Skip to content

expect_error(), expect_warning(), expect_message(), and expect_condition() check that code throws an error, warning, message, or condition with a message that matches regexp, or a class that inherits from class. See below for more details.

In the 3rd edition, these functions match (at most) a single condition. All additional and non-matching (if regexp or class are used) conditions will bubble up outside the expectation. If these additional conditions are important you'll need to catch them with additional expect_message()/expect_warning() calls; if they're unimportant you can ignore with suppressMessages()/suppressWarnings().

It can be tricky to test for a combination of different conditions, such as a message followed by an error. expect_snapshot() is often an easier alternative for these more complex cases.

Usage

expect_error(
  object,
  regexp = NULL,
  class = NULL,
  ...,
  inherit = TRUE,
  info = NULL,
  label = NULL
)

expect_warning(
  object,
  regexp = NULL,
  class = NULL,
  ...,
  inherit = TRUE,
  all = FALSE,
  info = NULL,
  label = NULL
)

expect_message(
  object,
  regexp = NULL,
  class = NULL,
  ...,
  inherit = TRUE,
  all = FALSE,
  info = NULL,
  label = NULL
)

expect_condition(
  object,
  regexp = NULL,
  class = NULL,
  ...,
  inherit = TRUE,
  info = NULL,
  label = NULL
)

Arguments

object

Object to test.

Supports limited unquoting to make it easier to generate readable failures within a function or for loop. See quasi_label for more details.

regexp

Regular expression to test against.

  • A character vector giving a regular expression that must match the error message.

  • If NULL, the default, asserts that there should be an error, but doesn't test for a specific value.

  • If NA, asserts that there should be no errors, but we now recommend using expect_no_error() and friends instead.

Note that you should only use message with errors/warnings/messages that you generate. Avoid tests that rely on the specific text generated by another package since this can easily change. If you do need to test text generated by another package, either protect the test with skip_on_cran() or use expect_snapshot().

class

Instead of supplying a regular expression, you can also supply a class name. This is useful for "classed" conditions.

...

Arguments passed on to expect_match

fixed

If TRUE, treats regexp as a string to be matched exactly (not a regular expressions). Overrides perl.

perl

logical. Should Perl-compatible regexps be used?

inherit

Whether to match regexp and class across the ancestry of chained errors.

info

Extra information to be included in the message. This argument is soft-deprecated and should not be used in new code. Instead see alternatives in quasi_label.

label

Used to customise failure messages. For expert use only.

all

DEPRECATED If you need to test multiple warnings/messages you now need to use multiple calls to expect_message()/ expect_warning()

Value

If regexp = NA, the value of the first argument; otherwise the captured condition.

Testing message vs class

When checking that code generates an error, it's important to check that the error is the one you expect. There are two ways to do this. The first way is the simplest: you just provide a regexp that match some fragment of the error message. This is easy, but fragile, because the test will fail if the error message changes (even if its the same error).

A more robust way is to test for the class of the error, if it has one. You can learn more about custom conditions at https://adv-r.hadley.nz/conditions.html#custom-conditions, but in short, errors are S3 classes and you can generate a custom class and check for it using class instead of regexp.

If you are using expect_error() to check that an error message is formatted in such a way that it makes sense to a human, we recommend using expect_snapshot() instead.

Examples

# Errors ------------------------------------------------------------------
f <- function() stop("My error!")
expect_error(f())
expect_error(f(), "My error!")

# You can use the arguments of grepl to control the matching
expect_error(f(), "my error!", ignore.case = TRUE)

# Note that `expect_error()` returns the error object so you can test
# its components if needed
err <- expect_error(rlang::abort("a", n = 10))
expect_equal(err$n, 10)

# Warnings ------------------------------------------------------------------
f <- function(x) {
  if (x < 0) {
    warning("*x* is already negative")
    return(x)
  }
  -x
}
expect_warning(f(-1))
expect_warning(f(-1), "already negative")
expect_warning(f(1), NA)

# To test message and output, store results to a variable
expect_warning(out <- f(-1), "already negative")
expect_equal(out, -1)

# Messages ------------------------------------------------------------------
f <- function(x) {
  if (x < 0) {
    message("*x* is already negative")
    return(x)
  }

  -x
}
expect_message(f(-1))
expect_message(f(-1), "already negative")
expect_message(f(1), NA)