Developer guide

This guide is meant for people who want to work on A/B Street's code. In most cases as a user, even for importing new cities, you shouldn't need to bother with any of this.

Getting started

You will first need:

One-time setup:

  1. Download the repository: git clone

  2. Grab the minimal amount of data to get started: cargo run --bin updater -- download --minimal

  3. Run the game: RUST_BACKTRACE=1 cargo run --bin game --release. On Windows, set environment variables like this: set RUST_BACKTRACE=1 && cargo run --bin game --release

Development tips

  • Generated API documentation
  • Compile faster by just doing cargo run. The executable will have debug stack traces and run more slowly. You can do cargo run --release to build in optimized release mode; compilation will be slower, but the executable much faster.
  • Some in-game features are turned off by default or don't have a normal menu to access them. The list:
    • To toggle developer mode: press Control+S in game, or cargo run -- --dev
    • To warp to an object by numeric ID: press Control+j
    • To enter debug mode with all sorts of goodies: press Control+D
  • You can start the game in different modes using flags:
    • cargo run --bin game -- --dev data/system/us/seattle/maps/downtown.bin starts on a particular map
    • cargo run --bin game -- data/system/us/seattle/scenarios/downtown/weekday.bin starts with a scenario (which is tied to a certain map)
    • cargo run --bin game -- --challenge=trafficsig/tut2 starts on a particular challenge. See the list of aliases by passing in a bad value here.
    • cargo run --bin game -- data/player/saves/us/seattle/montlake/no_edits_unnamed/00h00m20.3s.bin restores an exact simulation state. Savestates are found in debug mode (Control+D) -- they're probably confusing for the normal player experience, so they're hidden for now.
    • cargo run --bin game -- --tutorial=12 starts somewhere in the tutorial
    • Adding --edits='name of edits' starts with edits applied to the map.

Downloading more cities

Most easily, you can download new cities using the GUI directly.

As data formats change over time, things in the data/ directory not under version control will get out of date. At any time, you can run cargo run --bin updater -- download from the main repository directory to update only the files that have changed.

You can also opt into downloading updates for more cities from the command line by editing data/player/data.json. If you want to add data for Leeds, GB, for example, you could edit that file so that it contains the following:

  "runtime": ["gb/leeds", "us/seattle"],
  "input": ["gb/leeds", "us/seattle"]

If you want to opt into absolutely everything: cargo run --bin updater -- opt-into-all > data/player/data.json

Building map data

You can skip this section if you're just touching code in game, widgetry, and sim. If you're just trying to import a new city, follow the user guide. This section is relevant if you're working on the code that imports maps and scenarios.


To run all pieces of the importer, you'll need some extra dependencies:

The importer pipeline

The import process is split into two main steps. When you run the importer, you specify what work you want to do, so understanding these steps is important. First is the --raw step, which:

  1. Downloads all input data needed for a map -- a .osm.pbf file from Geofabrik, at minimum
  2. Uses osmium to clip the large OSM file into smaller pieces
  3. Reads in all of the input, producing a RawMap file. This is an intermediate format that's useful for debugging, using the map_editor tool.

If you look in data/input/us/nyc/ (substituting the country and city, of course), you'll see osm and raw_maps directories. In places like data/input/us/seattle/, there are many other input files.

The second step, --map, transforms the RawMap to the final Map file used by all of the applications. These files live in data/system/us/nyc/maps.

Running the importer

The importer tool has many CLI flags not described here; --help is the best reference. But we can start with a few common examples.

After creating configuration files for a city, you can import it, running both the --raw and --map steps:

./ --raw --map --city=br/sao_paulo

If your city is split into many maps, you can operate on just one of them (the west boundary):

./ --raw --map --city=fr/paris west

If you want to regenerate a map with the latest OSM data, you need to first delete the OSM input files cached locally. Then running with --raw will download them again. Note the OSM files are downloaded from Geofabrik, so they may be up to one day out-of-date. You can additionally use osmupdate if needed.

rm -rfv data/input/fr/paris/osm/
./ --raw --map --city=fr/paris

Some more tips for running specific stages of the importer:

  • If you're modifying the initial OSM data -> RawMap conversion in convert_osm, you need ./ --raw --map.
  • If you're modifying map_model but not the OSM -> RawMap conversion, then you just need ./ --map.
  • If you're modifying the demand model for Seattle, you can add --scenario to regenerate.
  • By default, all maps are regenerated. You can also specify a single map: ./ --map downtown.
  • By default, Seattle is assumed as the city. You have to specify otherwise: ./ --city=us/detroit --map downtown.

Building contraction hierarchies for pathfinding occurs in the --map stage. It can take a few minutes for larger maps. To view occasional progress updates, you can run the importer with

RUST_LOG="fast_paths=debug/contracted node [0-9]+0000 "

You can also make the importer import a new city.

Understanding stuff

The docs listed at explain things like map importing and how the traffic simulation works.

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
  • kml: extract shapes from KML and CSV shapefiles
  • map_model: the final representation of the map, also conversion from the intermediate map format into the final format
  • map_editor: GUI for modifying geometry of maps and creating maps from scratch. pretty abandoned as of June 2020
  • importer: the entire import pipeline
  • cli: a collection of command-line tools to manage importing
  • updater: tool to download/upload large files used in the import pipeline

Traffic simulation:

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


  • game: the GUI and main gameplay
  • map_gui: common code to interact with map_model maps
  • widgetry: a GUI and 2D OpenGL rendering library, using glium + winit + glutin

Common utilities:

  • abstutil: a grab-bag timing and logging utilities
  • abstio: Reading/writing files on native/web
  • geom: types for GPS and map-space points, lines, angles, polylines, polygons, circles, durations, speeds


  • collisions: an experimental data format for real-world collision data
  • traffic_seitan: a bug-finding tool that randomly generates live map edits
  • tests: integration tests
  • santa: 15-minute Santa, an arcade game about delivering and zoning
  • parking_mapper: a standalone tool to help map street parking in OSM
  • osm_viewer: a standalone tool to render OSM in detail
  • fifteen_min: a standalone tool to explore 15-minute neighborhoods
  • popdat: use census data to produce traffic simulation input
  • traffic_signal_data: manual timing overrides for some traffic signals
  • piggyback: a small WebAssembly API to layer parts of A/B Street on top of Mapbox or other web maps

Code conventions

All code is automatically formatted using; please run cargo fmt before sending a PR.

cargo fmt can't yet organize imports, but we follow a convention to minimize conflict with what some IDEs do. Follow existing code to group imports: std, external crates, other crates in the project, the current crate, then finally any module declarations.

We also use to adhere to more Rust best practices. No need to run it for every commit, and sometimes we'll disable a warning. If in doubt, just ask on a PR.

See the testing strategy page.

Error handling

The error handling is unfortunately inconsistent. The goal is to gracefully degrade instead of crashing the game. If a crash does happen, make sure the logs will have enough context to reproduce and debug. For example, giving up when some geometry problem happens isn't ideal, but at least make sure to print the road / agent IDs or whatever will help find the problem. It's fine to crash during map importing, since the player won't deal with this, and loudly stopping problems is useful. It's also fine to crash when initially constructing all of the renderable map objects, because this crash will consistently happen at startup-time and be noticed by somebody developing before a player gets to it.

Since almost none of the code ever needs to distinguish error cases, use anyhow. Most of the errors generated within A/B Street are just strings anyway; the bail! macro is a convenient way to return them.


Prefer using info!, warn!, error!, etc from the log crate rather than println.

Adjust the log level without recompiling via the RUST_LOG env variable.

RUST_LOG=debug cargo run --bin game

This can be done on a per lib basis:

RUST_LOG=my_lib=debug cargo run --bin game

Or a module-by-module basis:

RUST_LOG=my_lib::module=debug cargo run --bin game

You can mix and match:

# error logging by default, except the foo:bar module at debug level
# and the entire baz crate at info level
RUST_LOG=error,foo::bar=debug,baz=info cargo run --bin game

For some special cases, you might want to use regex matching by specifying a pattern with the "/":

# only log once every 10k
RUST_LOG="fast_paths=debug/contracted node [0-9]+0000 " mike import_la

See the env_logger documentation for more usage examples.


Use, just running it on the binaries you build normally.