[How to] Write a purrr-like adverb

Create your own safely, compose and friends!

What is an adverb

If you read carefully the purrr documentation, you’ll find this simple explanation :

Adverbs modify the action of a function; taking a function as input and returning a function with modified action as output.

In other words, adverbs take a function, and return this function modified. Yes, just as an adverb modifies a verb. So if you do :

library(purrr)
safe_log <- safely(log)

The returned object is another function that you can use just as a regular one.

class(safe_log)

## [1] "function"

safe_log("a")

## $result
## NULL
## 
## $error
## <simpleError in log(x = x, base = base): argument non numérique pour une fonction mathématique>

In computer science, these adverbs are what is called “high-order functions”.

How to write your own?

I’ve been playing with adverbs in {attempt}, notably through these adverbs :

library(attempt)

# Silently only return the errors, and nothing if the function succeeds
silent_log <- silently(log)
silent_log(1)

# Surely make a function always work, without stopping the process
sure_log <- surely(log)
sure_log(1)

## [1] 0

sure_log("a")

# with_message and with_warning
as_num_msg <- with_message(as.numeric, msg = "We're performing a numeric conversion")
as_num_warn <- with_warning(as.numeric, msg = "We're performing a numeric conversion")
as_num_msg("1")

## We're performing a numeric conversion

## [1] 1

as_num_warn("1")

## Warning in as_num_warn("1"): We're performing a numeric conversion

## [1] 1

So, how to implement this kind of behavior? Let’s take a simple example with sleepy, also shared on Twitter.

sleepy <- function(fun, sleep){
  function(...){
    Sys.sleep(sleep)
    fun(...)
  }
}

sleep_print <- sleepy(Sys.time, 5)
class(sleep_print)

## [1] "function"

# Let's try
Sys.time()

## [1] "2018-09-03 21:17:05 CEST"

sleep_print()

## [1] "2018-09-03 21:17:10 CEST"

Let’s decompose what we’ve got here.

First of all, the function should return another function, so we need to start with :

talky <- function(){
  function(){
    
  }
}

What this function will take as a first argument is another function, that will be executed when our future new function is called.

So let’s do this:

talky <- function(fun){
  function(){
    fun()
  }
}

Because you know, with R referential transparency, you can create a variable that is a function:

plop <- mean
plop(1:10)

## [1] 5.5

This simple skeleton will work if we take a function without any args:

sys_time <- talky(Sys.time)
sys_time()

## [1] "2018-09-03 21:17:10 CEST"

But hey, this is not what we want: we need this new function to be able to take arguments. So let’s use our friend ....

talky <- function(fun){
  function(...){
    fun(...)
  }
}

Now, our new adverb creates a function that can take arguments. But as you’ve notice, this is still not really an adverb: we need to modify something. Now you’re only limited by your imagination ;)

# Print the time
talky <- function(fun){
  function(...){
    print(Sys.time())
    fun(...)
  }
}
talky_sqrt <- talky(sqrt)
talky_sqrt(10)

## [1] "2018-09-03 21:17:10 CEST"

## [1] 3.162278

# Or with a kind message ? 
talky <- function(fun, mess){
  function(...){
    message(mess)
    fun(...)
  }
}
talky_sqrt<- talky(fun = sqrt, mess = "Hey there! You Rock!")
talky_sqrt(1)

## Hey there! You Rock!

## [1] 1

# Run it or not ?
maybe <- function(fun){
  function(...){
    num <- sample(1:100, 1)
    if (num > 50) {
      fun(...)
    }
  }
}
maybe_sqrt <- maybe(fun = sqrt)
maybe_sqrt(1)
maybe_sqrt(1)

## [1] 1

maybe_sqrt(1)

## [1] 1

# Create a log file of a function 
log_calls <- function(fun, file){
  function(...){
    write(as.character(Sys.time()), file, append = TRUE, sep = "\n")
    fun(...)
  }
}
log_sqrt <- log_calls(sqrt, file = "logs")
log_sqrt(10)

## [1] 3.162278

log_sqrt(13)

## [1] 3.605551

readLines("logs")

## [1] "2018-09-03 21:17:10" "2018-09-03 21:17:10"