Restrict the schemas to valid values

[hugoledoux]

Discussed in https://github.com/cityjson/specs/discussions/186

^{Originally posted by **balazsdukai** November 17, 2023} I experimented with the popular [JSON Schema Faker](https://json-schema-faker.js.org/) tool to mock CityJSON data. I used the [combined v2.0 schema](https://3d.bk.tudelft.nl/schemas/cityjson/2.0.0/cityjson.min.schema.json) to fake values for the CityJSON members. It does generate a seemingly schema-valid CityJSON object, however many members have invalid values, for instance negative vertex indices in the boundaries. The reason for this is that in several cases the schema does not restrict the values sufficiently. Below is are the issues and potential fixes that I discovered from the generated CityJSON. - The version number is invalid. It should be restricted to the CityJSON version that is described by the schema. ```json "version": "3.4" ``` - CityObject family references point to nonexistent objects. This is probably not possible to prescribe solely with JSON Schema. ```json "CityObjects": { "et_": { "type": "BuildingInstallation", "parents": [ "ex fugiat", "sed veniam ut dolore" ], "children": [ "ipsum nostrud" ] } } ``` - Geometry LoD has invalid value. The `lod` should be restricted to the allowed values. ```json "lod": "9" ``` - Vertex indices in the boundaries have negative values. The vertex indices should be restricted to positive integers. [This example](https://json-schema.org/draft/2020-12/json-schema-core#name-schema-re-use-with-defs) shows that this should be possible. ```json "boundaries": [ [ 44174456, 49920625, -95552225, -13564149 ] ] ``` - Semantic object does not have `type`. In the example below, the second semantic object does not have a `type` member, which is required according to the specs. ```json "semantics": { "surfaces": [ { "type": "ClosureSurface", "laboris_4": "sit enim in", "amet62": true }, { "deserunt_af": 18338638, "quis_2": -44755250, "in_e": 51750499, "adb_": "enim irure", "velit__": 73779662, "tempor_73": false } ] } ``` I don't see drawbacks from applying these restrictions where possible, but maybe others see it differently?

[hugoledoux]

I fixed the 4th above (semantic surfaces) there: https://github.com/cityjson/specs/commit/5544e1bd0c54c508603ec239938d78e07334e963

in a branch for v2.0.1

[hugoledoux]

Point 1 above (version of CityJSON) has been fixed in 1dbd85685dcf7aa5709b48f00b1eccf79d0fdd45

[hugoledoux]

@balazsdukai for the #3 above ("The lod should be restricted to the allowed values"), I am not sure we want to do this... The specs state:

The value must be a string with the LoD identifying the level-of-detail (LoD) of the geometry. This can be either a single digit (following the CityGML standards), or "X.Y"-formatted if the improved LoDs by TU Delft are used.

If we restrict, we go up to what? LoD3.3? What if someone wants LoD4.2?

Maybe @fbiljecki has something smart to add here?

[fbiljecki]

If it follows the two standards, then it is better to constrain it with their possible values and ranges, that is, 3.3 being the maximum in the latter case.

[hugoledoux]

this is the ones that would be accepted with CityGML3 (no LoD4 here)

  "Lods": {
      "enum": [
        "0",   "1",   "2",   "3", 
        "0.0", "0.1", "0.2", "0.3", 
        "1.0", "1.1", "1.2", "1.3", 
        "2.0", "2.1", "2.2", "2.3", 
        "3.0", "3.1", "3.2", "3.3"
      ]
  }

[balazsdukai]

What if someone wants LoD4.2?

Then it would be an extended CityJSON file, with its own schema that allows LoD4.2 I think. :+1: for 29dfdc4

[hugoledoux]

Hmmm, no that would mean no geometry types since "lod" is a child of "geometry" and this is not possible to modify.

But CityJSON is CityGML (thus only 0-1-2-3) and the TUDelft-LoDs, so restricting is fine. If you want more than you can use IFC I guess