This is a new version of ltn.abstreet.org. It doesn't have all features from v1 yet, but it solves many problems with the original. Please try both versions and give feedback.
You'll need: npm, wasm-pack, and cargo.
cd web
, and then:
npm ci
to install dependencies (ci
to make sure the versions inpackage-lock.json
are used)npm run wasm
to rebuild the Rust backend- vite doesn't automatically rebuild when you edit things
npm run dev
to run locally- Changes to the Svelte/CSS usually auto-reload in your browser
npm run fmt
to auto-format codenpm run check
to see TypeScript errors
The tool operates on fixed study areas, generated from clips of OSM data. Custom areas imported by the user are always retrieved from the Overpass API, with the latest OSM data. For speed and for deterministic tests, there are also "built-in" study areas, consisting of pre-clipped osm.pbf files that do not automatically use the latest OSM data. These are manually managed and hosted by Dustin on assets.od2net.org.
If you're developing locally, you can avoid hitting od2net.org by setting up
two directories in web/public/
: osm
and boundaries
. If
web/public/osm/areas.json
exists, then the Svelte app will load from
localhost, not od2net.org. You can copy from od2net.org to set this up,
choosing what study areas to cache:
AREAS="bristol edinburgh strasbourg ut_dallas"
cd web/public
mkdir boundaries osm
cd osm
wget https://assets.od2net.org/severance_pbfs/areas.json
for x in $AREAS; do
wget https://assets.od2net.org/severance_pbfs/$x.pbf
done
cd ../boundaries
for x in $AREAS; do
wget https://assets.od2net.org/boundaries/$x.geojson
done
(Don't spend too much time looking after these particular tests; they've been helpful to spot unexpected changes in calculating existing modal filters, but they're often just noisy when cell geometry slightly changes.)
The tests
directory has some diff-based tests. cd backend; cargo test --release
will run them. You'll need to follow the instructions above to
ensure you have bristol
and strasbourg
areas set up. To accept the diffs,
just commit the changed files.
There's a few ways to understand / verify the diffs.
First, you can manually try the tool
before and after (running locally). You
can load a .geojson
file from tests/
as a project, then click the
neighbourhood boundary (Bristol has two tests, Strasbourg just one). Then just
visually compare things -- cells, shortcuts, and existing filters.
You can also try just diffing the output GeoJSON file as text. They're stored
without pretty-printed newlines/indentation to save size in git, but you can do
something like this to view, assuming you have jq
and meld
(or another diff
tool):
function json_diff {
cat $1 | jq > /tmp/after.json
git show HEAD^:./$1 | jq > /tmp/before.json
meld /tmp/before.json /tmp/after.json
}
cd tests/output
json_diff bristol_west.geojson
Finally, if you have the before and after output GeoJSON file (json_diff
creates /tmp/before.geojson
and /tmp/after.geojson
), then you can try
exploring in GeoDiffr. You'll want to
"remove unchanged features", then try to understand the remaining changes. This
tool is not very sophisticated; it's only helpful sometimes when filters
change, but it's less useful if big cell polygons change.
The Rust backend
crate gets compiled to WASM, with generally type-unsafe APIs
in lib.rs
. On the frontend side, wasm.ts
wraps these APIs in nicer TS APIs.
The Svelte frontend calls APIs in the main thread (though moving to web workers
later is a possibility). The "important" state is kept in the backend, while
the frontend has more ephemeral UI state. The main state is serialized as
GeoJSON and kept in the user's browser local storage. The frontend is
effectively a single page app, with the Mode
toggling between different
pages.