Transfer patterns.

What a pattern is

The stations a journey calls at. Written out in full, the patterns from Norwich to Liverpool Street are:

$ NRW,LST              direct$ NRW,CBG,LST          changing at Cambridge$ NRW,CBG,ELY,LST      changing at Cambridge and Ely

The codes are the stop_code of the feed — CRS codes — so the two are read together: stops.txt is what turns NRW back into Norwich and into somewhere with a platform and a position. Patterns are found a day at a time and the published file holds a week of them, so it covers every shape of service the timetable has.

What is actually in the file

Not that. The commas above are there to read; the file has no separators, because every code is three characters and a line is cut up rather than split. And because the patterns are sorted, each line records only what it adds to the line above it — so a station is written once and never again.

The first character says how many leading stations to take from the previous line, counting up from 0. These are the opening lines of NRW.gz, and what each one means:

$ 0NRWACL      NRW ACL$ 1ALF         NRW ALF$ 1ALK         NRW ALK$ 1AUD         NRW AUD$ 2BIS         NRW AUD BIS$ 3ESM         NRW AUD BIS ESM

So NRW appears exactly once in a file of tens of thousands of patterns that all begin with it, and 2BIS is four bytes for a journey of three stations. The count carries on past 9 into the characters above it, because a pattern can run to fourteen stations. Nothing marks a pattern that a longer one runs through —NRW AUD is a journey in its own right as well as the start ofNRW AUD BIS — because every line is exactly one pattern.

The whole thing is then gzip compressed, so decompress it before reading — which a browser will do for you with DecompressionStream. Rather than unpicking any of this yourself, transfer-pattern-planner reads it back for you.

They are built after the feed they describe is published, so a release without them is a night whose scan did not finish rather than a feed with something wrong.Using this data says more.

Reading one station

Each file is named for its station and holds every pattern departing it, so nothing has to keep an index of where they are. It is a plain GET, which a browser and a cache can each make sense of on their own.

$ curl -O https://planarnetwork.github.io/gb-transit/transfer-patterns/NRW.gz

Which stations there are is transfer-patterns.json— every code, its name, its file and how large it is. Nothing needs it to fetch a station, since a station names its own file, but something does need it to know what to ask for. It is generated from the files the build published, so it cannot name one that is not there.