Make Form data/files initialize with an empty MultiValueDict rather than dict

[Sven R. Kunze]

You might have a look here:

​https://github.com/django/django/blob/362813d6287925b8f63f/django/forms/forms.py#L78

None is converted to a regular dict but not to a QueryDict.

Methods of the form might rely on the API of a QueryDict such as 'iterlists' or 'getlist' which a regular dict doesn't provide.

[Herbert Fortes]

Hi,

To be clear, ​QueryDict is a subclass of dict.

Is the idea to add a feature?

At a first look, the dictionary is there to avoid that the code breaks.
# edited to fix english

Regards,

[Sven R. Kunze]

Replying to Herbert Fortes:

Is the idea to add a feature?

Sorry, if that wasn't clear from the issue itself. Instead of:

        self.data = {} if data is None else data

this would be 100% compatible with what views usually pass into forms:

        self.data = QueryDict() if data is None else data

At a first look, the dictionary is there to avoid that the code breaks.

I think so as well and it works in many places until people think they can rely on the QueryDict API such as 'iterlists' or 'getlist' which is not always true.

[Tim Graham]

A similar issue was raised in #27989. The main concern I have is that this would couple django.forms to django.http.

[Sven R. Kunze]

Yes, I thought that this was the reason to change it from "data or {}" to ternary operator.

What do you think about:

1) either using MultiValueDict instead of QueryDict? That would couple django.forms to django.utils.datastructures

2) or moving QueryDict to django.utils.datastructures and then using it in the BaseForm constructor?

At least, MultiValueDict supports the getlist, setlist and lists APIs. So, those would behave the same because the dicts are empty anyway. Althought, I would rather tend to option 2.

What do you think?

[Simon Charette]

I like the MultiValueDict approach, +1.

[Herbert Fortes]

MultiValueDict +1

[Sven R. Kunze]

Okay seems like we have winner. :D

Just exploring the other option a bit more: why is moving QueryDict to datastructures not a feasible option? It actually sounds like a more reasonable place for it.

[Tim Graham]

I agree that QueryDict should remain in django.http. I don't feel that it's something meant for use outside of the request/response cycle.

[Sven R. Kunze]

I don't feel that it's something meant for use outside of the request/response cycle.

Sorry, for still nagging here, but I still don't understand why it cannot be moved to datastructures.

Its doc string: "A specialized MultiValueDict which represents a query string." <<< That sounds like a perfect valid datastructure handling encoding and query string syntax. Its source code has no reference to request/responses.

Also APIs MultiValueDict doesn't provide:

• urlencode
• encoding (getter and setter)
• fromkeys
• mutability
• isinstance check

I just want to make sure we don't need to touch this part in 3 years AGAIN when somebody else needs the real thing.

[Jeff]

Do we want to move ahead with the MultiValueDict, or delay for more discussion concerning the QueryDict?

[Herbert Fortes]

IMHO, an empty dict should not make too much noise on the code. For now at least.

[Tim Graham]

​PR. Please uncheck "Needs tests" after updating.

[Tim Graham <timograham@…>]

In 4c086d7d:

Fixed #29459 -- Initialized form data/files with empty MultiValueDicts.