The footprint for this is defiantly too large in comparison to the small benefit it'd give us.

Do you think it might be worth merging the regression tests at least?

I find the idea -the original proposal- pretty interesting.
@tomchristie what about keeping things as they currently are and add a call to the field's metadata if it exists ?

Something such as metadata.py:

def get_field_info(self, field):
    if hasattr(field, 'get_metadata'):
        return field.get_metadata()
    else:
        # keep the current code

Therefore we'll have a hook to customize the metadata easily on a per field basis while keeping the footprint reasonable.

@charettes will try to take some time to dig that topic - keeping the tests.

@tomchristie what about keeping things as they currently are and add a call to the field's metadata if it exists?

I think I'd still consider that API-creep. In this area I'd prefer us to keep the API surface area low. If there's complexity, let's force that to be in the metadata class.