Skip to content

OpenApiVisitorBase PathString is not correct for v2 documents #2806

@davidrea-MS

Description

@davidrea-MS

Describe the bug
The OpenApiVisitorBase.PathString is documented as 'Pointer to source of validation error in document', but when used with OpenApiWalker and the built-in validation system, it does not produce correct pointers when the input document is in the v2 format. This is because the walker operates on the shared data model which mirrors the v3 structure - which has already been manipulated by the v2 reader.

OpenApi File To Reproduce

{
    "swagger": "2.0",
    "info": {
        "title": "Test API",
        "version": "1.0.0"
    },
    "paths": {
        "/items": {
            "get": {
                "responses": {
                    "200": {
                        "description": "OK",
                        "schema": {
                            "type": "object",
                            "properties": {
                                "id": {
                                    "type": "integer"
                                },
                                "name": null
                            }
                        }
                    }
                }
            }
        }
    }
}

Expected behavior
I'd expect a validation error with the pointer:
#/paths/~1items/get/responses/200/schema/properties/name

Instead, we see a validation error with the pointer:
#/paths/~1items/get/responses/200/content/application~1octet-stream/schema/name

This example actually has two separate bugs in one:

  • It looks like a v3 pointer - the 'content/application~1octet-stream' is not present in the document.
  • It is also missing 'properties' path segment due to a bug in the ValidateSchemaPropertyHasValue rule.

Additional context
Ideally it seems like we need the walker to 'undo' the changes that the v2 reader makes to the structure so it can track what the pointer in the underlying document actually is.

At minimum we should document that the PathString is a path to the internal document representation, and not the original document - but IMO it's more useful to have a path to the location in the input document.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions