Repository
https://github.com/flutter/plugins
New Features
Pull request: https://github.com/flutter/plugins/pull/695
Implementing the cloud based API in ML Kit for Firebase.
The Flutter implementation of ML Kit for Firebase was only supporting the offline capable features. This pull request added the first cloud API integration to the plugin.
GitHub Account
https://github.com/DennisAlund
Background
I spoke at a conference where another speaker did a live coding of "Hot dog/not hot dog". I instantly wanted to try it myself. But I am not a iOS developer and my Android skills are rusty. But I knew that Flutter is known for ease of learning and speed of development, I decided to explore it.
A couple of hours later I already had a basic app up and running with the ML Kit flutter plugin. Only to learn that the on-device recognition wasn't accurate enough for my taste, and the plugin didn't support integration with the cloud API, at the time of writing.
So, next step was to dive deeper into Flutter, to learn how to develop or extend a plugin, so that I could have the functionality that I wanted.
Instead of developing the hot dog/not hot dog app that night, I made my first community contributon to Flutter.
Anatomy of Flutter plugins
Flutter plugins are dart packages that can either be purely Dart code or implementing integration with the specific platform. I.e. implementing support for native APIs that are not part of the Flutter core. The Firebase libraries are examples of this, including the ML Kit Plugin.
Flutter plugins that integrate with the underlying platform are doing so via an asynchronous message channel that bridges between the Flutter system and plugin code written natively for respective platform. This makes the relationship between Flutter and the platform very dynamic and cleanly separated. The asynchronous nature is ensuring that there are no blocking calls that affects the Flutter UI thread.
Image source: https://flutter.io/platform-channels/
The method channel API supports simple parameter values that can be serialized to JSON (scalar values, lists and maps). So we don't pass large chunks of data across that bridge. In the case of the cloud vision API we're actually just passing the path of the image on the native file system, which is later used in the platform specific native plugin code to make an instance of the corresponding class and make the API call.