Mocking is catching

[This article was first published on Posts on R-hub blog, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

When writing unit tests for a package, you might find yourself wondering about how to best test the behaviour of your package

  • when the data it’s supposed to munge has this or that quirk,

  • when the operating system is Windows,

  • when a package enhancing its functionality is not there,

  • when a web API returns an error;

or you might even wonder how to test at least part of that package of yours that calls a web API or local database… without accessing the web API or local database during testing.

In some of these cases, the programming concept you’re after is mocking, i.e. making a function act as if something were a certain way! In this blog post we shall offer a round-up of resources around mocking, or not mocking, when unit testing an R package.

Please keep reading, do not flee to Twitter! ???? (The talented Sharla did end up using mocking for her package!)

Packages for mocking

General mocking

Nowadays, when using testthat for testing, the recommended tool for mocking is the mockery package, not testthat’s own with_mock() function. To read how they differ in their implementation of mocking, refer to this issue and that section of mockery README. In brief, with mockery you can stub (i.e. replace) a function in a given environment e.g. the environment of a function. Let’s create a small toy example to illustrate that.

# a function that says encoding is a pain
# when the OS is Windows
is_encoding_a_pain <- function(){
  if ([["sysname"]] == "Windows"){
  } else {

# The post was rendered on Ubuntu[["sysname"]]

## [1] "Linux"

# So, is encoding a pain?

## [1] "no"

# stub/replace in is_encoding_a_pain()
# with a mock that answers "Windows"
mockery::stub(where = is_encoding_a_pain,
              what = "", 
              how = c(sysname = "Windows"))

# Different output

## [1] "YES"

# NOT changed[["sysname"]]

## [1] "Linux"

Let’s also look at a real life example, from keyring tests:

test_that("auto windows", {
  mockery::stub(default_backend_auto, "", c(sysname = "Windows"))
  expect_equal(default_backend_auto(), backend_wincred)

What happens after the call to mockery::stub() is that inside the test, when default_backend_auto() is called, it won’t use the actual but instead a mock that returns c(sysname = "Windows") so the test can assess what default_backend_auto() returns on Windows… without the test being run on a Windows machine. ????

Instead of directly defining the return value as is the case in this example, one could stub the function with a function, as seen in one of the tests for the remotes package.

To find more examples of how to use mockery in tests, you can use GitHub search in combination with R-hub’s CRAN source code mirror:

Web mocking

In the case of a package doing HTTP requests, you might want to test what happens when an error code is received for instance. To do that, you can use either httptest or webmockr (compatible with both httr and crul).

Temporarily modify the global state

To test what happens when, say, an environment variable has a particular value, one can set it temporarily within a test using the withr package. You could argue it’s not technically mocking, but it’s an useful trick. You can see it in action in keyring’s tests.

To mock or… not to mock

Sometimes, you might not need mocking and can resort to an alternative approach instead, using the real thing/situation. You could say it’s a less “unit” approach and requires more work.

Fake input data

For say a plotting or modelling library, you can tailor-make data. Comparing approaches or packages for creating fake data are beyond the scope of this post, so let’s just name a few packages:

Stored data from a web API / a database

As explained in this discussion about testing web API packages, when testing a package accessing and munging web data you might want to separate testing of the data access and of the data munging, on the one hand because failures will be easier to trace back to a problem in the web API vs. your code, on the other hand to be able to selectively turn off some tests based on internet connection, the presence of an API key, etc. Storing and replaying HTTP requests is supported by:

What about applying the same idea to packages using a database connection?

Different operating systems

Say you want to be sure your packages builds correctly on another operating system… you can use R-hub package builder ???? or maybe a continuous integration service.

Different system configurations or libraries

Regarding the case where you want to test your package when a suggested dependency is or is not installed, you can use the configuration script of a continuous integration service to have at least one build without that dependency:


In this post we offered a round-up of resources around mocking when unit testing R packages, as well as around not mocking. To learn about more packages for testing your package, refer to the list published on Locke Data’s blog. Now, what if you’re not sure about the best approach for that quirky thing you want to test, mocking or not mocking, and how exactly? Well, you can fall back on two methods: Reading the source code of other packages, and Asking for help! Good luck! ????

To leave a comment for the author, please follow the link and comment on their blog: Posts on R-hub blog. offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)