Error handling and Assertions

2026-03-05

Writing Code for Easy Debugging

1. Format your code well (styler package is useful)

2. Leave comments

   • describe what the code is supposed to be doing

   • what the input is supposed to be

   • what the output should look like
     …For each functional unit of code

3. Check inputs and verify outputs

   • stopifnot() in base R

   • assertthat package (https://github.com/hadley/assertthat)

     • error signaling, input validation, better error messages

   • checkmate package (https://github.com/mllg/checkmate)

     • fast and concise input validation

assertthat

Compare the error messages:

stopifnot(is.character(1:10))


library(assertthat)
assert_that(is.character(1:10))

Other assertthat functions

function	purpose
is.flag(x)	is x TRUE or FALSE? (a boolean flag)
is.string(x)	is x a length 1 character vector?
has_name(x, nm)	does x have a component named nm?
has_attr(x, attr)	does x have attribute attr?
is.count(x)	is x a single positive integer?
not_empty(x)	are all dimensions of x greater than 0?
noNA(x)	is x free from missing values?
is.dir(path)	is path a directory?
is.writeable(path)	is path writeable?
is.readable(path)	is path readable?
has_extension(path, ext)	does file have the specified extension ext?

Your Turn

Use assertthat functions to check inputs and output for the col_means function, ensuring that they make sense.

col_means <- function(df) {
  numeric <- sapply(df, is.numeric)
  numeric_cols <- df[, numeric]

  data.frame(lapply(numeric_cols, mean))
}

This is not an exhaustive list of checks…

library(assertthat)

col_means <- function(df) {
  assert_that(
    is.data.frame(df), # Check input is a data frame
    not_empty(df) # Check that df has at least one row and column
  )
  numeric <- sapply(df, is.numeric)
  assert_that(length(numeric) > 0) # >= one numeric column
  
  numeric_cols <- df[, numeric]
  # ensure numeric_cols has at least one row and column
  assert_that(length(dim(numeric)) == 2) 
  list_res <- lapply(numeric_cols, mean)
  res <- data.frame(list_res)
  assert_that(
    is.data.frame(res), # output is data frame
    are_equal(ncol(res), length(numeric_cols)) # correct # cols
  )
  
  res
}

checkmate

Package checkmate implements assertions and checks for speedy evalution …

Goals:

• Faster (C backend)

• Better memory handling - no intermediate objects created

• Single call to check e.g. length, missingness, and lower/upper bounds (more concise)

(Preview…) Also interfaces with testthat for unit tests.

checkmate

• assert* : single statement to check type, length, type-specific attrs
  also assert_* if you prefer that syntax

  • Throw an error if the check fails

  • Checked object returned invisibly on success

    • can be used with pipes!

• test* return TRUE or FALSE - best used to combine different checks

• expect* functions to be used in unit tests

• check* functions

  • return error message as a string if the conditions fail

  • return TRUE otherwise

  • used in assert*, test*, expect* functions

Your Turn

Use checkmate functions to check inputs and output for the col_means function, ensuring that they make sense.

col_means <- function(df) {
  numeric <- sapply(df, is.numeric)
  numeric_cols <- df[, numeric]

  data.frame(lapply(numeric_cols, mean))
}

This is not an exhaustive list of checks…

library(checkmate)

col_means <- function(df) {
  assert_data_frame(df, min.rows = 1, min.cols = 1) # Check input is a data frame
  
  numeric <- sapply(df, is.numeric)
  assert_atomic_vector(numeric, min.len = 1) # >= one numeric col
  
  numeric_cols <- df[, numeric]
  assert_data_frame(df, min.rows = 1, min.cols = 1) # numeric_cols is a data frame
  
  list_res <- lapply(numeric_cols, mean)
  res <- data.frame(list_res)
  assert_data_frame(
    res, # output is data frame
    ncols = length(numeric_cols) # correct # cols
  ) 
  
  res
}