Add basic valid and invalid tests for the json schema

[jorisvandenbossche]

Closes #135 (rework of #134 to focus on json schema for now)

[jorisvandenbossche]

cc @m-mohr do you think this is a sensible approach to test the json schema? (ensure it passes when it should and errors when it should)
Feedback would be very welcome (I never did this before)

[m-mohr]

I'll have a look in the next days :-)

[jorisvandenbossche]

OK, I updated this PR for the latest changes in the spec and updated to use pytest. This should be ready for review.

[m-mohr]

Looks reasonable to me. Why commit the test files when you can always generate them on the fly though? May lead to PRs where people update them by hand...

[m-mohr]

I'd like to propose some additional checks:

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["invalid_column_object"] = {"foo": "bar"}
invalid_cases["invalid_column_object"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = [0,0,0]
invalid_cases["3_element_bbox"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = [0,0,0,0]
valid_cases["4_element_bbox"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = ["0","0","0","0"]
invalid_cases["bbox_invalid_type"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = [0,0,0,0,0]
invalid_cases["5_element_bbox"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = [0.0,0.0,0.0,0.0,0.0,0.0]
valid_cases["6_element_bbox"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["bbox"] = [0,0,0,0,0,0,0]
invalid_cases["7_element_bbox"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"][""] = metadata["columns"]["geometry"]
invalid_cases["empty_column_name"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["primary_geometry"] = ""
invalid_cases["empty_primary_geometry"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["geometry_types"] = ["Point", "Point"]
invalid_cases["geometry_type_uniqueness"] = metadata

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["geometry"]["geometry_types"] = ["PointZ"]
invalid_cases["geometry_type_missing_space"] = metadata

Some of these may fail with the existing JSON Schema and may need to be added in #131, where we can fix this.
I could also add them all in #131. Now that #131 was merged, you should be able to just add them here.

[jorisvandenbossche]

@m-mohr thanks for those additional cases! This one I don't fully understand what it is testing for:

metadata = copy.deepcopy(metadata_template)
metadata["columns"]["primary_geometry"] = ""
invalid_cases["empty_primary_geometry"] = metadata

[jorisvandenbossche]

This one is still failing (it is xfailed for now to have the tests passing):

E           AssertionError: This is an invalid GeoParquet file, but no validation error occurred for 'missing_columns_entry':
E           {
E             "geo": {
E               "columns": {},
E               "primary_column": "geometry",
E               "version": "0.5.0-dev"
E             }
E           }

But I suppose that should be easy to solve to require a minimum length of one for columns?

[m-mohr]

Ah, I see. We can fix that, sure. Just add "minProperties": 1, in the columns schema.
By the way, the written spec doesn't forbid an empty column object.

[m-mohr]

Example:

    "columns": {
      "type": "object",
      "minProperties": 1,
      "patternProperties": {
        ".+": {

[m-mohr]

Or use/merge #158.

[m-mohr]

@jorisvandenbossche In #153 we require a minimum length of 1 for the primary geometry and I'm trying to check whether the schema fails for an empty primary geometry.

But I just now realize that it's wrong and should be:

metadata = copy.deepcopy(metadata_template)
metadata["primary_geometry"] = ""
invalid_cases["empty_primary_geometry"] = metadata

[jorisvandenbossche]

OK, that one is already included!

[tschaub]

Would there be support for moving the test_json_schema.py script to the scripts directory? It feels like that directory has the most complete story around managing python dependencies.

[jorisvandenbossche]

I wouldn't mind that, certainly now it is only a single file (I am also already making use of the python dependencies that are installed for the scripts to run the tests in CI)

[tschaub]

I moved this script in #159 and updated it to read the version constant from the schema file. So the tests now assert that metadata without a version or with a version that is different than the schema version is invalid.