Twitch For Android: From Meme to Dream

1. From Meme to Dream
2. About me Joaquim Verges Android Lead @ Twitch joenrv
3. Level 1: meme App with big dreams
4. State of the app in March 2017 • 9M MAU • < 1% crashes • 2-3 developers • Basic functionality
5. Everything was fine until... • New PM with Big ambitions • App redesign • New mobile features • New hires Suddenly, all the issues with the current app surfaced
6. Looking under the hood • Fragment & View classes with thousands of lines of code • No common pattern or structure • Manual JSON model parsing • No tests whatsoever • ….
7. Let the grind begin
8. Twitch’s MVP pattern • Data Layer: ◦ Fetcher ◦ Models ◦ Api • Presentation Layer ◦ Presenter ◦ Tracker • View Layer ◦ ViewDelegate ◦ AdapterBinder Data Layer Presentation Layer View Layer Requests data Data loaded Callbacks Binds View User interaction Callbacks
9. Unit tests as a culture • With a testable pattern in place, no excuse for not testing your code! • Push for injecting dependencies • Run tests on every commit • Weekly report of % of code coverage
10. Cleanup • Nullability annotations • Retrofit for all API calls, get rid of Volley • GSON, remove all manual JSON parsing • Parceler for object serialization
11. 3 months later • App redesign complete • ~70% main screens rewritten • ~15% code coverage • Big features added • Team of 11 developers
12. Level 2: Stronger, but can do better
13. Modernizing the code base
14. Migrating to Kotlin • First Kotlin class introduced in June 2017 • Never looked back • Lessons learned: ◦ We quickly needed a standard for Kotling coding style ◦ Built-in nullability is a blessing, but beware of java interop • Code distribution as of September 2018:
15. Introducing RxJava • Motivation ◦ Reduce callback hell of complex use cases ◦ Cancel background tasks • Lessons learned ◦ Importance of educating the team on operators and general philosophy ◦ Enforce disposing of subscriptions, Kotlin extensions ◦ RxPlugins: Global error handler, JUnit rule for easy unit testing • Today: ◦ Most API calls return Observables ◦ Heavy usage of Subjects
16. Dependency Injection with Dagger 2 • Motivation: ◦ Piping data through deeply nested layers of objects is cumbersome ◦ No compile time validation, No object scoping ◦ Reduce usage of global singletons • Lessons learned: ◦ Even harder sell than RxJava: steep learning curve, no immediate benefit ◦ AndroidInjection helps reducing some initial boilerplate ◦ Benefits start showing only after a good portion of the code base is migrated • Today: ◦ ~90% of our Presenters are constructor injected ◦ Ongoing work to get rid of global Singletons
17. Migrating to GraphQL • Motivation: ◦ Big company push to move from REST to GQL ◦ Less dependencies on the backend, more flexibility ◦ Partial object caching • Lessons learned: ◦ Relying on a pre-alpha GQL library is scary ◦ Schema based, awkward versioning ◦ Beware of your fragments usage • Today: ◦ ~80% of our API calls go through GQL
18. UI testing with Espresso • Motivation: ◦ Reduce workload on QA for regression testing ◦ Reduce our release cadence to 2 weeks ◦ Improve confidence for big code changes • Lessons learned: ◦ Having a easy DSL and base framework helped a lot ◦ Team-wide “UI testing day” to improve coverage ◦ UI tests flakiness is a real pain • Today: ◦ ~70 unique tests, covers the main user flows ◦ Runs nightly, and on every alpha/beta push to google play
19. Level 3: We made it here, now what?
20. Upcoming big efforts
21. Modularizing the app • Prepare for multiple apps & decentralized teams • Improve build times • Prepare for App Bundles, Instant apps, downloadable features…
22. Twitch MVP pattern 2.0 (?) • Current pattern still holding strong 1+ year later • But can still be improved: ◦ Loosen coupling between Presentation and View layers ◦ Leverage ViewModel and LifecycleOwner from architecture components ◦ Automatic cleanup, easy way to persist Presenters Helium: Reactive MVP Framework. 100% Kotlin https://github.com/joaquim-verges/helium
23. Latency / Performance monitoring • Framework to automatically get latency metrics for every new screen • Daily/Weekly reports on latency changes • Frame rate monitoring tools and reports • Video performance monitoring tools and reports
24. Questions??