Shell Testing

Shell Testing with BATS

Running Tests — Targeted Only

NEVER run the full test suite. Always run only the test files affected by the current changes.

1. Map source → test file: lib/helpers.sh → tests/helpers.bats, lib/ui.sh → tests/ui.bats, cmd/theme/* → tests/theme.bats
2. Run specific files: bats tests/helpers.bats tests/theme.bats
3. Filter by name: bats tests/helpers.bats --filter "format_insert"
4. Filter by tag: bats tests/ --filter-tags "unit"

When multiple source files changed, run only their corresponding test files. If unsure which tests cover a function, grep for the function name across tests/*.bats.

Framework Decision

• BATS (primary): All bash code, polyglot bash/zsh code, CLI commands
• Zunit: Only for pure zsh features (completions, zle widgets, zstyle) that cannot be tested via BATS
• bats --shell zsh (BATS 1.8.0+): Run BATS tests under zsh for polyglot verification

Test File Template

#!/usr/bin/env bats

load test_helper/bats-support/load
load test_helper/bats-assert/load
load test_helper/bats-file/load   # Only if testing filesystem

setup() {
  source "${BATS_TEST_DIRNAME}/../lib/module_under_test.sh"
}

# bats test_tags=unit
@test "function_name returns expected output for valid input" {
  run function_name "valid_input"
  assert_success
  assert_output "expected"
}

# bats test_tags=unit
@test "function_name fails with clear error on bad input" {
  run function_name "bad_input"
  assert_failure
  assert_output --partial "error"
}

Test Isolation Requirements (MANDATORY)

These are non-negotiable requirements for all shell tests. Violating these rules creates side effects and pollutes the real environment.

Never use real project/framework directories in tests

# NEVER — writes to real framework directory
MANIFEST_DIR="${DOTTY_ROOT}/local"
BACKUP_DIR="${PROJECT_ROOT}/backups"

# ALWAYS — create parallel structure in test temp space
MANIFEST_DIR="${TEST_TMPDIR}/mock-local"
BACKUP_DIR="${TEST_TMPDIR}/mock-backups"
mkdir -p "$MANIFEST_DIR" "$BACKUP_DIR"

If production code needs to write to framework directories, add override env var:

# In production code (e.g., cmd/dotty-undo)
DOTTY_LOCAL="${DOTTY_LOCAL_OVERRIDE:-${DOTTY_ROOT}/local}"

# In tests
export DOTTY_LOCAL_OVERRIDE="${TEST_TMPDIR}/mock-local"

Mock all external commands that modify state

# Mock git submodule
cat << 'EOF' > "${TEST_TMPDIR}/git"
#!/usr/bin/env bash
if [[ "$1" == "submodule" ]]; then
  echo "[mock] git submodule $*" >&2
  exit 0
fi
exec /usr/bin/git "$@"  # Pass through other commands
EOF
chmod +x "${TEST_TMPDIR}/git"
export PATH="${TEST_TMPDIR}:${PATH}"

# Mock package managers, installers, etc.
mock_brew() {
  cat << 'EOF' > "${TEST_TMPDIR}/brew"
#!/usr/bin/env bash
echo "[mock] brew $*" >&2
exit 0
EOF
  chmod +x "${TEST_TMPDIR}/brew"
}

Add safety guards to production commands

Every command that could be called by tests must detect test mode:

# At the top of cmd/dotty-install, cmd/dotty-update, etc.
if [[ -n "${BATS_TEST_TMPDIR:-}" ]]; then
  echo "[test mode] Skipping git submodule operations" >&2
else
  git submodule sync --quiet --recursive vendor/dotbot
  git submodule update --init --recursive vendor/dotbot
fi

Verify test environment isolation

Add to all test files before any tests run:

# Verify we're in test mode
if [[ -z "${BATS_TEST_TMPDIR:-}" ]]; then
  echo "ERROR: Not running in BATS environment" >&2
  exit 1
fi

# Verify dotfiles dir is not real (adapt to your project)
if [[ "${DOTTY_DOTFILES_DIR:-}" == "${HOME}/.dotfiles" ]]; then
  echo "ERROR: DOTTY_DOTFILES_DIR points to real dotfiles" >&2
  exit 1
fi

Skip tests gracefully when external tools unavailable

# NEVER — fails if tool missing
@test "tool is available" {
  run command -v mytool
  assert_success
}

# ALWAYS — skips gracefully
@test "feature works with mytool" {
  if ! command -v mytool &>/dev/null; then
    skip "mytool not installed (optional)"
  fi
  run use_mytool
  assert_success
}

Core Rules

Always use bats-assert over raw [ ]

# YES
run my_func; assert_success; assert_output --partial "done"

# NO — cryptic failure messages
run my_func; [ "$status" -eq 0 ]; [[ "$output" == *"done"* ]]

Use run when testing exit codes or output; direct calls for side effects

# run: capture status + output
run parse_config "$file"
assert_success
assert_output --partial "key=value"

# direct: need side effects in test scope
source lib/config.sh
load_config           # sets $CONFIG_VALUE in current scope
[ "$CONFIG_VALUE" = "loaded" ]

Name tests as behavior descriptions

Format: "<subject> <verb describing behavior> [when <condition>]"

@test "parse_config returns key-value pairs from valid file" { ... }
@test "parse_config fails with clear error on missing file" { ... }

Use $BATS_TEST_TMPDIR for temp files (auto-cleaned)

@test "creates output file" {
  run my_func --output "$BATS_TEST_TMPDIR/result.txt"
  assert_success
  assert_file_exist "$BATS_TEST_TMPDIR/result.txt"
}

Guard any manual rm -rf in teardown

teardown() {
  [ -n "$TEST_TMPDIR" ] && [ -d "$TEST_TMPDIR" ] && rm -rf "$TEST_TMPDIR"
}

Tag tests for filtering

# bats test_tags=unit
@test "isolated function test" { ... }

# bats test_tags=integration
@test "full workflow test" { ... }

Run: bats tests/ --filter-tags "unit" or bats tests/ --filter-tags "!integration"

run Semantics

After run: $status, $output, ${lines[@]}, $stderr (if --separate-stderr), ${stderr_lines[@]}

Assertions Quick Reference

assert_success                         # status == 0
assert_failure [code]                  # status != 0 (or == code)
assert_output "exact text"             # exact match
assert_output --partial "substring"    # contains
assert_output --regexp "^pattern$"     # regex
assert_line "text"                     # any line equals
assert_line --index 0 "first line"     # specific line
assert_line --partial "text"           # any line contains
refute_output "text"                   # does NOT match
refute_line "text"                     # no line matches
assert_file_exist "$path"             # bats-file
assert_dir_exist "$path"              # bats-file

Lifecycle Hooks

setup_file()     →  Once before all tests in file (expensive shared setup)
  setup()        →  Before each test (source code, create per-test state)
    @test        →  Runs in subshell (isolated)
  teardown()     →  After each test, even on failure
teardown_file()  →  Once after all tests in file

Mocking Strategy

See "Test Isolation Requirements (MANDATORY)" above for critical mocking rules.

Additional mocking patterns — prefer function overrides over bats-mock:

@test "handles API response" {
  curl() { echo '{"status":"ok"}'; }
  export -f curl
  run my_api_caller
  assert_output --partial "ok"
}

For CLI stubs, use PATH manipulation:

@test "uses external tool" {
  mkdir -p "$BATS_TEST_TMPDIR/bin"
  printf '#!/bin/bash\necho "stubbed"' > "$BATS_TEST_TMPDIR/bin/mytool"
  chmod +x "$BATS_TEST_TMPDIR/bin/mytool"
  run env PATH="$BATS_TEST_TMPDIR/bin:$PATH" my_script
  assert_success
}

Testing Zsh Code

Polyglot code — test with BATS directly, optionally verify under zsh:

bats tests/my.bats              # default (bash)
bats --shell zsh tests/my.bats  # verify under zsh

Zsh-only features — invoke zsh explicitly or use zunit:

@test "zsh completion works" {
  run zsh -c "source lib/completions.zsh; _my_completion"
  assert_success
}

For zunit patterns, see references/zsh-testing.md.

File Organization

One test file per source file. Shared helpers in test_helper.bash. Fixtures in fixtures/.

tests/
  helpers.bats            # tests for lib/helpers.sh
  ui.bats                 # tests for lib/ui.sh
  commands.bats           # smoke tests for CLI commands
  test_helper.bash        # shared setup, custom assertions
  fixtures/
    sample_config.txt

Detailed References

• Patterns & examples: references/patterns.md — sourcing, multi-line output, stdin, stderr, TTY testing, fixtures, parallel safety, helper libraries, CI setup
• Anti-patterns & gotchas: references/gotchas.md — dangerous teardown, negation bug, variable scope, platform portability, flaky tests
• Zsh-specific testing: references/zsh-testing.md — zunit syntax, zsh-test-runner, polyglot patterns, testing completions/widgets