# Testing Guide
This guide covers test execution, categorization, and CI integration for
`duckduckgo-search-cli`.
## v0.6.5 Test Additions
The v0.6.5 release added 11 tests, all addressing previously open gaps:
- **WS-11** (5 tests) — property-based invariants for the HTML parser in
`extraction.rs`. Validates that empty inputs yield empty `Vec`, positions
are dense and 1-based, URLs are normalized to absolute paths, the parser
is deterministic, and malformed HTML does not panic. These tests would
have caught the v0.6.3 → v0.6.4 migration regressions.
- **WS-12** (4 tests) — per-host circuit breaker in `content_fetch.rs`.
Validates the closed-state allows requests, the threshold opens the
breaker, a single success resets the failure counter, and the half-open
state is reachable after the cooldown window.
- **WS-23** (1 test) — wiremock integration test for the `Retry-After`
header on HTTP 429 responses. Validates the backoff delay is at least
`Retry-After` seconds, with a 500ms slack for CI scheduler overhead.
- **Existing 322 tests preserved** — the v0.6.5 changes are purely additive.
No tests removed, no test signatures changed, no test fixtures renamed.
### v0.6.5 gaps closed by these tests
- **MP-26** (Windows HANDLE) — validated by `cargo test --all-features`
on `windows-latest` CI runner (added in this release).
- **CI-01** (6 clippy errors) — `cargo clippy --all-targets --all-features -- -D warnings`
now passes, which is itself a "test" that no lint regression exists.
- **WS-12** (circuit breaker) — covered by 4 unit tests in
`src/content_fetch.rs`.
- **WS-23** (Retry-After) — covered by 1 wiremock test in
`tests/integration_wiremock.rs`.
## Why Categorized Tests
The test suite is split into four categories to balance speed, isolation,
and coverage:
| Unit | < 1 s | per-fn | none | 243 |
| Integration | < 30 s | per-test | localhost | 84 |
| Doc | < 5 s | per-doc | none | 6 |
| Loom | n/a | n/a | n/a | 0 (gated) |
## Test Categories
### Unit Tests
Located in `src/**/tests` modules (mod tests). Fast, in-process, no I/O.
Run with:
```bash
cargo test --lib
```
### Integration Tests
Located in `tests/*.rs` files. Use wiremock (no real HTTP), assert_cmd (no real
subprocess spawn), and tempfile (no real FS writes outside tmpdir).
```bash
# All integration tests
cargo test --tests
# Single integration test file
cargo test --test integration_wiremock
```
### Doc Tests
Located in `///` examples throughout `src/`. Compiled and executed by `cargo test --doc`.
```bash
cargo test --doc
```
### Loom Tests
Located in `tests/loom_atomics.rs`. Gated by `--cfg loom`. NOT compiled by
default — requires explicit opt-in.
```bash
RUSTFLAGS="--cfg loom" cargo test --test loom_atomics --release
```
> **Known limitation**: Loom conflicts with `hyper-util` and currently
> compiles but does not run cleanly. Issue tracked upstream.
## How to Run
### Local Development
```bash
# Quick feedback loop
timeout 300 cargo test --all-features --locked
# Specific category
cargo test --lib --locked
cargo test --tests --locked
cargo test --doc --locked
```
### With Coverage
```bash
# Install cargo-llvm-cov
cargo install cargo-llvm-cov
# Run with HTML report
cargo llvm-cov --all-features --locked --html --open
# Run with text summary only
cargo llvm-cov --all-features --locked --summary-only
```
Minimum line coverage: **80%**. CI fails below this threshold.
### Property-Based Tests (v0.6.5, WS-11)
5 invariants in `src/extraction.rs`:
```bash
cargo test ws11_
# Run all 5 property tests:
# - ws11_invariant_empty_inputs_yield_empty_results
# - ws11_invariant_positions_are_dense_and_one_based
# - ws11_invariant_urls_are_normalized_to_absolute
# - ws11_invariant_extraction_is_idempotent
# - ws11_invariant_malformed_html_does_not_panic
```
### WireMock Retry-After Test (v0.6.5, WS-23)
```bash
cargo test --test integration_wiremock test_retry_after_header_respected
```
### Circuit Breaker Tests (v0.6.5, WS-12)
```bash
cargo test ws12_
# Tests: ws12_breaker_allows_when_closed,
# ws12_breaker_opens_after_threshold_failures,
# ws12_breaker_resets_on_success,
# ws12_breaker_half_opens_after_cooldown
```
## Environment Variables
| `RUST_TEST_THREADS` | Number of parallel test threads (default 1) |
| `RUST_BACKTRACE` | Set to `1` or `full` for detailed backtraces |
| `RUST_LOG` | Tracing filter (`debug`, `info`, `warn`, `error`) |
| `CARGO_TERM_COLOR` | Force ANSI colors (`always`, `never`, `auto`) |
| `LOOM_MAX_PREEMPTIONS` | Max preemption bound for loom tests |
| `WIREMOCK_LOG` | WireMock request/response logging |
## CI Profiles
Three CI jobs run the test suite:
1. **`validate` matrix** — `cargo test --all-features --locked` on Linux, macOS, Windows
2. **`msrv`** — `cargo check --all-targets --all-features --locked` on Rust 1.75
3. **`coverage`** — `cargo llvm-cov --all-features --locked --fail-under-lines 80` on Linux
Plus a manual `cargo nextest` profile available locally:
```toml
# .config/nextest.toml (not in repo, per project convention)
[profile.default]
retries = 2
test-threads = 1
```
## Troubleshooting
### `flaky::lazy_template` failures
Loom tests may be flaky. Re-run with:
```bash
RUSTFLAGS="--cfg loom" cargo test --test loom_atomics --release -- --test-threads=1
```
### `wiremock::MockServer` startup timeout
Increase the wait:
```bash
WIREMOCK_LOG=info cargo test --test integration_wiremock
```
### Coverage drops below 80%
Check the HTML report for uncovered lines:
```bash
cargo llvm-cov --html --open
```
The diff will show which lines are not exercised by the test suite. Add
unit or integration tests to cover the missing branches.
### Tests pass locally but fail in CI
- Check for environment-specific behavior (paths, timeouts, locale)
- Check for `Instant::now()` non-determinism in code under test
- Use `cargo nextest` with retries to detect flaky tests:
```bash
cargo nextest run --retries 3
```