With version 3.0.0 of testthat
, mocking capabilities provided by
testthat::with_mock()
and
testthat::local_mock()
have been deprecated under edition 3. This leaves implementation of
function mocking for unit testing to third-party packages, of which two
have been published on CRAN:
mockery
and
mockr
. While all currently
available mocking implementations have their limitations, what sets
mockthat
apart from mockery
and mockr
is coping with S3 dispatch
(see example below).
You can install the development version of mockthat
from
GitHub with:
# install.packages("devtools")
devtools::install_github("nbenn/mockthat")
A release version will be submitted to CRAN shortly.
Mocking in the context of unit testing refers to temporarily replacing a piece of functionality (that might be part of the package being tested or potentially even part of a downstream dependency) in order to cope with limited infrastructure in testing environments (for example absence of a live Internet connection).
For a function download_data()
implemented as
download_data <- function(url) {
curl::curl_fetch_memory(url)
}
we do not want to have to rely on a live internet connection for writing
a unit test. With help of mockthat
, we can substitute
curl::curl_fetch_memory()
with a stub that simply returns a constant.
library(mockthat)
url <- "https://eu.httpbin.org/get?foo=123"
with_mock(
`curl::curl_fetch_memory` = function(...) '["mocked request"]',
download_data(url)
)
#> [1] "[\"mocked request\"]"
As mentioned above, the main point of differentiation of mockthat
over
the other available packages mockery
and mockr
is stubbing out
functions in the context of S3 dispatch. Assuming the following set-up,
gen <- function(x) UseMethod("gen")
met <- function(x) foo(x)
foo <- function(x) stop("foo")
.S3method("gen", "cls", met)
x <- structure(123, class = "cls")
gen(x)
#> Error in foo(x): foo
mockthat::with_mock()
can be used to catch the call to foo()
and
therefore prevent the error from being thrown.
mockthat::with_mock(
foo = function(x) "bar",
met(x)
)
#> [1] "bar"
This is not possible with the current implementation of
mockr::with_mock()
.
mockr::with_mock(
foo = function(x) "bar",
met(x)
)
#> Warning: Replacing functions in evaluation environment: `foo()`
#> Warning: The code passed to `with_mock()` must be a braced expression to get
#> accurate file-line information for failures.
#> [1] "bar"
And with the current API of
mockery::stub()
it is
unclear how the depth
argument should be chosen, as the function
gen()
does not contain a call to met()
. Trying a range of sensible
values does not yield the desired result.
for (depth in seq_len(3L)) {
mockery::stub(gen, "foo", "bar", depth = depth)
tryCatch(met(x), error = function(e) message("depth ", depth, ": nope"))
}
#> depth 1: nope
#> depth 2: nope
#> depth 3: nope
Borrowing from mockery
, mockthat
also allows for creating mock
objects (with class attribute mock_fun
), which allow capture of the
call for later examination.
mk <- mock("mocked request")
dl <- function(url) curl::curl(url)
with_mock(`curl::curl` = mk, dl(url))
#> [1] "mocked request"
mock_call(mk)
#> curl::curl(url = url)
mock_args(mk)
#> $url
#> [1] "https://eu.httpbin.org/get?foo=123"
#>
#> $open
#> [1] ""
#>
#> $handle
#> <curl handle> (empty)
In addition to with_mock()
, mockthat
also offers a local_mock()
function, again, mimicking the deprecated testthat
function, which
keeps the mocks in place for the life-time of the environment passed as
local_env
argument (or if called from the global environment, until
withr::deferred_run()
is executed). Mock objects as shown above are
created (and returned invisibly) for all non-function objects passed as
...
.
tmp <- new.env()
mk <- local_mock(`curl::curl` = "mocked request", local_env = tmp)
dl(url)
#> [1] "mocked request"
mock_arg(mk, "url")
#> [1] "https://eu.httpbin.org/get?foo=123"