[jravani]

Part of #210: the Netherlands, after Germany in #211.

use_type and residential were Germany only: they map a source function code, and Dutch 3D data carries none. BAG records the use (gebruiksdoel) and floor area (oppervlakte) of every unit (verblijfsobject) of a building, which gives the building's use, including mixed use.

Change

• Attribute datasets: data/attributes/<country>/<dataset>/ holds an attribution.json (validated and credited like a 3D dataset's) and building_use.csv:

  Column	Content
  object_id	The building's object id as the 3D data carries it
  use_code	One provider use code; several codes are several rows
  floor_area	m2 the code covers, >= 0

• Import (-create-db, -import-data, on-request runs): each dataset's rows in city2tabula.building_use are replaced via CopyFrom in one transaction. The table is created if missing, so a database built earlier gains it on its next import. A malformed file stops the import with file and line; every file is parsed before any row is written. dataset_id must be unique across 3D and attribute datasets.

• Serving: where a building's function code gives no use, the buildings queries read its rows as one JSON column. The use type covering the most floor area wins, with equal areas going to the first by name. residential is true when any row is a dwelling. The dataset is added to attributions. A database without the table returns null, as before.

• Mapping internal/onrequest/usetype/netherlands.csv:

  • the 11 BAG gebruiksdoelen;
  • plus geen verblijfsobject, which the extract gives a building without units (a shed, garage or barn): ancillary;
  • bijeenkomstfunctie, celfunctie and sportfunctie: public;
  • logiesfunctie: hospitality.

• Deployment: environment/http/docker-compose.yml mounts data/attributes, and .gitignore excludes it.

• Docs: data/README.md, the attribution page, OpenAPI (use_type, attributions), CHANGELOG.md.

The response shape is unchanged; Dutch buildings now carry values.

Tests

• Unit:

  • TestUseTypeFromUnits: largest floor area, codes of one use type adding up, ties, unmapped codes, building without units.
  • TestParseBuildingUse, TestDiscoverAttributeDatasets: header, joined codes, bad area, missing file or folder.

• Integration: TestServe_UseTypeFromBuildingUnits on the real Dutch seed covers the import, the served use type, residential, credits and restaging. With the fallback removed it fails (use_type null, residential <nil>, want retail and true).

• go vet ./... and go test ./... pass. The integration suites for importer, process, onrequest, db and api pass.

• End to end on the Loenen and Heeten BAG extract (20,039 rows) and 3DBAG tiles, through create-db, extract and the API:

  use_type	Buildings	Median footprint
  ancillary	4,482 (40.9%)	35 m2
  residential	4,013 (36.7%)	95 m2
  hospitality	1,759 (16.1%)	72 m2
  other uses	562 (5.1%)
  null	131 (1.2%)

  The 131 without a use type have left BAG since the 3DBAG snapshot. 177 buildings hold dwellings under another main use. All 10,947 buildings come back in one bbox request in 0.75 s.

[codecov]

Codecov Report

❌ Patch coverage is 78.77095% with 38 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/importer/building_use.go	78.72%	10 Missing and 10 partials ⚠️
internal/onrequest/query.go	75.75%	4 Missing and 4 partials ⚠️
internal/importer/dataset.go	66.66%	2 Missing and 2 partials ⚠️
internal/config/region.go	75.00%	1 Missing and 1 partial ⚠️
internal/db/setup.go	0.00%	1 Missing and 1 partial ⚠️
internal/onrequest/usetype.go	91.66%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!