Write testthat Tests

Create comprehensive tests for R package functions using testthat edition 3.

When to Use

• Adding tests for new package functions
• Increasing test coverage for existing code
• Writing regression tests for bug fixes
• Setting up test infrastructure for a new package

Inputs

• Required: R functions to test
• Required: Expected behavior and edge cases
• Optional: Test fixtures or sample data
• Optional: Target coverage percentage (default: 80%)

Procedure

Step 1: Set Up Test Infrastructure

If not already done:

usethis::use_testthat(edition = 3)

This creates tests/testthat.R and tests/testthat/ directory.

Got: tests/testthat.R and tests/testthat/ directory created. DESCRIPTION has Config/testthat/edition: 3 set.

If fail: If usethis is not available, manually create tests/testthat.R containing library(testthat); library(packagename); test_check("packagename") and add tests/testthat/ directory.

Step 2: Create Test File

usethis::use_test("function_name")

This creates tests/testthat/test-function_name.R with a template.

Got: Test file created at tests/testthat/test-function_name.R with a placeholder test_that() block ready to fill in.

If fail: If usethis::use_test() is not available, manually create the file. Follow the naming convention test-<function_name>.R.

Step 3: Write Basic Tests

test_that("weighted_mean computes correct result", {
  expect_equal(weighted_mean(1:3, c(1, 1, 1)), 2)
  expect_equal(weighted_mean(c(10, 20), c(1, 3)), 17.5)
})

test_that("weighted_mean handles NA values", {
  expect_equal(weighted_mean(c(1, NA, 3), c(1, 1, 1), na.rm = TRUE), 2)
  expect_true(is.na(weighted_mean(c(1, NA, 3), c(1, 1, 1), na.rm = FALSE)))
})

test_that("weighted_mean validates input", {
  expect_error(weighted_mean("a", 1), "numeric")
  expect_error(weighted_mean(1:3, 1:2), "length")
})

Got: Basic tests cover correct output for typical inputs, NA handling behavior, and input validation error messages.

If fail: If tests fail immediately, verify the function is loaded (devtools::load_all()). If error messages do not match, use a regex pattern in expect_error() instead of an exact string.

Step 4: Test Edge Cases

test_that("weighted_mean handles edge cases", {
  # Empty input
  expect_error(weighted_mean(numeric(0), numeric(0)))

  # Single value
  expect_equal(weighted_mean(5, 1), 5)

  # Zero weights
  expect_true(is.nan(weighted_mean(1:3, c(0, 0, 0))))

  # Very large values
  expect_equal(weighted_mean(c(1e15, 1e15), c(1, 1)), 1e15)

  # Negative weights
  expect_error(weighted_mean(1:3, c(-1, 1, 1)))
})

Got: Edge cases are covered: empty input, single values, zero weights, extreme values, and invalid inputs. Each edge case has a clear expected behavior.

If fail: If the function does not handle an edge case as expected, decide whether to fix the function or adjust the test. Document the intended behavior for ambiguous cases.

Step 5: Use Fixtures for Complex Tests

Create tests/testthat/fixtures/ for test data:

# tests/testthat/helper.R (loaded automatically)
create_test_data <- function() {
  data.frame(
    x = c(1, 2, 3, NA, 5),
    group = c("a", "a", "b", "b", "b")
  )
}

# In test file
test_that("process_data works with grouped data", {
  test_data <- create_test_data()
  result <- process_data(test_data)
  expect_s3_class(result, "data.frame")
  expect_equal(nrow(result), 2)
})

Got: Fixtures provide consistent test data across multiple test files. Helper functions in tests/testthat/helper.R are loaded automatically by testthat.

If fail: If helper functions are not found, ensure the file is named helper.R (not helpers.R) and is located in tests/testthat/. Restart the R session if needed.

Step 6: Mock External Dependencies

test_that("fetch_data handles API errors", {
  local_mocked_bindings(
    api_call = function(...) stop("Connection refused")
  )
  expect_error(fetch_data("endpoint"), "Connection refused")
})

test_that("fetch_data returns parsed data", {
  local_mocked_bindings(
    api_call = function(...) list(data = list(value = 42))
  )
  result <- fetch_data("endpoint")
  expect_equal(result$value, 42)
})

Got: External dependencies (APIs, databases, network calls) are mocked so tests run without real connections. Mock return values exercise the function's data processing logic.

If fail: If local_mocked_bindings() fails, ensure the function being mocked is accessible in the test scope. For functions in other packages, use the .package argument.

Step 7: Snapshot Tests for Complex Output

test_that("format_report produces expected output", {
  expect_snapshot(format_report(test_data))
})

test_that("plot_results creates expected plot", {
  expect_snapshot_file(
    save_plot(plot_results(test_data), "test-plot.png"),
    "expected-plot.png"
  )
})

Got: Snapshot files are created in tests/testthat/_snaps/. First run creates the baseline; subsequent runs compare against it.

If fail: If snapshots fail after an intentional change, update them with testthat::snapshot_accept(). For cross-platform differences, use the variant parameter to maintain platform-specific snapshots.

Step 8: Use Skip Conditions

test_that("database query works", {
  skip_on_cran()
  skip_if_not(has_db_connection(), "No database available")

  result <- query_db("SELECT 1")
  expect_equal(result[[1]], 1)
})

test_that("parallel computation works", {
  skip_on_os("windows")
  skip_if(parallel::detectCores() < 2, "Need multiple cores")

  result <- parallel_compute(1:100)
  expect_length(result, 100)
})

Got: Tests that require special environments (network, database, multiple cores) are guarded with skip conditions. These tests run locally but are skipped on CRAN or restricted CI environments.

If fail: If tests fail on CRAN or CI but pass locally, add the appropriate skip_on_cran(), skip_on_os(), or skip_if_not() guard at the top of the test_that() block.

Step 9: Run Tests and Check Coverage

# Run all tests
devtools::test()

# Run specific test file
devtools::test_active_file()  # in RStudio
testthat::test_file("tests/testthat/test-function_name.R")

# Check coverage
covr::package_coverage()
covr::report()

Got: All tests pass with devtools::test(). Coverage report shows the target percentage is met (aim for >80%).

If fail: If tests fail, read the test output for specific assertion failures. If coverage is below target, use covr::report() to identify untested code paths and add tests for them.

Validation

• All tests pass with devtools::test()
• Coverage exceeds target percentage
• Every exported function has at least one test
• Error conditions are tested
• Edge cases are covered (NA, NULL, empty, boundary values)
• No tests depend on external state or order of execution

Pitfalls

• Tests depending on each other: Each test_that() block must be independent
• Hardcoded file paths: Use testthat::test_path() for test fixtures
• Floating point comparison: Use expect_equal() (has tolerance) not expect_identical()
• Testing private functions: Test through the public API when possible. Use ::: sparingly.
• Snapshot tests in CI: Snapshots are platform-sensitive. Use variant parameter for cross-platform.
• Forgetting skip_on_cran(): Tests requiring network, databases, or long runtime must skip on CRAN

Examples

# Pattern: test file mirrors R/ file
# R/weighted_mean.R -> tests/testthat/test-weighted_mean.R

# Pattern: descriptive test names
test_that("weighted_mean returns NA when na.rm = FALSE and input contains NA", {
  result <- weighted_mean(c(1, NA), c(1, 1), na.rm = FALSE)
  expect_true(is.na(result))
})

# Pattern: testing warnings
test_that("deprecated_function emits deprecation warning", {
  expect_warning(deprecated_function(), "deprecated")
})

Related Skills

• create-r-package - set up test infrastructure as part of package creation
• write-roxygen-docs - document the functions you test
• setup-github-actions-ci - run tests automatically on push
• submit-to-cran - CRAN requires tests to pass on all platforms