Functional Input Validation.

Development has moved to rong

valaddin

Dealing with invalid function inputs is a chronic pain for R users, given R’s weakly typed nature. valaddin provides pain relief—a lightweight R package that enables you to transform an existing function into a function with input validation checks, in situ, in a manner suitable for both programmatic use and interactive sessions.

Installation

Install from CRAN

install.packages("valaddin")

or get the development version from GitHub using the devtools package

# install.packages("devtools")
devtools::install_github("egnha/valaddin", ref = "dev", build_vignettes = TRUE)

Why use valaddin

Fail fast—save time, spare confusion

You can be more confident your function works correctly, when you know its arguments are well-behaved. But when they aren’t, its better to stop immediately and bring them into line, than to let them pass and wreak havoc, exposing yourself to breakages or, worse, silently incorrect results. Validating the inputs of your functions is good defensive programming practice.

Suppose you have a function secant()

secant <- function(f, x, dx) (f(x + dx) - f(x)) / dx

and you want to ensure that the user (or some code) supplies numerical inputs for x and dx. Typically, you’d rewrite secant() so that it stops if this condition is violated:

secant_numeric <- function(f, x, dx) {
  stopifnot(is.numeric(x), is.numeric(dx))
  secant(f, x, dx)
}

secant_numeric(log, 1, .1)
#> [1] 0.9531018

secant_numeric(log, "1", ".1")
#> Error in secant_numeric(log, "1", ".1"): is.numeric(x) is not TRUE

The standard approach in R is problematic

While this works, it’s not ideal, even in this simple situation, because

• it’s inconvenient for interactive use at the console: you have to declare a new function, and give it a new name (or copy-paste the original function body)

• it doesn’t catch all errors, only the first that occurs among the checks

• you’re back to square one, if you later realize you need additional checks, or want to skip them altogether.

valaddin rectifies these shortcomings

valaddin provides a function firmly() that takes care of input validation by transforming the existing function, instead of forcing you to write a new one. It also helps you by reporting every failing check.

library(valaddin)

# Check that `x` and `dx` are numeric
secant <- firmly(secant, list(~x, ~dx) ~ is.numeric)

secant(log, 1, .1)
#> [1] 0.9531018

secant(log, "1", ".1")
#> Error: secant(f = log, x = "1", dx = ".1")
#> 1) FALSE: is.numeric(x)
#> 2) FALSE: is.numeric(dx)

To add additional checks, just apply the same procedure again:

secant <- firmly(secant, list(~x, ~dx) ~ {length(.) == 1L})

secant(log, "1", c(.1, .01))
#> Error: secant(f = log, x = "1", dx = c(0.1, 0.01))
#> 1) FALSE: is.numeric(x)
#> 2) FALSE: (function(.) {length(.) == 1L})(dx)

Or, alternatively, all in one go:

secant <- loosely(secant)  # Retrieves the original function
secant <- firmly(secant, list(~x, ~dx) ~ {is.numeric(.) && length(.) == 1L})

secant(log, 1, .1)
#> [1] 0.9531018

secant(log, "1", c(.1, .01))
#> Error: secant(f = log, x = "1", dx = c(0.1, 0.01))
#> 1) FALSE: (function(.) {is.numeric(.) && length(.) == 1L})(x)
#> 2) FALSE: (function(.) {is.numeric(.) && length(.) == 1L})(dx)

Check anything using a simple, consistent syntax

firmly() uses a simple formula syntax to specify arbitrary checks—not just type checks. Every check is a formula of the form <where to check> ~ <what to check>. The “what” part on the right is a function that does a check, while the (form of the) “where” part on the left indicates where to apply the check—at which arguments or expressions thereof.

valaddin provides a number of conveniences to make checks for firmly() informative and easy to specify.

Use custom error messages

Use a custom error message to clarify the purpose of a check:

bc <- function(x, y) c(x, y, 1 - x - y)

# Check that `y` is positive
bc_uhp <- firmly(bc, list("(x, y) not in upper half-plane" ~ y) ~ {. > 0})

bc_uhp(.5, .2)
#> [1] 0.5 0.2 0.3

bc_uhp(.5, -.2)
#> Error: bc_uhp(x = 0.5, y = -0.2)
#> (x, y) not in upper half-plane

Easily apply a check to all arguments

Leave the left-hand side of a check formula blank to apply it to all arguments:

bc_num <- firmly(bc, ~is.numeric)

bc_num(.5, ".2")
#> Error: bc_num(x = 0.5, y = ".2")
#> FALSE: is.numeric(y)

bc_num(".5", ".2")
#> Error: bc_num(x = ".5", y = ".2")
#> 1) FALSE: is.numeric(x)
#> 2) FALSE: is.numeric(y)

Or fill in a custom error message:

bc_num <- firmly(bc, "Not numeric" ~ is.numeric)

bc_num(.5, ".2")
#> Error: bc_num(x = 0.5, y = ".2")
#> Not numeric: `y`

Check conditions with multi-argument dependencies

Use the isTRUE() predicate to implement checks depending on multiple arguments or, equivalently, the check maker vld_true():

in_triangle <- function(x, y) {x >= 0 && y >= 0 && 1 - x - y >= 0}
outside <- "(x, y) not in triangle"

bc_tri <- firmly(bc, list(outside ~ in_triangle(x, y)) ~ isTRUE)

# Or more concisely:
bc_tri <- firmly(bc, vld_true(outside ~ in_triangle(x, y)))

# Or more concisely still, by relying on an auto-generated error message:
# bc_tri <- firmly(bc, vld_true(~in_triangle(x, y)))

bc_tri(.5, .2)
#> [1] 0.5 0.2 0.3

bc_tri(.5, .6)
#> Error: bc_tri(x = 0.5, y = 0.6)
#> (x, y) not in triangle

Make your code more intelligible

To make your functions more intelligible, declare your input assumptions and move the core logic to the fore. You can do this using firmly(), in several ways:

• Precede the function header with input checks, by explicitly assigning the function to firmly()’s .f argument:

  bc <- firmly(
    ~is.numeric,
    ~{length(.) == 1L},
    vld_true(outside ~ in_triangle(x, y)),
    .f = function(x, y) {
      c(x, y, 1 - x - y)
    }
  )
  
  bc(.5, .2)
  #> [1] 0.5 0.2 0.3
  
  bc(.5, c(.2, .1))
  #> Error: bc(x = 0.5, y = c(0.2, 0.1))
  #> 1) FALSE: (function(.) {length(.) == 1L})(y)
  #> 2) Error evaluating check (function (x) is.logical(x) && length(x) == 1L && !is.na(x) && x)(in_triangle(x, y)): 'length = 2' in coercion to 'logical(1)'
  
  bc(".5", 1)
  #> Error: bc(x = ".5", y = 1)
  #> 1) FALSE: is.numeric(x)
  #> 2) (x, y) not in triangle
  

• Use the magrittr%>% operator to deliver input checks, by capturing them as a list with firmly()’s .checklist argument:

  library(magrittr)
  
  bc2 <- list(
    ~is.numeric,
    ~{length(.) == 1L},
    vld_true(outside ~ in_triangle(x, y))
  ) %>%
    firmly(function(x, y) {
      c(x, y, 1 - x - y)
    },
    .checklist = .)
  

• Better yet, use the %checkin% operator:

  bc3 <- list(
    ~is.numeric,
    ~{length(.) == 1L},
    vld_true(outside ~ in_triangle(x, y))
  ) %checkin%
    function(x, y) {
      c(x, y, 1 - x - y)
    }
  

Learn more

See the package documentation ?firmly, help(p = valaddin) for detailed information about firmly() and its companion functions, and the vignette for an overview of use cases.

Related packages

• assertive, assertthat, and checkmate provide handy collections of predicate functions that you can use in conjunction with firmly().

• argufy takes a different approach to input validation, using roxygen comments to specify checks.

• ensurer and assertr provide a means of validating function values. Additionally, ensurer provides an experimental replacement for function() that builds functions with type-validated arguments.

• typeCheck, together with Types for R, enables the creation of functions with type-validated arguments by means of special type annotations. This approach is orthogonal to that of valaddin: whereas valaddin specifies input checks as predicate functions with scope, typeCheck specifies input checks as arguments with type.

License

MIT Copyright © 2016–2023 Eugene Ha.