tdd-workflow

name	tdd-workflow
trigger	writing or reviewing tests
description	Test-driven development workflow. Use when writing any R code (writing new features, fixing bugs, refactoring, or reviewing tests).

TDD workflow

Core principle

Write a failing test first, then implement the minimal code to make it pass, then refactor. Never write implementation code without a failing test driving it.

File naming

Tests for R/{name}.R go in tests/testthat/test-{name}.R. Place new tests next to similar existing ones.

Running tests

# Full suite
devtools::test(reporter = "check")

# Single file
devtools::test(filter = "name", reporter = "check")

Testing functions load code automatically. You do not need to call library() or devtools::load_all() separately.

Coverage

Goal: 100% for every edited file. After editing R/file_name.R, verify:

covr_res <- devtools:::test_coverage_active_file("R/file_name.R")
which(purrr::map_int(covr_res, "value") == 0)

Files excluded from the coverage requirement:

• R/*-package.R
• R/aaa-shared_params.R
• Files matching R/import-standalone-*.R

Test types

Unit tests

Test individual functions in isolation:

test_that("fetch_records() returns a tibble (#2)", {
  result <- fetch_records(sample_input)
  expect_s3_class(result, "tbl_df")
})

Integration tests

Test end-to-end pipelines through multiple functions:

test_that("build_report() produces expected output (#15)", {
  input <- data.frame(value = c(1.123, 2.456, NA))
  result <- build_report(input, tempfile())
  expect_equal(nrow(result), 2L)
})

Snapshot tests

For complex outputs that are hard to specify with equality assertions:

test_that("build_summary print method is stable (#123)", {
  expect_snapshot(print(build_summary(sample_data)))
})

When snapshots change intentionally, check the content of the file corresponding to the edited test file, then accept:

testthat::snapshot_accept("test_name")

Snapshots are stored in tests/testthat/_snaps/. The filename corresponds to the R file being tested, ending with .md.

Test design principles

• Self-sufficient: each test contains its own setup, execution, and assertion. Tests must be runnable in isolation.
• Duplication over factoring: repeat setup code rather than extracting it. Clarity beats DRY in tests.
• One concept per test: a failing test should tell you exactly what broke.
• Minimal with few comments: keep tests lean. Avoid over-commenting.
• Issue reference in description: the desc of every new test_that() call should end with one or more parenthetical issue references for the issue(s) verified by those tests — typically the issue currently being solved. Never guess or invent issue numbers; if no tracked issue applies, use #noissue:

  test_that("fetch_records() returns correct columns (#1)", { ... })
  test_that("build_summary() returns correct columns (#2, #3)", { ... })
  test_that(".check_record() errors on empty input (#noissue)", { ... })
  

testthat Edition 3 — deprecated patterns

# Deprecated → Modern
context("Data validation")        # Remove — filename serves this purpose
expect_equivalent(x, y)          # expect_equal(x, y, ignore_attr = TRUE)
with_mock(...)                    # local_mocked_bindings(...)
expect_is(x, "data.frame")       # expect_s3_class(x, "data.frame")

Essential expectations

Equality & identity

expect_equal(x, y)                        # with numeric tolerance
expect_equal(x, y, tolerance = 0.001)
expect_equal(x, y, ignore_attr = TRUE)
expect_identical(x, y)                    # exact match

Conditions

Errors thrown by this package (via .stbl_abort()) should always be tested with expect_pkg_error_snapshot() (defined in tests/testthat/helper-expectations.R), which captures both the error class hierarchy and the user-facing message in one snapshot:

test_that("process_data() errors on empty input (#42)", {
  expect_pkg_error_snapshot(
    process_data(data.frame()),
    "empty_input"
  )
})

Pass transform = .transform_path(path) to scrub volatile values (e.g. temp paths) from the snapshot before comparison.

Errors from other packages can be tested with expect_error().

expect_warning(code)
expect_no_warning(code)
expect_message(code)
expect_no_message(code)

Collections

expect_setequal(x, y)           # same elements, any order
expect_in(element, set)
expect_named(x, c("a", "b"))

Type & structure

expect_type(x, "double")
expect_s3_class(x, "tbl_df")
expect_length(x, 10)
expect_null(x)

Logical

These expectations are a last resort when more-specific checks aren't available.

expect_true(x)
expect_false(x)

withr patterns for temporary state

withr::local_options(list(stbl.verbose = TRUE))
withr::local_envvar(MY_VAR = "value")
withr::local_tempfile(lines = c("a", "b"))

Fixtures

Store static test data in tests/testthat/fixtures/ and access via:

test_path("fixtures", "sample.rds")

Mocking

Mock functions that might have unstable output, hit external servers, etc.

local_mocked_bindings(
  .other_fn = function(...) "mocked_result"
)
result <- my_function_that_calls_other_fn()

Common mistakes

• Do not modify tests to make them pass. Fix the implementation.
• Do not write tests that depend on other tests' state. Each test must be independently runnable.
• Ask for help if test is bad. If you think a test might be invalid, do not loop through trying to make impossible tests pass. Ask for help if possible.