For R packages that have Shiny applications, there are generally two ways that the applications will be present in the package. The first is to have an app.R or ui.R/server.R in a subdirectory of inst/. The second way is to have a function which returns a Shiny app object.
See https://github.com/rstudio/shinytestPackageExample for an example package. This document closely mirrors the content of that package: it explains how to set up tests for both types of applications, and then has a section that describes some setup tasks that are common to both types.
An application could live in a subdirectory of inst/, as shown below:
/
├── DESCRIPTION
├── NAMESPACE
├── R
├── inst
│ └── appdir
│ ├── app.R
│ └── tests/
│ ├── mytest.R
│ └── mytest-expected
│ ├── 001.json
│ └── 001.png
└── tests
├── testthat
│ └── test-app-file.R
└── testthat.RIn this case, you can run recordTest() and testApp() as normal. After you create and run the tests, there will be a tests/ subdirectory in the application directory that stores the test scripts and results.
Assuming you are using testthat for automated tests, you would create a test driver script in tests/testthat/. In this example, it’s named test-app-file.R and contains the following:
context("app-file")
# This file is for testing the applications in the inst/ directory.
library(shinytest)
test_that("sampleapp works", {
# Don't run these tests on the CRAN build servers
skip_on_cran()
# Use compareImages=FALSE because the expected image screenshots were created
# on a Mac, and they will differ from screenshots taken on the CI platform,
# which runs on Linux.
appdir <- system.file(package = "shinytestPackageExample", "sampleapp")
expect_pass(testApp(appdir, compareImages = FALSE))
})For more information about this script, see the Test driver script notes section below.
The second way have an application in an R package is by having a function that returns a Shiny application object. In this example, there’s a function helloWorldApp(), which lives in R/helloworld.R:
/
├── .Rbuildignore
├── DESCRIPTION
├── NAMESPACE
├── R
│ └── helloworld.R
│
└── tests
├── testthat
│ ├── apps/
│ │ └── helloworld/
| | └── app.R
│ └── test-app-function.R
└── testthat.RThe function simply returns an object from shinyApp():
helloWorldApp <- function() {
utils::data(cars)
shinyApp(
ui = fluidPage(
sliderInput("n", "n", 1, nrow(cars), 10),
plotOutput("plot")
),
server = function(input, output) {
output$plot <- renderPlot({
plot(head(cars, input$n), xlim = range(cars[[1]]), ylim = range(cars[[2]]))
})
}
)
}To use shinytest to test the application, the key is to add an app.R file that simply calls the function, and then use shinytest on that app.R. You can see the example application. In this case, the application lives at tests/testthat/apps/helloworld/app.R, and contains the following:
library(shinytestPackageExample)
helloWorldApp()You can call recordApp() and testApp() on this application as normal to create the test script and expected results.
Assuming you are using testthat for automated tests, you would then create a test driver script in tests/testthat/. In this example, it’s named test-app-function.R and contains the following:
context("app-function")
# This file is for testing the applications in the apps/ directory.
library(shinytest)
test_that("helloWorldApp() works", {
# Don't run these tests on the CRAN build servers
skip_on_cran()
# Use compareImages=FALSE because the expected image screenshots were created
# on a Mac, and they will differ from screenshots taken on the CI platform,
# which runs on Linux.
expect_pass(testApp(test_path("apps/helloworld/"), compareImages = FALSE))
})For more information about this script, see the Test driver script notes section below.
There are a few steps that are needed for both types of tests.
You will need to add shinytest to the Suggests section in your DESCRIPTION file.
Suggests:
shinytestWhen all of these items are in place, you can test your package using testthat::test() or by running R CMD check on your package. If you are using the RStudio IDE, you can also run Build -> Test Package or Build -> Check Package.
Here is one of the test driver scripts (which lives in tests/testthat/) from above:
context("app-function")
# This file is for testing the applications in the apps/ directory.
library(shinytest)
test_that("helloWorldApp() works", {
# Don't run these tests on the CRAN build servers
skip_on_cran()
# Use compareImages=FALSE because the expected image screenshots were created
# on a Mac, and they will differ from screenshots taken on the CI platform,
# which runs on Linux.
expect_pass(testApp(test_path("apps/helloworld/"), compareImages = FALSE))
})It has a few unusual features:
If you will submit the package to CRAN, it is best to configure the application tests to not run on the CRAN build servers. If you are using testthat, use skip_on_cran() in the test block, as in the example.
If you will be testing the package on multiple different platforms, then when calling testApp(), use compareImages = FALSE. This is because the the screenshots on will likely have small differences across different platforms, and result in spurious test failures.
You may have noticed the expect_pass() function, which is from the shinytest package (instead of most expect_ functions, which are from the testthat package).
Also note that the shinytest scripts require your package to be installed. testthat::test_local() (and related wrappers) eventually call pkgload::load_all() to temporarily source the local R package. You can use test_local() to test non-shinytest tests, but you will need to install your R package to safely execute your shinytest tests. If not installed, it will create a confusing situation where your shinytest tests run on a different version of your R package (whichever was last installed), than the rest of your tests (the current source).
If you would like your package to be tested with every commit, you can set it up with Travis CI as described in Hadley Wickham’s R Packages book.