proposal: changing fields.Field's `**metadata` to `metadata=None`

This is backward incompatible change.

pros

typo check is working correctly.

Sometimes, people that unfamiliar with marshmallow, want to add validation with typo.

e.g.

class S(Schema):
    # this is typo. (correct keyword is `validate`)
    v = fields.Integer(validator=validate.Range(max=10))

Current marshmallow’s fields.Field implementation is such as like below.

class Field(FieldABC):
    def __init__(self, default=missing_, attribute=None, load_from=None, dump_to=None,
                 error=None, validate=None, required=False, allow_none=None, load_only=False,
                 dump_only=False, missing=missing_, error_messages=None, **metadata):
        pass

So, above typo is also valid schema definition. adding Range validator as metadata, and validation on deserialization is working without error when passing invalid data.

Sometimes, sometimes, the code such like this is passed our code review, then we are disappointed ours poor heads(this is careless human error).

But, if, metadata is just keyword argument(metadata=None), this typo-ed code is raising TypeError.

[sloria]

Thanks @podhmo for the suggestion.

You raise a fair point–the current way does run the risk of passing invalid kwargs.

My worry is that we’ll be breaking lots of use cases. apispec and webargs, for example, both make use of metadata to add new field params not offered by marshmallow. Passing metadata as this way also allows users to add their own field kwargs. For example, I suggested using metadata to add a key parameter here: https://github.com/marshmallow-code/marshmallow/pull/714#issuecomment-355451867

I’ll have to think more about the tradeoffs with this one.

[sloria]

Closing this for now, as I don’t think we’ll add it in the near future. I’m open to the allowed_field_metadata idea, but I don’t think it warrants the added complexity at this point.