Test Shiny App

Set up comprehensive testing for Shiny applications using shinytest2 (end-to-end) and testServer() (unit tests).

When to Use

• Adding tests to an existing Shiny application
• Setting up a testing strategy for a new Shiny project
• Writing regression tests before refactoring Shiny code
• Integrating Shiny app tests into CI/CD pipelines

Inputs

• Required: Path to the Shiny application
• Required: Test scope (unit tests, end-to-end, or both)
• Optional: Whether to use snapshot testing (default: yes for e2e)
• Optional: CI platform (GitHub Actions, GitLab CI)
• Optional: Modules to test in isolation

Procedure

Step 1: Install Testing Dependencies

install.packages("shinytest2")

# For golem apps, add as a Suggests dependency
usethis::use_package("shinytest2", type = "Suggests")

# Set up testthat infrastructure if not present
usethis::use_testthat(edition = 3)

Got: shinytest2 installed and testthat directory structure in place.

If fail: shinytest2 requires chromote (headless Chrome). Install Chrome/Chromium on the system. On WSL: sudo apt install -y chromium-browser. Verify with chromote::find_chrome().

Step 2: Write testServer() Unit Tests for Modules

Create tests/testthat/test-mod_dashboard.R:

test_that("dashboard module filters data correctly", {
  testServer(dataFilterServer, args = list(
    data = reactive(iris),
    columns = c("Species", "Sepal.Length")
  ), {
    # Set inputs
    session$setInputs(column = "Species")
    session$setInputs(value_select = "setosa")
    session$setInputs(apply = 1)

    # Check output
    result <- filtered()
    expect_equal(nrow(result), 50)
    expect_true(all(result$Species == "setosa"))
  })
})

test_that("dashboard module handles empty data", {
  testServer(dataFilterServer, args = list(
    data = reactive(iris[0, ]),
    columns = c("Species")
  ), {
    # Module should not error on empty data
    expect_no_error(session$setInputs(column = "Species"))
  })
})

Key patterns:

• testServer() tests module server logic without a browser
• Pass reactive arguments via the args list
• Use session$setInputs() to simulate user interactions
• Access reactive return values directly by name
• Test edge cases: empty data, NULL inputs, invalid values

Got: Module tests pass with devtools::test().

If fail: If testServer() errors with "not a module server function", ensure the function uses moduleServer() internally. If session$setInputs() doesn't trigger reactives, add session$flushReact() after setting inputs.

Step 3: Write shinytest2 End-to-End Tests

Create tests/testthat/test-app-e2e.R:

test_that("app loads and displays initial state", {
  # For golem apps
  app <- AppDriver$new(
    app_dir = system.file(package = "myapp"),
    name = "initial-load",
    height = 800,
    width = 1200
  )
  on.exit(app$stop(), add = TRUE)

  # Wait for app to load
  app$wait_for_idle(timeout = 10000)

  # Check that key elements exist
  app$expect_values()
})

test_that("filter interaction updates the table", {
  app <- AppDriver$new(
    app_dir = system.file(package = "myapp"),
    name = "filter-interaction"
  )
  on.exit(app$stop(), add = TRUE)

  # Interact with the app
  app$set_inputs(`filter1-column` = "cyl")
  app$wait_for_idle()

  app$set_inputs(`filter1-apply` = "click")
  app$wait_for_idle()

  # Snapshot the output values
  app$expect_values(output = "table")
})

Key patterns:

• AppDriver$new() launches the app in headless Chrome
• Always use on.exit(app$stop()) to clean up
• Module input IDs use the format "moduleId-inputId"
• app$expect_values() creates/compares snapshot files
• app$wait_for_idle() ensures reactive updates complete

Got: End-to-end tests create snapshot files in tests/testthat/_snaps/.

If fail: If Chrome isn't found, set CHROMOTE_CHROME environment variable to the Chrome binary path. If snapshots fail on CI but pass locally, check for platform-dependent rendering differences — use app$expect_values() for data snapshots rather than app$expect_screenshot() for visual ones.

Step 4: Record a Test Interactively (Optional)

shinytest2::record_test("path/to/app")

This opens the app in a browser with a recording panel. Interact with the app, then click "Save test" to auto-generate test code.

Got: A test file is generated in tests/testthat/ with recorded interactions.

If fail: If the recorder doesn't open, check that the app runs successfully with shiny::runApp() first. The recorder requires a working app.

Step 5: Set Up Snapshot Management

For snapshot-based tests, manage expected values:

# Accept new/changed snapshots after review
testthat::snapshot_accept("test-app-e2e")

# Review snapshot differences
testthat::snapshot_review("test-app-e2e")

Add snapshot directories to version control:

tests/testthat/_snaps/    # Committed — contains expected values

Got: Snapshot files tracked in git for regression detection.

If fail: If snapshots change unexpectedly, run testthat::snapshot_review() to see the diffs. Accept intentional changes with testthat::snapshot_accept().

Step 6: Integrate with CI

Add to .github/workflows/R-CMD-check.yaml or create a dedicated workflow:

- name: Install system dependencies
  run: |
    sudo apt-get update
    sudo apt-get install -y chromium-browser

- name: Set Chrome path
  run: echo "CHROMOTE_CHROME=$(which chromium-browser)" >> $GITHUB_ENV

- name: Run tests
  run: |
    Rscript -e 'devtools::test()'

For golem apps, ensure the app package is installed before testing:

- name: Install app package
  run: Rscript -e 'devtools::install()'

Got: Tests pass in CI with headless Chrome.

If fail: Common CI issues: Chrome not installed (add the apt-get step), display server missing (shinytest2 uses headless mode by default so this usually isn't an issue), or timeout on slow runners (increase timeout in AppDriver$new()).

Validation

• devtools::test() runs all tests without errors
• testServer() tests cover module server logic
• shinytest2 tests cover key user workflows
• Snapshot files are committed to version control
• Tests pass in CI environment
• Edge cases tested (empty data, NULL inputs, error states)

Pitfalls

• Testing UI rendering instead of logic: Prefer testServer() for logic and app$expect_values() for data. Only use app$expect_screenshot() when visual appearance matters — screenshots are brittle across platforms.
• Module ID format in e2e tests: When setting module inputs via AppDriver, use "moduleId-inputId" format (hyphen-separated), not "moduleId.inputId".
• Flaky timing: Always call app$wait_for_idle() after app$set_inputs(). Without it, assertions may run before reactive updates complete.
• Snapshot drift: Don't commit snapshots generated on different platforms (Mac vs Linux). Standardize on the CI platform for snapshot generation.
• Missing Chrome on CI: shinytest2 requires Chrome/Chromium. Always include the installation step in CI workflows.

Related Skills

• build-shiny-module — create testable modules with clear interfaces
• scaffold-shiny-app — set up app structure with testing infrastructure
• write-testthat-tests — general testthat patterns for R packages
• setup-github-actions-ci — CI/CD setup for R packages (golem apps)