Beyond the Buzzwords — A Field Manual for Functional, Integration, Smoke and Regression Testing

“Quality is a design parameter, not a phase.”
This article dissects four pillars every SDET leans on—Functional, Integration, Smoke, Regression—and shows where they live in a pipeline, how they fail, and how to keep them cheap.

0 · Why Another Glossary?

Because most definitions are either too academic (“black‑box verification of an externally visible behaviour…”) or too shallow (“does the page load?”).
Below you’ll find the working definitions my teams and I have converged on after shipping several B2C mobile apps and microservice backends.

1 · Functional Tests

Validate a single feature in isolation—UI optional

1.1 Example — iOS SwiftUI ViewModel

final class CartViewModelTests: XCTestCase {
    func test_addItem_updates_total() {
        let repo = FakeCatalogRepository()
        let vm   = CartViewModel(catalog: repo)

        vm.add(itemID: "capuccino", quantity: 2)

        XCTAssertEqual(vm.total, 7.0)    // local currency, tax‑free
        XCTAssertEqual(repo.hitCount, 1) // assert single repo call
    }
}

The repository is a fake — no SQLite, no CoreData, no HTTP. That keeps the test deterministic and 6 × faster than touching real storage.

1.2 When Functional Tests Fail

• Spec drift — PM changed acceptance criteria but BDD/Gherkin not updated.
• UI refactor — UI test locators coupled to class names rather than semantic IDs.

Guardrail: enforce semantic accessibility identifiers ("CartTotalLabel", not "label3").

2 · Integration Tests

Prove that contracts between two real components hold

2.1 Example — Swift Consumer Pact Test

@PactTest(for: "PricingService", provider: "pricing-api")
func test_price_contract() throws {
    given("SKU capuccino exists")
        .uponReceiving("Price lookup")
        .path("/v1/prices/capuccino")
        .method("GET")
    willRespondWith {
        status(200)
        json([
            "sku": "capuccino",
            "price": 3.5,
            "currency": "USD"
        ])
    }

    let client = PricingClient(base: mockServer.url)
    let price  = try await client.price(for: "capuccino")
    XCTAssertEqual(price.amount, 3.5)
}

When the provider publishes its own verification, the Pact Broker guarantees you never ship a schema mismatch.

2.2 Integration Anti‑Patterns

3 · Smoke Tests

Binary confidence check—no business logic, just “does it launch?”

3.1 Typical Mobile Smoke Suite (GitHub Actions)

Rollback rule: any red in this job aborts the deploy; no flaky retries.

4 · Regression Tests

Codified bug history—keep the zombies down

Regression ≠ layer. It’s a label attached to tests that reproduce previous defects.

4.1 Git Hygiene

# pytest.ini
[pytest]
markers =
    regression: previously reported defects

• On hot‑fix PRs, add @pytest.mark.regression to the new test.
• Nightly job ‑‑runs pytest -m regression first; if that suite goes green the full matrix follows.

4.2 Metrics to Watch

5 · Putting It All Together

5.1 Pipeline Blueprint (mobile + backend)

graph TD
    A[Dev Commit]
    A --> B(Unit + Static Analysis)
    B --> C(Functional Tag‑Select)
    C -->|<60 s| D(GitHub Review)
    D --> E(Stage Deploy)
    E --> F(Smoke Suite)
    E --> G(Integration Contracts)
    G --> H(Nightly Full Regression)
    F -->|green| I(Prod Deploy Canary)
    I --> J(Full Smoke on Real Devices)

Every test type lives once, has a clear gate responsibility, and a hardened time budget.

5.2 Cost ✕ Confidence Matrix

When deadlines loom, you cut from the right‑hand columns, never the left.

6 · Advanced Patterns (for Those Who Already Have the Basics)

1. Mutation Testing – run stryker‑swift, pitest, or mutmut nightly; mutants caught by functional tests reveal dead assertions.
2. Contract Diff Alerts – Git hook that fails PR if OpenAPI or Expo config changes without a matching pact update.
3. Test Selection by Affected Paths – run regression subset where git diff intersects the call‑graph; slashes PR runtime by 70 %.
4. Chaos in Canary – inject 30 % packet loss during smoke; if the app still loads, the UX team wins cookies.
5. Shadow Production – clone prod traffic to a dev flight build; treat unexpected 5xx as integration failures in real time.

7 · Common Pitfalls (And Cheap Fixes)

8 · Cheat‑Sheet for Interview Whiteboards

Layer        Boundary            Mocks?   Trigger      Budget   Failure Action
-----------  ------------------  -------  -----------  -------  -------------------
Unit         Single class/file   All      PR           < 30 s   fix before merge
Functional   Feature slice       External PR & Nightly < 4  m   fix / revert PR
Integration  Two real services   None     Nightly      < 10 m   block release
Smoke        Full stack, narrow  None     Deploy       < 2  m   auto‑rollback
Regression   Historical bug      Context  Branch‑/Night Varies  hot‑fix must pass

Memorise the middle column; recruiters love boundary talk.

9 · Key Takeaways

• Test type = risk vector – choose the layer that kills the costliest uncertainty first.
• Labels, not silos – “regression” and “smoke” cross‑cut layers; don’t duplicate code, duplicate intent.
• Time budgets keep CI honest; every extra second steals throughput from your team.
• Accessibility IDs pay double—stable locators for automation, inclusive UX for users.