Tools for Defensive Programming

Tools for defensive programming, inspired by 'purrr' mappers and based on 'rlang'.'attempt' extends and facilitates defensive programming by providing a consistent grammar, and provides a set of easy to use functions for common tests and conditions. 'attempt' only depends on 'rlang', and focuses on speed, so it can be easily integrated in other functions and used in data analysis.

A Friendlier Condition Handler for R, inspired by {purrr} mappers and based on {rlang}.

{attempt} is designed to handle the cases when something / someone attempts to do something it shouldn’t.

For example :

• an attempt to run a log("a") (error)
• an attempt to connect to a web API without an internet connexion (error)
• an attempt to paste() "good morning" and iris (message/warning)
• …

{attempt} provides several condition handlers, from try catch to simple message printing.

{attempt} only depends on {rlang}, and every function is design to be fast, making it easy to implement in other functions and packages.

Install

From CRAN:

install.packages("attempt")

The dev version:

devtools::install_github("ColinFay/attempt")

Reference

library(attempt)

attempt

attempt is a wrapper around base try that allows you to insert a custom messsage on error.

attempt(log("a"))
# Error: argument non numérique pour une fonction mathématique
attempt(log("a"), msg = "Nop !")
# Error: Nop !

You can make it verbose (i.e. returning the expression):

attempt(log("a"), msg = "Nop !", verbose = TRUE)
# Error in log("a"): Nop !

Of course the result is returned if there is one:

attempt(log(1), msg = "Nop !", verbose = TRUE)
# [1] 0

As with try, the result cant be saved as an error object:

a <- attempt(log("a"), msg = "Nop !", verbose = TRUE)
a
# [1] "Error in log(\"a\"): Nop !\n"
# attr(,"class")
# [1] "try-error"
# attr(,"condition")
# <simpleError in log("a"): Nop !>

silent_attempt

silent_attempt is a wrapper around silently (see further down for more info) and attempt. It attempts to run the expr, stays silent if the expression succeeds, and returns error or warnings if any.

silent_attempt(log("a"))
# Error: argument non numérique pour une fonction mathématique
silent_attempt(log(1))

try catch

You can write a try catch with these params :

• expr the expression to be evaluated
• .e a mapper or a function evaluated when an error occurs
• .w a mapper or a function evaluated when a warning occurs
• .f a mapper or an expression which is always evaluated before returning or exiting

In .e and .f, the .x refers to the error / warning object.

With mappers

try_catch(expr = log("a"), 
          .e = ~ paste0("There is an error: ", .x), 
          .w = ~ paste0("This is a warning: ", .x))
#[1] "There is an error: Error in log(\"a\"): argument non numérique pour une fonction mathématique\n"
 
try_catch(log("a"), 
          .e = ~ stop(.x), 
          .w = ~ warning(.x))
# Error in log("a") : argument non numérique pour une fonction mathématique
 
try_catch(matrix(1:3, nrow= 2), 
          .e = ~ print(.x), 
          .w = ~ print(.x))
#<simpleWarning in matrix(1:3, nrow = 2): la longueur des données [3] n'est pas un diviseur ni un multiple du nombre de lignes [2]>
 
try_catch(expr = 2 + 2 , 
          .f = ~ print("Using R for addition... ok I'm out!"))
# [1] "Using R for addition... ok I'm out!"
# [1] 4

As usual, the handlers are set only if you call them :

try_catch(matrix(1:3, nrow = 2), .e = ~ print("error"))
#      [,1] [,2]
# [1,]    1    3
# [2,]    2    1
# Warning message:
# In matrix(1:3, nrow = 2) :
#   la longueur des données [3] n'est pas un diviseur ni un multiple du nombre de lignes [2]

try_catch(matrix(1:3, nrow = 2), .w = ~ print("warning"))
# [1] "warning"

Traditionnal way

{attempt} is flexible in how you can specify your arguments.

You can, as you do with {base} tryCatch, use a plain old function:

try_catch(log("a"), 
          .e = function(e){
            print(paste0("There is an error: ", e))
            print("Ok, let's save this")
            time <- Sys.time()
            a <- paste("+ At",time, ", \nError:",e)
            # write(a, "log.txt", append = TRUE) # commented to prevent log.txt creation 
            print(paste("log saved on log.txt at", time))
            print("let's move on now")
          })
 
# [1] "There is an error: Error in log(\"a\"): argument non numérique pour une fonction mathématique\n"
# [1] "Ok, let's save this"
# [1] "log saved on log.txt at 2018-01-30 16:59:13"
# [1] "let's move on now"

You can even mix both:

try_catch(log("a"), 
          .e = function(e){
            paste0("There is an error: ", e)
          },
          .f = ~ print("I'm not sure you can do that pal !"))
# [1] "I'm not sure you can do that pal !"
# [1] "There is an error: Error in log(\"a\"): argument non numérique pour une fonction mathématique\n"
 
try_catch(log("a"), 
          .e = ~ paste0("There is an error: ", .x),
          .f = function() print("I'm not sure you can do that pal !"))
# [1] "I'm not sure you can do that pal !"
# [1] "There is an error: Error in log(\"a\"): argument non numérique pour une fonction mathématique\n"

try_catch_df

try_catch_df returns a tibble with the call, the error message if any, the warning message if any, and the value of the evaluated expression or “error”. The values will always be contained in a list-column.

res_log <- try_catch_df(log("a"))
res_log
#>       call                                                 error warning
#> 1 log("a") argument non numérique pour une fonction mathématique      NA
#>   value
#> 1 error
res_log$value
#> [[1]]
#> [1] "error"
 
res_matrix <- try_catch_df(matrix(1:3, nrow = 2))
res_matrix
#>                    call error
#> 1 matrix(1:3, nrow = 2)    NA
#>                                                                                    warning
#> 1 la longueur des données [3] n'est pas un diviseur ni un multiple du nombre de lignes [2]
#>        value
#> 1 1, 2, 3, 1
res_matrix$value
#> [[1]]
#>      [,1] [,2]
#> [1,]    1    3
#> [2,]    2    1
 
res_success <- try_catch_df(log(1))
res_success
#>     call error warning value
#> 1 log(1)    NA      NA     0
res_success$value
#> [[1]]
#> [1] 0

map try_catch

map_try_catch and map_try_catch_df allow you to map on a list of arguments l, to be evaluated by the function in fun.

map_try_catch(l = list(1, 3, "a"), fun = log, .e = ~ .x)
#> [[1]]
#> [1] 0
#> 
#> [[2]]
#> [1] 1.098612
#> 
#> [[3]]
#> <simpleError in .Primitive("log")("a"): argument non numérique pour une fonction mathématique>
 
map_try_catch_df(list(1,3,"a"), log)
#>                     call
#> 1   .Primitive("log")(1)
#> 2   .Primitive("log")(3)
#> 3 .Primitive("log")("a")
#>                                                   error warning    value
#> 1                                                  <NA>      NA        0
#> 2                                                  <NA>      NA 1.098612
#> 3 argument non numérique pour une fonction mathématique      NA    error

Adverbs

Adverbs take a function and return a modified function.

silently

silently transforms a function so that when you call this new function, it returns nothing unless there is an error or a warning (contrary to attempt that returns the result). In a sense, the new function stay silent unless error or warning.

silent_log <- silently(log)
silent_log(1)
silent_log("a")
# Error in .f(...) : argument non numérique pour une fonction mathématique

With silently, the result is never returned.

silent_matrix <- silently(matrix)
silent_matrix(1:3, 2)
#Warning message:
#In .f(...) :
#  la longueur des données [3] n'est pas un diviseur ni un multiple du nombre de lignes [2]

surely

surely transforms a function so that when you call this new function, it calls attempt() - i.e. in the code below, calling sure_log(1) is the same as calling attempt(log(1)). In a sense, you’re sure this new function will always work.

sure_log <- surely(log)
sure_log(1)
# [1] 0
sure_log("a")
# Error: argument non numérique pour une fonction mathématique

with_message and with_warning

These two functions take a function, and add a warning or a message to it.

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

without_message and withou_warning

These two functions do the opposite, as they remove warnings and messages:

matrix(1:3, ncol = 2)
no_warning_matrix <- without_warning(matrix)
no_warning_matrix(1:3, ncol = 2)

if_ conditions

if_none, if_any and if_all test the elements of the list.

if_all(1:10, ~ .x < 11, ~ return(letters[1:10]))
#>  [1] "a" "b" "c" "d" "e" "f" "g" "h" "i" "j"
 
if_any(1:10, is.numeric, ~ "Yay!")
#> [1] "Yay!"
 
if_none(1:10, is.character, ~ rnorm(10))
#>  [1]  0.5082694 -0.4049499 -2.1738221  0.4168055  0.7928220  0.4901836
#>  [7] -1.7092720  0.0174628  1.3408672  0.1922164

The defaut for all .p is isTRUE. So you can:

a <- c(FALSE, TRUE, TRUE, TRUE)
 
if_any(a, .f = ~ "nop!")
#> [1] "nop!"

if_then performs a simple “if this then do that”:

if_then(1, is.numeric, ~ "nop!")
#> [1] "nop!"

if_not runs .f if .p(.x) is not TRUE :

if_not(.x = 1, .p = is.character, ~ ".x is not a character")
#> [1] ".x is not a character"

And if_else is a wrapper around base::ifelse().

If you want these function to return a value, you need to wrap these values into a mapper / a function. E.g, to return a vector, you’ll need to write if_then(1, is.numeric, ~ "Yay").

a <- if_else(1, is.numeric, ~ "Yay", ~ "Nay")
a
#> [1] "Yay"

warnings and messages

The stop_if, warn_if and message_if are easy to use functions that send an error, a warning or a message if a condition is met. Each function has its counterpart with _not that returns a message if the condition is not met.

stop_if_not is quite the same as assert_that from the {assertthat} package, except that it can takes mappers. It is not the same as base stopifnot(), as it doesn’t take a list of expression.

These functions are also flexible as you can pass base predicates (is.numeric, is.character…), a custom predicate built with mappers, or even your own predicate function.

You can either choose a custom message or just let the built-in messages be printed:

x <- 12
# Stop if .x is numeric
stop_if(.x = x, 
        .p = is.numeric)
#> Error: Test `is.numeric` on `x` returned an error.
 
y <- "20"
# stop if .x is not numeric
stop_if_not(.x = y, 
            .p = is.numeric, 
            msg = "y should be numeric")
#> Error: y should be numeric
a  <- "this is not numeric"
# Warn if .x is charcter
warn_if(.x = a, 
        .p = is.character)
#> Warning: Test `is.character` on `a` returned a warning.
 
b  <- 20
# Warn if .x is not equal to 10
warn_if_not(.x = b, 
        .p = ~ .x == 10 , 
        msg = "b should be 10")
#> Warning: b should be 10
 
c <- "a"
# Message if c is a character
message_if(.x = c, 
           .p = is.character, 
           msg = "You entered a character element")
#> You entered a character element
 
# Build more complex predicates
d <- 100
message_if(.x = d, 
           .p = ~ sqrt(.x) < 42, 
           msg = "The square root of your element must be more than 42")
#> The square root of your element must be more than 42
 
# Or, if you're kind of old school, you can still pass classic functions
 
e <- 30
message_if(.x = e, 
           .p = function(vec){
             return(sqrt(vec) < 42)
           }, 
           msg = "The square root of your element must be more than 42")
#> The square root of your element must be more than 42

If you need to call a function that takes no argument at .p (like curl::has_internet()), use this function as .x.

stop_if(.x = curl::has_internet(), msg = "You shouldn't have internet to do that")
#> Error: You shouldn't have internet to do that
 
warn_if(.x = curl::has_internet(), 
            msg = "You shouldn't have internet to do that")
#> Warning: You shouldn't have internet to do that
 
message_if(.x = curl::has_internet(), 
            msg = "Huray, you have internet \\o/")
#> Huray, you have internet \o/

If you don’t specify a .p, the default test is isTRUE.

a <- is.na(airquality$Ozone)
message_if_any(a, msg = "NA found")
#> NA found

In function

That can come really handy inside a function :

my_fun <- function(x){
  stop_if_not(.x = curl::has_internet(), 
              msg = "You should have internet to do that")
  warn_if_not(x, 
          is.character, 
          msg =  "x is not a character vector. The output may not be what you're expecting.")
  paste(x, "is the value.")
}
 
my_fun(head(iris))
#> Warning: x is not a character vector. The output may not be what you're
#> expecting.
#> [1] "c(5.1, 4.9, 4.7, 4.6, 5, 5.4) is the value."  
#> [2] "c(3.5, 3, 3.2, 3.1, 3.6, 3.9) is the value."  
#> [3] "c(1.4, 1.4, 1.3, 1.5, 1.4, 1.7) is the value."
#> [4] "c(0.2, 0.2, 0.2, 0.2, 0.2, 0.4) is the value."
#> [5] "c(1, 1, 1, 1, 1, 1) is the value."

none, all, any

stop_if, warn_if and message_if all have complementary tests with _all, _any and _none, which combine the if_* and the warn_*, stop_* and message_* seen before. They take a list as first argument, and a predicate. They test if any, all or none of the elements validate the predicate.

stop_if_any(iris, is.factor, msg = "Factors here. This might be due to stringsAsFactors.")
#> Error: Factors here. This might be due to stringsAsFactors.
 
warn_if_none(1:10, ~ .x < 0, msg = "You need to have at least one number under zero.")
#> Warning: You need to have at least one number under zero.
 
message_if_all(1:100, is.numeric, msg = "That makes a lot of numbers.")
#> That makes a lot of numbers.

Misc

Acknowledgments

Thanks to Romain for the name suggestion.

Contact

Questions and feedbacks welcome!

You want to contribute ? Open a PR :) If you encounter a bug or want to suggest an enhancement, please open an issue.

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.