I literally implemented this in my toy project with Claude. The prompt was very dumb with something like "integrate the Metro dependency injection framework" and it did all the right things except for creating its own AppScope class instead of using the built-in class. It worked for me.
One underrated improvement Metro brings is its easy setup. No more configuring KAPT, runtime modules, keeping versions in sync, etc. You apply the Gradle plugin, setup a tiny graph and the rest is driven through constructor injection, @ContributesBinding and its other features. This was all I needed
Square did the same 7 years ago. This module structure was created with Bazel in mind, it's also mentioned in the original design doc. The goal was to scale horizontally. But then Bazel didn't happen 🙃 This talk got me a little excited again: www.youtube.com/watch?v=cSYe...
I wish we'd consider Bazel more seriously too, but it comes with other downsides. Build times are not the only metric.
With this structure we have now 7,000 Gradle modules in our codebase. Configuration time is painful. But with configure-on-demand, configuration caching and other strategies like using dev apps it's somewhat mitigated.
I agree with you. Modularizing by type instead of functionality often leads to issues you mentioned. Other bad examples are putting all feature flags in an enum or all Android Activities into the same module.
But I think Jesse's point here was showing how to break circular dependencies.
No, quite the opposite in my opinion. The UI layer is just one consumer of the navigation state. If I run my app headless, then there's no UI layer.
On this scale, what does version 3.0 alpha 8 for 1.5 years mean?
Actually, OpenJDK 25 is 18% faster than OpenJDK 17 and 21 and a whopping 84% faster than OpenJDK 11. Upgrade now!
I recently migrated our codebase away from Ktlint to Ktfmt and Detekt. I summarized the reasons and migration steps in a blog post: engineering.block.xyz/blog/adoptin...
Kotlin compiler plugins are amazing!
✅ Expressive syntax
✅ Compile-time checked
✅ Fast at runtime
For real? We had so many power outages and they never offered anything.
I proposed 95% of what Metro became 4 years ago. In particular I called out error messages. I didn't get funding, perhaps I undersold the value.
In hindsight I'm happy. @zacsweers.dev did such a good job and pushes the boundaries of the compiler. 4 years ago this would have been worse.
We're getting there, but it'll take another few months. The migration steps are almost identical for us. There are quite a few extra hurdles for historical reasons, like migrating quite a few Java classes to Kotlin first.
Cash Android has recently switched to Metro, a new, fast dependency injection framework built for K2. That was a fun project and we're really happy with the results! Sharing details on how we approached the migration to help teams adopt Metro: code.cash.app/cash-android...
This is fantastic and @zacsweers.dev deserves all the praise but it also is gonna give you a false idea of what Z is like as a maintainer and author, what the project is like, and how you might achieve something like this yourself
I adopted to always type "assDeb" for assembleDebug. Please don't judge me during pairing sessions.
I'm quite proud of what I accomplished with my team at Amazon and will miss them. I'll keep working on App Platform as much as I can thanks to open source! amzn.github.io/app-platform/
After three wonderful years at Amazon I'm happy to share that I started this week at ... Square / Block again on my old team 🪃 I'm excited to think about developer productivity more deeply again.
Metro is conceptional very similar to Dagger + Anvil and kotlin-inject-anvil, which made this migration easy. I expect most new development to happen for Metro, so I wouldn't use kotlin-inject-anvil for new projects anymore.
Our KMP framework App Platform now supports Metro out of the box: github.com/amzn/app-pla...
@zacsweers.dev is pushing the boundaries with Metro and there are still some Kotlin compiler bugs that need to be resolved, but on the JVM / Android it works.
I've been using ktfmt over KtLint for a few months for some projects. It's a blessing and so much easier to deal with. I forgot how much I missed 2 space indent in Kotlin.
facebook.github.io/ktfmt/
We published a new version of App Platform with back button support. We also added recipes for saving presenter state, app bar integration and backstack navigation.
Working with Compose in the business logic (Molecule) and UI (Compose Multiplatform) is so much fun.
amzn.github.io/app-platform...
Today I learned that Cleanse from Square was deprecated. I haven't tried Needle yet, but it makes the same promises as Dagger.
github.com/uber/needle
Singleton in the scope by default, based on the large codebases I've seen. I've written @SingleIn way too often and explained way too often why a class doesn't need to be a Singleton. I want this to be explicit.
Anvil was quite successful inside and outside of Square, but now it's time for the next big improvement. Metro is 95% of what I pitched as Dagger 3 within Square. I'm excited.
Version 0.34.0 of the gradle-maven-publish-plugin finally comes with configuration cache support. Dropping support for Sonatype OSSRH unblocked us from waiting for Gradle to resolve the underlying issue on their side.
github.com/vanniktech/g...
