <- count + 1
    }
    count / n
}

set.seed(1)
prsim (rexp(10,0.1), rnorm(10,1), 1000)


# Creating a variable that behaves like a quoted argument.

quoted_assign("x",quote(runif(1)))

set.seed(1)
cat (paste0("Two evaluations of ",deparse(x),": "),
     quoted_eval(x), quoted_eval(x),
     "\n")


# Examples of when quotation of an argument is passed on.

qfun1 <- function (x) { quoted_arg(x); list(x,quoted_eval(x)) }

qfun2 <- function (y) {
    quoted_arg(y)
    a <- y
    quoted_assign ("b", y)
    list(qfun1(y),qfun1((y)),qfun1(a),qfun1(b))
}

qfun2(1+2)


# Example of how quoted_arg and quoted_eval can be used to avoid
# copying of a large object.

sum_first_last1 <- function (v) {
    v[1] + v[length(v)]
}

sum_first_last2 <- function (v) {
    quoted_arg(v)
    quoted_eval(v)[1] + quoted_eval(v)[length(quoted_eval(v))]
}

f <- function (sumfl) {
    x <- 1:100000
    r <- sumfl(x)
    x[2] <- 0L
    r
}

f(sum_first_last1)  # x[2] <- 0L first copies x (in current R implementations)
f(sum_first_last2)  # x[2] <- 0L does not result in x being copied


# Example of using quotedargs to build functions that take as
# arguments expressions that may reference columns of a data
# frame and variables accessible in the caller's environment.
# The data frame columns take precedence, except that the data
# frame is skipped for expressions enclosed in O(...).

dfeval <- function (df, expr) {    # Find value of expression
    quoted_arg(expr)
    env <- new.env (parent = quoted_environment(expr), hash=FALSE)
    env$O <- function (z) { quoted_arg(z); eval(z,parent.env(environment())) }
    environment(env$O) <- env
    eval (expr, df, env)
}

dfchange <- function (df, expr) {  # Return data frame changed by assignments
    quoted_arg(expr)
    env <- new.env (parent = quoted_environment(expr), hash=FALSE)
    env$O <- function (z) { quoted_arg(z); eval(z,parent.env(environment())) }
    environment(env$O) <- env
    dfenv <- as.environment(df)
    parent.env(dfenv) <- env
    eval (expr, dfenv)
    as.data.frame (as.list (dfenv))
}

dfchange_var <- function (df, expr) {  # Actually change df variable passed
    quoted_arg(df,expr)
    newdf <- dfchange (quoted_eval(df), expr)
    assign (as.character(df), newdf, quoted_environment(df))
}

tstdf <- as.data.frame (list (x = 1:4, y = c("a","b","c","d")))

y <- 100
dfeval (tstdf, paste0(y,x))
dfeval (tstdf, x * O(y))

dfchange (tstdf, { z <- 10*x; x <- x + O(y) })

dfchange_var (tstdf, x <- 1000+x)
tstdf