Gå til hovedinnhold

C-12: Places

Summary​

A geo axis (C-02) holds place ids from the project's gazetteer. The gazetteer is the built-in places, which are the world, the UN M49 regions, every ISO 3166-1 country and every ISO 3166-2 subdivision, plus the places the recipe declares under places:: a city, a neighbourhood, a metropolitan area, a sales territory. Each place sits within one or more places or is made of places. Containment is computed from these two declarations only. A rule scoped to Norway applies to content at Oslo because the recipe declares Oslo within the county no-03, and the built-in data puts no-03 within Norway.

core/place holds the gazetteer and the containment rules over it. The recipe loader, the axis space, the operation log, the terms store, the host and the atlas all call it. The map in @neokapi/context-atlas applies the same rules to the place graph in the atlas document, and a conformance fixture that the Go and TypeScript tests both run keeps the two implementations in agreement. A place's shape is used only for drawing. The map draws built-in places from Natural Earth outlines, and a declared place from the GeoJSON file it names.

Context​

The markets a project writes for often span several administrative levels. A campaign can run in one city and one of its neighbourhoods. A sales territory can cover five countries, and a delivery area can cover a metropolitan area across two counties. A rule recorded for one of these places has to apply to content written for the places inside it. Two rules whose places overlap have to be detected as overlapping.

Containment could be computed from geometry, by testing whether one outline lies inside another. The result would then depend on the boundary data, its version and simplification, a geometry library, and the presence of the files. Declared containment gives the same result on every machine, offline, for the cost of a map lookup. Every change to it appears in the recipe's diff, where a reviewer sees it beside the content it affects.

Decision​

Nordicsnordics · territoryGreater Oslogreater-oslo · metroWorldworld · worldEuropeeurope · regionNorthern Europenorthern-europe · regionNorwayno · countrySwedense · countryDenmarkdk · countryFinlandfi · countryIcelandis · countryOslono-03 · subdivisionAkershusno-32 · subdivisionOslooslo · localityGrünerløkkagrunerlokka · neighbourhoodsits withinmade ofdeclared in the recipe

The places of the KapiMart sample. Built-in places form a tree under the world; a declared place sits within places or is made of them.

A place​

type Place struct {
    ID       string            // a slug, matched byte for byte: "oslo", "no-03", "no", "northern-europe", "world"
    Name     string            // English display name, from CLDR for a built-in place
    Names    map[string]string // display names per BCP 47 locale, for a declared place
    Type     Type
    Within   []string          // the places it sits inside
    Of       []string          // the places it is made of
    Refs     []string          // ids in other gazetteers: "wikidata:Q585", "geonames:3143244"
    Centroid *LatLon           // [lat, lon], for drawing
    Shape    string            // a declared place's GeoJSON file, project-relative
    Declared bool
}

A place sets exactly one of within: and of:. Type is a named kind of place rather than a numbered level, because administrative levels differ between countries:

TypeHolds
worldthe one place world
regiona UN M49 region, subregion or intermediate region
countryan ISO 3166-1 country or territory
subdivisionan ISO 3166-2 subdivision, built in, or declared for a municipality ISO 3166-2 does not list
territorya market or sales territory, usually made of places; the default for a declared place with of:
metroa metropolitan area, usually made of places
localitya city, town or village
districta borough or city district
neighbourhooda neighbourhood within a locality or district
postal-areathe area of a postal code
areaany other area; the default for a declared place with within:

Type.Rank orders the types as the table lists them, widest first. Place lists sort by it, so a country comes before its subdivisions.

Names are for display. Matching compares ids only, so renaming a place changes no content and no rule.

The built-in gazetteer​

place.Builtin() is shared and immutable:

  • world, which contains every place.
  • The UN M49 regions, subregions and intermediate regions, by slug (northern-europe, sub-saharan-africa, latin-america-and-the-caribbean), each within the region above it, a continent within the world.
  • Every ISO 3166-1 country and territory, by its alpha-2 code in lower case, with its English short name, an approximate centroid, and the M49 region it sits in: the intermediate region where M49 names one (ke within eastern-africa), else the subregion. Antarctica sits directly within the world.
  • Every ISO 3166-2 subdivision, by its code in lower case with the hyphen (no-03, us-ca, gb-sct), with its English name and the place it sits in, including second levels: fr-69m sits within the region fr-ara, which sits within fr.

The subdivisions come from Unicode CLDR at a pinned release (release-48-2): common/validity/subdivision.xml for the ids CLDR marks regular, common/supplemental/subdivisions.xml for containment, and common/subdivisions/en.xml for the names. ISO 3166-2 changes that the pinned release does not record yet are listed in core/place/internal/gen/amendments.tsv, one line per added or retired id, each naming its source. The generator (go generate ./core/place) downloads the three files, applies the amendments, and writes core/place/subdivisions.tsv, which is committed and embedded. The generator fails on an amendment that the pinned release already contains, so a release bump also removes that amendment line. The CLDR data is under the Unicode License v3, which ships beside it as core/place/UNICODE-LICENSE.txt. The file has that name because a file named LICENSE marks a licence boundary in this repository.

Every built-in place except the world sits within exactly one place, so the built-in places form a tree. A test checks the data: the file parses, ids are unique, every parent resolves, and no subdivision id collides with a region slug or a country code.

Declared places​

A recipe adds places under the top-level places: key, beside axes::

places:
  oslo:
    name: Oslo
    type: locality
    within: no-03                     # one id or a list
    centroid: [59.91, 10.75]
    refs: [wikidata:Q585, geonames:3143244]
  grunerlokka:
    name: Grünerløkka
    type: neighbourhood
    within: oslo
  nordics:
    name: Nordics
    of: [no, se, dk, fi, is]          # type defaults to territory
  greater-oslo:
    name: Greater Oslo
    type: metro
    of: [no-03, no-32]

axes:
  market:
    kind: geo
    label: Market

place.New(declared) builds the project's gazetteer on top of the built-in one and validates every declared place. Each error names the recipe key (places.<id>...). The rules are:

  • the id is a slug that no built-in place uses;
  • the place sets within: or of:, and not both;
  • every id it lists is a place;
  • no chain of within: and of: leads from a place back to itself;
  • the type is one a recipe may declare;
  • names are keyed by locale;
  • refs use a known scheme (wikidata, geonames, nuts, osm, wof), with a value in that scheme's format;
  • the centroid is a valid latitude and longitude;
  • shape is a .geojson or .json path relative to the project and inside it.

The recipe load then checks that the shape file exists. The project file reference lists the error messages.

KapiProject.Gazetteer() returns the gazetteer built at load, and KapiProject.AxisSpace() is axis.NewSpace(p.Axes, p.Gazetteer()), so every caller that interprets a coordinate uses the project's places.

Containment and overlap​

Gazetteer.Contains(outer, inner) applies these rules to the declarations:

  • every non-empty id contains itself, known or not;
  • the world contains every known place;
  • a place made of places sits within outer when every one of its members does;
  • a place made of places contains inner when one of its members contains it, or when inner is declared within it;
  • otherwise inner sits within outer when one of the places it sits within does.

An id that is not in the gazetteer contains only itself and sits within only itself. A rule recorded at a place that another checkout declares therefore still applies to content at that same id, and to no other place.

Gazetteer.Overlaps(a, b) is true when one place contains the other, when a member of a place made of places overlaps the other place, or when a declared place sits within both. The built-in places form a tree, so two built-in places overlap only when one contains the other. Only a declared place with several within: parents connects two branches. Suggestions at greater-oslo and at oslo overlap, and so do suggestions at nordics and at no-03. Suggestions at oslo and at bergen do not.

For a geo axis, axis.Space.Contains and axis.Space.Overlaps call these two functions. Scope coverage of a point (C-11), the terms-store match of a concept at a point (terms.AtPointIn) and the overlap test between records in the operation log's fold all go through them.

Paths, ancestors and children​

A surface that shows a place in context needs a single chain from the world down to it. Gazetteer.Path(id) follows each place's first within: parent. For a place made of places, the parent is the nearest place on its first member's path that contains every member: northern-europe for nordics, and no for greater-oslo. Path("grunerlokka") is [world europe northern-europe no no-03 oslo grunerlokka] and Path("nordics") is [world europe northern-europe nordics].

Ancestors(id) lists every place that contains id, nearest first, including each place made of places that contains it. Children(id) lists the places whose path parent is id, plus the members when id is a place made of places, sorted by name. The atlas uses these to collect the places a map can drill into.

Looking a place up​

Gazetteer.Search(query, limit) matches places whose id starts with the query, then places whose name starts with it, then places whose name contains it. It ignores case and accents, so grunerlokka finds Grünerløkka, and a declared place's names in other locales count as names. Within each tier the exact id comes first, then wider kinds of place, then ids in order.

A person or an agent looks up an id with kapi context places [query] [--within <id>] (host.App.ContextPlaces). With a query, the command lists the matches, limited to places inside --within when it is given. With only --within, it lists the places directly below that place. With neither, it lists the declared places. Each line gives the id, the name, the type and the path. If a coordinate value is not a place id, the recipe fails to load and the error carries a hint from place.Hint. The hint is the lower-case or hyphenated spelling when that is a place id, otherwise up to three matches and the kapi context places command for finding more.

A locale can name a place​

place.OfLocale(tag) returns the place a locale tag identifies. That is the subdivision in a -u-sd- extension (nb-NO-u-sd-no03 is no-03), else the country or M49 region of the region subtag (fr-CA is ca, es-419 is latin-america-and-the-caribbean). A tag with neither returns no place. The atlas uses it to derive the region axis of every record that has a locale. A recipe's target_languages may use the -u-sd- extension.

The same rules in Go and TypeScript​

The map filters and counts in the browser, so @neokapi/context-atlas implements the same rules over the atlas's places (placeIndex in model/places.ts: contains, overlaps, path, children). core/place/testdata/containment.json is a small gazetteer with the cases for containment, overlap and paths. The Go test and the TypeScript test both run every case, and the Go test also checks that each built-in place in the fixture matches the built-in gazetteer's id, type and parent. A change to the rules updates both implementations and the fixture in the same commit.

Drawing a place​

The atlas document carries the places a map needs (Context atlas): every place a record names, everything that contains them, the members of each place made of places, the declared places, and the children of each country and subdivision in that set. A declared place with a shape: carries its outline as GeoJSON geometry. If the shape file cannot be read, is not a single Polygon or MultiPolygon, or is larger than 1 MiB, the outline is omitted and the atlas adds a note naming the recipe key.

The built-in outlines ship in @neokapi/context-atlas (src/geo/): Natural Earth (public domain) at a pinned release, the admin-0 countries in one file and each country's admin-1 subdivisions in a file of its own, keyed by place id and loaded when the map drills into that country. A script fetches the pinned files, checks their SHA-256 digests and writes the TopoJSON (vp run @neokapi/context-atlas#build-geometry). Where Natural Earth carries a subdivision under an older code, or only as the parts of a current one, scripts/geometry-successors.json composes the current id from those parts and cites the source of each row: the boundary or code change, or the parent relation. The package's src/geo/README.md lists the sources, the licence and the ids that have no outline.

The map draws a declared place from its geometry, else as a circle at its centroid, else lists it beside the map. It flags a declared place whose centroid falls outside the outline of the place it sits within, which usually means the within: is wrong. Drawing has no effect on containment.

Consequences​

  • A project names its own markets at any grain, from a neighbourhood to a sales territory, and a rule kept for a wider place holds in every place inside it without a copy per place.
  • Containment depends only on the recipe and the pinned data, so it gives the same result on every machine and in every surface, with or without outline files.
  • A wrong within: changes where rules apply. The change appears in the recipe's diff, and the map flags a centroid outside its parent.
  • Updating the built-in subdivisions is a reviewed change to the generator's pinned release and the amendments file. The diff of the generated table lists each changed id.
  • The map's outlines lag ISO 3166-2 wherever Natural Earth does. The map lists a place without an outline instead of drawing it. Containment is unaffected.

See also​