Issue warning when map_entry can't find field for query

[lafrech]

While working on map_query, I realized a shortcoming of the field/name mapping in the inheritance case.

In the framework tests, we use the parent class to find a child element based on a query on a field that does not exist in the parent class.

InheritanceSearchParent.find_one({'sc1f': 1})

It works, but I believe it is wrong. The query mapper does not have access to the field, so it can not do its job. In fact the query will fail if the field is an EmbeddedField (I could reproduce that) because in map_entry we won't pass in the dedicated case as field is None:

elif isinstance(field, EmbeddedField):

I don't think we can pass the child classes fields here, and it would be a bad idea because different children could have different fields for the same name, so the field is undefined.

Since it works in some (in fact, most) cases anyway, I propose to just add a non-breaking warning at least for now. But I can make this an Exception. Or if we plan to forbid this in the next major version, I can use a DeprecationWarning.

Or do we want users to be able to pass query entries that don't match a field known to the model?

This is a corner case, but since the wrong use was shown in the tests, users may be relying on it.

I can review the warning/exception type, the message string (add a note about inheritance ?), and perhaps rework the framework tests to isolate this specific test.

Feedback welcome.

[coveralls]

Coverage increased (+0.005%) to 96.18% when pulling a719444 on warning_map_entry_field_none into 0424772 on master.

[lafrech]

This raises a warning in at least one legit case:

    {'$near': {'$geometry': {'type': 'Point', 'coordinates': [122.0, 31.0]}}}

Maybe we should be totally loose and not try to warn the user at all.

I still think the framework tests should be reworked so as not to expose a use case that can't really be supported.