Support extra="allow" for real

[RobertoRoos]

Should resolve #304

• Keep extra elements and attributes, mixed with predefined ones #304

pydantic-xml doesn't really honor extra="allow", it only does extra="ignore'.

I would love to see behavior like in pydantic itself, see for example this test: https://github.com/pydantic/pydantic/blob/26da044ff35a7d9b7b26c798b8d3c44457468005/tests/test_main.py#L267

So in pydantic you can do something like this:

    class Model(BaseModel):
        model_config = ConfigDict(extra='allow')
        a: float

    m = Model(a='10.2', b=12)
    assert m.model_extra == {'b': 12}
    m.c = 42
    assert m.model_dump() == {'a': 10.2, 'b': 12, 'c': 42}

Here for XML we could do the same, where we store attributes and extra child elements also under .model_extra. Attributes we could all load as string, the elements we could store as raw XML.

[RobertoRoos]

The version I have now only works on attributes, not on child elements.

[RobertoRoos]

I dislike this flow a little. The data is first crawled over through the model. Then, a bit later, we go over the leftovers to store them elsewhere. Is this the most sensible approach?

[RobertoRoos]

Actually, I did just add another commit that also handles extra subelements. But I don't know if this is nice enough.

[dapper91]

@RobertoRoos Hi,

extra="allow" feature is much more complicated to implement for xml than json since xml format respects element order so that extra elements order must be kept during serialization.

For example:

class TestModel(BaseXmlModel, tag='model', extra='allow', search_mode='ordered'):
    element2: str = element()

xml = '''
<model>
    <element1>1</element1>
    <element2>2</element2>
    <element3>3</element3>
</model>
'''

elements must be serialized back in the same order not as

<model>
    <element2>2</element2>
    <element1>1</element1>
    <element3>3</element3>
</model>

[RobertoRoos]

Hm, good point. We'll need to store some ordering information too then.

My biggest doubt is in this comment: #305 (comment) - is there some way we can avoid parsing the input twice?