firmly: Apply a function firmly

In valaddin: Functional Input Validation

Apply a function firmly

Description

firmly transforms a function into a function with input validation checks. loosely undoes the application of firmly, by returning the original function (without checks). is_firm is a predicate function that checks whether an object is a firmly applied function, i.e., a function created by firmly.

Use %checkin% to apply firmly as an operator. Since this allows you to keep checks and arguments adjacent, it is the preferred way to use firmly in scripts and packages.

Usage

firmly(.f, ..., .checklist = list(), .warn_missing = character(),
       .error_class = character())

.checks %checkin% .f

loosely(.f, .keep_check = FALSE, .keep_warning = FALSE)

is_firm(x)

Arguments

.f	Interpreted function, i.e., closure.
...	Input-validation check formula(e).
.checklist	List of check formulae. (These are combined with check formulae provided via ....)
.warn_missing	Arguments of .f whose absence should raise a warning (character).
.error_class	Subclass of the error condition to be raised when an input validation error occurs (character).
.checks	List of check formulae, optionally containing character vectors named .warn_missing, .error_class, corresponding to the similarly named arguments.
.keep_check, .keep_warning	Should existing checks, resp. missing-argument warnings, be kept?
x	Object to test.

Check Formulae

An input validation check is specified by a check formula, a special formula of the form

<scope> ~ <predicate>

where the right-hand side expresses what to check, and the left-hand side expresses where to check it.

The right-hand side <predicate> is a predicate function, i.e, a one-argument function that returns either TRUE or FALSE. It is the condition to check/enforce. The left-hand side <scope> is an expression specifying what the condition is to be applied to: whether the condition is to be applied to all (non-...) arguments of .f (the case of “global scope”), or whether the condition is to be selectively applied to certain expressions of the arguments (the case of “local scope”).

According to scope, there are two classes of check formulae:

• Check formulae of global scope

  <string> ~ <predicate>

  ~<predicate>

  \item \strong{Check formulae of local scope}
    \preformatted{list(<check_item>, <check_item>, ...) ~ <predicate>}
  

Check Formulae of Global Scope

A global check formula is a succinct way of asserting that the function <predicate> returns TRUE when called on each (non-...) argument of .f. Each argument for which <predicate> fails—returns FALSE or is itself not evaluable—produces an error message, which is auto-generated unless a custom error message is supplied by specifying the string <string>.

\subsection{Example}{
  The condition that all (non-\code{\dots}) arguments of a function must
  be numerical can be enforced by the check formula
  \preformatted{~is.numeric}
  or
  \preformatted{"Not numeric" ~ is.numeric}
  if the custom error message \code{"Not numeric"} is to be used (in lieu
  of an auto-generated error message).
}

Check Formulae of Local Scope

A local check formula imposes argument-specific conditions. Each check item <check_item> is a formula of the form ~ <expression> (one-sided) or <string> ~ <expression>; it imposes the condition that the function <predicate> is TRUE for the expression <expression>. As for global check formulae, each check item for which <predicate> fails produces an error message, which is auto-generated unless a custom error message is supplied by a string as part of the left-hand side of the check item (formula).

\subsection{Example}{
  The condition that \code{x} and \code{y} must differ for the function
  \code{function(x, y) {1 / (x - y)}} can be enforced by the local
  check formula
  \preformatted{list(~x - y) ~ function(.) abs(.) > 0}
  or
  \preformatted{list("x, y must differ" ~ x - y) ~ function(.) abs(.) > 0}
  if the custom error message \code{"x, y must differ"} is to be used (in
  lieu of an auto-generated error message).
}

Anonymous Predicate Functions

Following the magrittr package, an anonymous (predicate) function of a single argument . can be concisely expressed by enclosing the body of such a function within curly braces { }.

\subsection{Example}{
  The (onsided, global) check formula
  \preformatted{~{. > 0}}
  is equivalent to the check formula \code{~function(.) {. > 0}}
}

Value

firmly

firmly does nothing when there is nothing to do: .f is returned, unaltered, when both .checklist and .warn_missing are empty, or when .f has no named argument and .warn_missing is empty.

Otherwise, \code{firmly} again returns a function that behaves
\emph{identically} to \code{.f}, but also performs input validation:
before a call to \code{.f} is attempted, its inputs are checked, and if
any check fails, an error halts further execution with a message
tabulating every failing check. (If all checks pass, the call to
\code{.f} respects lazy evaluation, as usual.)

\subsection{Subclass of the input-validation error object}{
  The subclass of the error object is \code{.error_class}, unless
  \code{.error_class} is \code{character()}. In the latter case, the
  subclass of the error object is that of the existing error object, if
  \code{.f} is itself a firmly applied function, or it is
  \code{"simpleError"}, otherwise.
}

\subsection{Formal Arguments and Attributes}{
  \code{firmly} preserves the attributes and formal arguments of
  \code{.f} (except that the \code{"class"} attribute gains the component
  \code{"firm_closure"}, unless it already contains it).
}

%checkin%

%checkin% applies the check formula(e) in the list .checks to .f, using firmly. The .warn_missing and .error_class arguments of firmly may be specified as named components of .checks.

loosely

loosely returns .f, unaltered, when .f is not a firmly applied function, or both .keep_check and .keep_warning are TRUE.

Otherwise, \code{loosely} returns the underlying (original) function,
stripped of any input validation checks imposed by \code{firmly}, unless
one of the flags \code{.keep_check}, \code{.keep_warning} is switched on:
if \code{.keep_check}, resp. \code{.keep_warning}, is \code{TRUE},
\code{loosely} retains any existing checks, resp. missing-argument
warnings, of \code{.f}.

is_firm

is_firm returns TRUE if x is a firmly applied function (i.e., has class "firm_closure"), and FALSE, otherwise.

See Also

firmly is enhanced by a number of helper functions:

• To verify that a check formula is syntactically correct, use the predicates is_check_formula, is_checklist.

• To make custom check-formula generators, use localize.

• Pre-made check-formula generators are provided to facilitate argument checks for types, scalar objects, and other common data structures and input assumptions. These functions are prefixed by vld_, for convenient browsing and look-up in editors and IDE's that support name completion.

• To access the components of a firmly applied function, use firm_core, firm_checks, firm_error, firm_args, (or simply print the function to display its components).

Examples

## Not run: 

dlog <- function(x, h) (log(x + h) - log(x)) / h

# Require all arguments to be numeric (auto-generated error message)
dlog_fm <- firmly(dlog, ~is.numeric)
dlog_fm(1, .1)    # [1] 0.9531018
dlog_fm("1", .1)  # Error: "FALSE: is.numeric(x)"

# Require all arguments to be numeric (custom error message)
dlog_fm <- firmly(dlog, "Not numeric" ~ is.numeric)
dlog_fm("1", .1)  # Error: "Not numeric: `x`"

# Alternatively, "globalize" a localized checker (see ?localize, ?globalize)
dlog_fm <- firmly(dlog, globalize(vld_numeric))
dlog_fm("1", .1)  # Error: "Not double/integer: `x`"

# Predicate functions can be specified anonymously or by name
dlog_fm <- firmly(dlog, list(~x, ~x + h, ~abs(h)) ~ function(x) x > 0)
dlog_fm <- firmly(dlog, list(~x, ~x + h, ~abs(h)) ~ {. > 0})
is_positive <- function(x) x > 0
dlog_fm <- firmly(dlog, list(~x, ~x + h, ~abs(h)) ~ is_positive)
dlog_fm(1, 0)  # Error: "FALSE: is_positive(abs(h))"

# Describe checks individually using custom error messages
dlog_fm <-
  firmly(dlog,
         list("x not positive" ~ x, ~x + h, "Division by 0 (=h)" ~ abs(h)) ~
           is_positive)
dlog_fm(-1, 0)
# Errors: "x not positive", "FALSE: is_positive(x + h)", "Division by 0 (=h)"

# Specify checks more succinctly by using a (localized) custom checker
req_positive <- localize("Not positive" ~ is_positive)
dlog_fm <- firmly(dlog, req_positive(~x, ~x + h, ~abs(h)))
dlog_fm(1, 0)  # Error: "Not positive: abs(h)"

# Combine multiple checks
dlog_fm <- firmly(dlog,
                  "Not numeric" ~ is.numeric,
                  list(~x, ~x + h, "Division by 0" ~ abs(h)) ~ {. > 0})
dlog_fm("1", 0)  # Errors: "Not numeric: `x`", check-eval error, "Division by 0"

# Any check can be expressed using isTRUE
err_msg <- "x, h differ in length"
dlog_fm <- firmly(dlog, list(err_msg ~ length(x) - length(h)) ~ {. == 0L})
dlog_fm(1:2, 0:2)  # Error: "x, h differ in length"
dlog_fm <- firmly(dlog, list(err_msg ~ length(x) == length(h)) ~ isTRUE)
dlog_fm(1:2, 0:2)  # Error: "x, h differ in length"

# More succinctly, use vld_true
dlog_fm <- firmly(dlog, vld_true(~length(x) == length(h), ~all(abs(h) > 0)))
dlog_fm(1:2, 0:2)
# Errors: "Not TRUE: length(x) == length(h)", "Not TRUE: all(abs(h) > 0)"

dlog_fm(1:2, 1:2)  # [1] 0.6931472 0.3465736

# loosely recovers the underlying function
identical(loosely(dlog_fm), dlog)  # [1] TRUE

# Use .warn_missing when you want to ensure an argument is explicitly given
# (see vignette("valaddin") for an elaboration of this particular example)
as_POSIXct <- firmly(as.POSIXct, .warn_missing = "tz")
Sys.setenv(TZ = "EST")
as_POSIXct("2017-01-01 03:14:16")  # [1] "2017-01-01 03:14:16 EST"
                                   # Warning: "Argument(s) expected ... `tz`"
as_POSIXct("2017-01-01 03:14:16", tz = "UTC")  # [1] "2017-01-01 03:14:16 UTC"
loosely(as_POSIXct)("2017-01-01 03:14:16")     # [1] "2017-01-01 03:14:16 EST"

# Use firmly to constrain undesirable behavior, e.g., long-running computations
fib <- function(n) {
  if (n <= 1L) return(1L)
  Recall(n - 1) + Recall(n - 2)
}
fib <- firmly(fib, list("`n` capped at 30" ~ ceiling(n)) ~ {. <= 30L})
fib(21)  # [1] 17711 (NB: Validation done only once, not for every recursive call)
fib(31)  # Error: `n` capped at 30

# Apply fib unrestricted
loosely(fib)(31)  # [1] 2178309 (may take several seconds to finish)

# firmly won't force an argument that's not involved in checks
g <- firmly(function(x, y) "Pass", list(~x) ~ is.character)
g(c("a", "b"), stop("Not signaled"))  # [1] "Pass"

# In scripts and packages, it is recommended to use the operator %checkin%
vec_add <- list(
  ~is.numeric,
  list(~length(x) == length(y)) ~ isTRUE,
  .error_class = "inputError"
) %checkin%
  function(x, y) {
    x + y
  }

# Or call firmly with .f explicitly assigned to the function
vec_add2 <- firmly(
  ~is.numeric,
  list(~length(x) == length(y)) ~ isTRUE,
  .f = function(x, y) {
    x + y
  },
  .error_class = "inputError"
)

all.equal(vec_add, vec_add2)  # [1] TRUE

## End(Not run)

valaddin documentation built on Oct. 26, 2023, 1:07 a.m.