Sinatra is a companion app for the Canberra MyWay+ network made by locals for locals. Our goal is to make it easier to find route and stop information, offering a fresh, streamlined experience alongside the official app.
While Sinatra doesn’t include ticketing or full routing features, we focus on delivering the key info you need, fast and frustration-free.
Made with 🩷 in Canberra, Australia.
Universal Android APKs are automatically built for both the develop and main branch, the latest of which can be found here. The build named "debug" automatically points at the development environment, while "release" points at production. Note that release builds built on the develop branch may not be functional, while develop builds may quickly become out of date.
To build from source, the app requires the provisioning of a Google Maps API key on Android, as well
as a Firebase project with both Crashlytics and Remote Config enabled. The values gathered from
Remote Config can be found in RemoteConfigRepository.kt and most are optional, however api_url
must be set to a valid GTFS api endpoint. Likewise, to enable address resolution,
nominatim_api_url must also be provided.
Sinatra is a Compose/Kotlin Multiplatform app that targets both Android and iOS. The majority of the native implementation is centred around the maps UI, for which a cross platform abstraction has been developed on top of Android's Google Maps and iOS's Apple Maps.
The app uses two different APIs to provide content and address resolution respectively. Content (such as routes, stops, timetables, etc) are provided by a statically generated API based on the publically available GTFS data. The script for generating that can be found here. This script also generates the custom byte format used for in-app navigation.
To resolve addresses, Nominatim is used.
Both APIs are configured through Firebase Remote Config, although another implementation may be provided.
Routing is calculated on device and uses a simple Dijkstra algorithm backed by a stripped back Fibonacci Heap priority queue implementation (credit to Keith Schwarz, on which the implementation is based).
The algorithm explores a time-dependent graph that represents the entire transport network. In this implementation, each stop is represented by a single node and each pair of (route, heading) associated with the stop is a unique node. From there, edges connect each stop by "transfer" (walking, biking, etc) and each (route, heading) associated with a stop into trips with information about the conditions under which the edge is active.
The router uses a custom byte format to store the graph on disk, and lazily deserializes nodes and edges on demand. The specification for the byte format can be found here. Essentially, the network graph is always stored in memory as an unstructured byte array, with only some metadata being permanently deserialized. Node and edge objects are in fact facades with a pointer to a position in the byte array and knowledge of how to fetch fields from the array. When a field is requested, the necessary data is fetched, reconstructed, and then cached.
Compared to a previously experimented format that used a more conventional protobuf format, this implementation proved to significantly reduce memory pressure.
