anyOf

[patrickkwang]

No description provided.

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (merging@b45a141). Click here to learn what that means.
The diff coverage is n/a.

@@            Coverage Diff             @@
##             merging      #19   +/-   ##
==========================================
  Coverage           ?   97.64%           
==========================================
  Files              ?        2           
  Lines              ?       85           
  Branches           ?        0           
==========================================
  Hits               ?       83           
  Misses             ?        2           
  Partials           ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b45a141...2d0c69f. Read the comment docs.

[patrickkwang]

This is a simplified treatment. The documentation is not super clear in this area, but I'm pretty sure that you can actually have type and anyOf defined in the same schema. What I don't understand is what that's supposed to mean. However, it doesn't seem like a very common use case, so I'm inclined to handle that later when we're able to generate some test cases.

[alongreyber]

I added another test for a nested anyOf, and it doesn't seem to work:

def test_anyof_nested():
    """Test generating an object from an anyOf schema."""
    schema = {
        "properties": {
            "a": {
                "anyOf": [
                    {
                        "type": "string",
                    },
                    {
                        "type": "integer",
                    },
                ],
            }
        },
    }
    output = generate_json(schema)
    assert isinstance(output["a"], (str, int))

Is this a use case we need to support? It seems like it should be, but I'm not sure how we would go about implementing this.

[patrickkwang]

It looks like that fails because the top-level schema ought to have "type": "object" and doesn't. To ensure the test passes, you should probably also add "required": ["a"].

[patrickkwang]

After digging a bit more, I think I understand. This:

{
    "type": "object",
    "properties": {
        "a": {},
        "b": {}
    },
    "anyOf": [
        {"required": ["a"]},
        {"required": ["b"]}
    ]
}

is equivalent to:

{
    "anyOf": [
        {
            "type": "object",
            "properties": {
                "a": {},
                "b": {}
            },
            "required": ["a"]
        },
        {    
            "type": "object",
            "properties": {
                "a": {},
                "b": {}
            },
            "required": ["b"]
        }
    ]
}

So I think we ought to do basically that transformation, followed by the simple anyOf procedure I've implemented here. @alongreyber has already taken a swing at schema merging in #20, so perhaps we should finish that up first.

[alongreyber]

This isn't exactly how you were planning to do this, but I think it is equivalent. The idea is to sample the list of anyOf schemas provided to pick a sublist (because anyOf can be one or more of the provided schemas). We merge those subschemas together and check if those subschemas were merged without error. If they were merged without error, we merge the combined schemas back into the base. If not, we pick a new sample. There will always be a valid sample because picking a single subschema is always valid. Let me know if you have questions about this process.