abstreet/docs/onboarding.md
2019-12-03 11:42:02 -08:00

4.1 KiB

Onboarding for new developers

I recommend you glance through all the other docs listed on the front page to initially get the game running from source and understand how the map importing and traffic simulation modeling work.

Code organization

If you're going to dig into the code, it helps to know what all the crates are. The most interesting crates are map_model, sim, and game.

Constructing the map:

  • convert_osm: extract useful data from OpenStreetMap and other data sources, emit intermediate map format
  • gtfs: simple library to just extract coordinates of bus stops
  • kml: extract shapes from KML shapefiles
  • map_model: the final representation of the map, also conversion from the intermediate map format into the final format
  • precompute: small tool to run the second stage of map conversion and write final output
  • popdat: importing daily trips from PSRC's Soundcast model, specific to Seattle
  • map_editor: GUI for modifying geometry of maps and creating maps from scratch

Traffic simulation:

  • sim: all of the agent-based simulation logic
  • headless: tool to run a simulation without any visualization

Graphics:

  • game: the GUI and main gameplay
  • ezgui: a GUI and 2D rendering library built on glium

Common utilities:

  • abstutil: a grab-bag of IO helpers, timing and logging utilities, etc
  • geom: types for GPS and map-space points, lines, angles, polylines, polygons, circles, durations, speeds
  • tests: a custom test runner and some tests using it

Example guide for implementing a new feature

A/B Street's transit modeling only includes buses as of September 2019. If you wanted to start modeling light rail, you'd have to touch many layers of the code. This is a nice, hefty starter project to understand how everything works. For now, this is just an initial list of considerations -- I haven't designed or implemented this yet.

Poking around the .osm extracts in data/input/, you'll see a promising relation with route = light_rail. The relation points to individual points (nodes) as stops, and segments of the track (ways). These need to be represented in the initial version of the map, RawMap, and the final version, Map. Stations probably coincide with existing buildings, and tracks could probably be modeled as a special type of road. To remember the order of stations and group everything, there's already a notion of bus route from the gtfs crate that probably works. The convert_osm crate is the place to extract this new data from OSM. It might be worth thinking about how the light rail line gets clipped, since most maps won't include all of the stations -- should those maps just terminate trains at the stations, or should trains go to and from the map border?

Then there are some rendering questions. How should special buildings that act as light rail stations be displayed? What about the track between stations, and how to draw trains moving on the track? The track is sometimes underground, sometimes at-grade with the road (like near Colombia City -- there it even has to somehow be a part of the existing intersections!), and sometimes over the road. How to draw it without being really visually noisy with existing stuff on the ground? Should trains between stations even be drawn at all, or should hovering over stations show some kind of ETA?

For modeling the movement of the trains along the track, I'd actually recommend using the existing driving model. Tracks can be a new LaneType (that gets rendered as nice train tracks, probably), and trains can be a new VehicleType. This way, trains queueing happens for free. There's even existing logic to make buses wait at bus stops and load passengers; maybe that should be extended to load passengers from a building? How should passengers walking to the platform be modeled and rendered -- it takes a few minutes sometimes!

Finally, you'll need to figure out how to make some trips incorporate light rail. Pedestrian trips have the option to use transit or not -- if light rail is modeled properly, it hopefully fits into the existing transit pathfinding and everything, so it'll just naturally happen.