[FIX] replicate swagger-ui problem and hacky solution

[jdkent]

Fixes #1890 .

Changes proposed in this pull request:

• revert Specification use raw spec on clone #1830 and clone using _spec
• add external reference to simple app
• try to clone with raw spec, unless there are external references
  • I'm unsure what the specific problem was in Connexion3 Failed to load API definition in the Swagger UI #1829 so I don't know how to differentiate use cases.

[RobbeSneyders]

Thanks @jdkent, but the tests are still failing with your fix.

Note that we do have tests for relative refs, so not sure why this is failing. If there is an issue, I assume it is in the ref resolving, and we should fix it there.

[jdkent]

thanks for pointing towards an existing test, for a more minimal example, try cloning the specification in the test:

def test_relative_refs(relative_refs, spec):
    spec_path = relative_refs / spec
    specification = Specification.load(spec_path)
    assert "$ref" not in specification.raw
    # clone specification
    specification_clone = specification.clone()
    assert "$ref" not in specification_clone.raw

The clone step will fail since base_uri is no longer set and clone is being applied to the "raw spec" before dereferencing is done instead of being applied to the "loaded spec" where dereferencing is already done.

I could see a solution where base_uri is an optional argument passed with clone, or some reversion of this pull request (#1830), but I do not know what problem that pull request was specifically fixing.

[jdkent]

if I change the clone function back to:

def clone(self):
        return type(self)(copy.deepcopy(self._spec))

and run this test from test_swagger_ui.py:

def test_simple(swagger_ui_app):
    app_client = swagger_ui_app.test_client()
    response = app_client.get("/v1.0/spec.json")
    assert response.status_code == 200

the tests still pass, so I'm unsure what the minimal example that makes deepcopying self._spec not viable in #1829, but deepcopying self._raw_spec fixes.

[coveralls]

coverage: 94.147%. remained the same
when pulling 185c88c on jdkent:fix/clone
into a930303 on spec-first:main.

[eharvey71]

This seems related to my issue. Unsure. #1909

[chrisinmtown]

Hacky or not, I see that PR #2002 made exactly this change to the spec.py file's clone method, and I confirmed that a split spec works again. This change has not yet been released in a version at PyPI.

However, PR #2002 did /not/ add a new test, and this PR does. I think adding a test is valuable. If you rebase this PR and squash your 3 commits down to just one, maybe it will show just the added test case and you'll get approval & merge? Please consider.