Schema descriptor with path property raises a "resource-error"

[pierrecamilleri]

Overview

If a schema descriptor has a path property, validating the descriptor with python code returns a "resource-error". At the command line, the file is considered valid.

For the context, the path property is defined for a data resource, but not in the Table Schema specification. It is also not in the Schema class properties.

I have no clue what is going on here.

Reproduce

Schema descriptor in file "schema.json" :

{
  "name": "schema",
  "path": "abc",
  "fields": [
    {
      "name": "id",
      "type": "integer"
    }
  ]
}

import frictionless

schema = frictionless.Schema("./schema.json")
descr = schema.to_descriptor()
print(frictionless.validate(descr, type="schema"))
# {'valid': False,
#  'stats': {'tasks': 0, 'errors': 1, 'warnings': 0, 'seconds': 0},
#  'warnings': [],
#  'errors': [{'type': 'resource-error',
#              'title': 'Resource Error',
#              'description': 'A validation cannot be processed.',
#              'message': "The data resource has an error: [{'name': 'id', "
#                         "'type': 'integer'}] is not of type 'integer' at "
#                         "property 'fields'",
#              'tags': [],
#              'note': "[{'name': 'id', 'type': 'integer'}] is not of type "
#                      "'integer' at property 'fields'"}],
#  'tasks': []}

What I tested

• Validating at the command line frictionless validate --type=schema --json schema.json returns VALID
• Validating with file path returns 'valid': True
• Changing the name of path to patch does not reproduce the behavior.
• Changing the path value to an URL does not solve the issue.

[pierrecamilleri]

Short investigation shows Resource is the culprit : it has a path property and a fields: Optional[int] = None which explains the error.

[pierrecamilleri]

Wow there is some magic going on there.

So :

• resource Factory class calls Detector.detect_metadata_type(path, format=options.get("format")) (with path = source a couple of lines above), despite the type being explicitely set.
• The detect_metadata_type uses METADATA_TRAITS to select the type, and ...
• Detects the Resource type if path is present : 

    ...
     "resource": {
        "names": ["resource.json", "resource.yaml"],
        "props": ["path", "data"],
    },
    ...