Skip to main content

DI testing, graph validation, startup, and leaks

Test behavior without a graph when possible​

val vm = CheckoutViewModel(
    savedStateHandle = SavedStateHandle(),
    placeOrder = FakePlaceOrder(result = acceptedOrder),
)

Constructor injection makes unit tests cheap. Use Hilt tests when the test question is wiring, Android integration, qualifiers, replacements, or scopes.

Replace bindings deliberately​

Test modules can uninstall/replace production modules or bind test values. Keep replacements narrow; a giant test graph that silently swaps many dependencies can pass while production wiring is wrong.

Useful graph tests:

  • production variants compile all roots
  • each qualifier resolves expected implementation
  • scoped identity is same within scope and different across scope instances
  • optional/multibound plugins are present exactly where intended
  • debug/release/flavor graphs do not accidentally expose test endpoints

Startup cost is often provider work​

Dagger-generated lookup is normally cheap; constructors/provider methods can be expensive. Avoid database opening, disk reads, key generation, SDK initialization, or network work merely because a singleton is requested at startup.

Use lazy/provider only when delayed or repeated creation semantics are understood:

class Feature @Inject constructor(private val sdk: dagger.Lazy<HeavySdk>) {
    fun open() = sdk.get().show()
}

Lazy caches once for that injected Lazy; Provider.get() may create according to binding scope each call.

Scope leak scenarios​

  • singleton holds Activity, View, NavController, or callback
  • ActivityRetained object holds old Activity after rotation
  • flow component remains in Application after logout
  • listener registration outlives component disposal
  • multibound startup task captures feature UI

Memory profiler/LeakCanary finds symptoms; scope design prevents them.

Build performance​

Large generated graphs affect annotation processing and incremental builds. Prefer constructor injection, avoid monolithic modules/components, keep module dependencies acyclic, use supported KSP/tooling where appropriate, and measure build scans rather than guessing.

Security/configuration​

Qualifiers can accidentally wire debug endpoint, logging interceptor, or mock auth into release. Add variant-specific graph tests and release configuration assertions. DI is configuration code and deserves security review.

False confidence

A compiling graph proves constructibility, not correct scope, endpoint, thread behavior, privacy, or business semantics. Those require targeted tests and runtime evidence.

Interview exercises​

  1. Investigate slow startup caused by eager singleton construction.
  2. Write a scope identity test across two Activity instances.
  3. Prevent debug network bindings from entering release.