QueryDict.update eats data when other_dict is a MultiValueDict

[Jeremy Dunck]

django.http.QueryDict supports .update, but it seems to assume that other_dict is a (builtin) dict.

This is dangerous when mixing two QueryDicts.

>>> x = QueryDict('a=1&a=2')
>>> y = QueryDict('a=3&a=4')
>>> print x
<QueryDict: {u'a': [u'1', u'2']}>
>>> print y
<QueryDict: {u'a': [u'3', u'4']}>
>>> x._mutable = True
>>> x.update(y)
>>> x._mutable=False
>>> print x
<QueryDict: {u'a': [u'1', u'2', u'4']}>

Note that a=3 was lost.

As far as I can see, there's no really nice fix to this. QueryDict.update doesn't just defer to MultiValueDict because it needs to force keys and values into unicode.

Defering to MultiValueDict first and then forcing means duplicate keys for items not yet coerced.
Forcing individual items results in lists from QueryDict becoming (single) items in MultiValueDict.

i.e.

dict([(f(k), f(v)) for k, v in other_dict.items()])
->
dict([(f(k), [f(v) for v in values]) for k, values in other_dict.lists()])

This seems to be a leaky abstraction here; I think QueryDict is not really a MultiValueDict, since MultiValueDict can have any (hashable) objects as keys and values, while QueryDict insists on unicode objects only.

No patch from me since the solution isn't obvious to me. Marking as 1.0 since this is a bug, though.

[Julien Phalip]

Patch and regression tests

[Julien Phalip]

I think there is a combination of two issues here:

• dict eats duplicate keys:

>>> dict([(u'a', u'1'), (u'a', u'2')])
dict: {u'a': u'2'} 

• And MultiValueDict is not prepared to receive values as lists.

Hopefully the attached patch fixes that.

[Julien Phalip]

Sorry, I meant: MultiValueDict.update() is not prepared...

[Chris Beaven]

From a quick scan of your patch:

1. Your tests don't actually check for the regression of updating a QueryDict with a normal dict (which leads to...)

2. It looks like this will break for updating normal dicts

3. It'd be better to use QueryDict('...', mutable=True) rather than messing around with internal variables in the test

[Julien Phalip]

Replying to SmileyChris:

From a quick scan of your patch:

1. Your tests don't actually check for the regression of updating a QueryDict with a normal dict (which leads to...)

2. It looks like this will break for updating normal dicts

3. It'd be better to use QueryDict('...', mutable=True) rather than messing around with internal variables in the test

I agree, I realised the patch was wrong shortly after I posted. Trying to give it another go.

[Julien Phalip]

Ok, I've amended the path and now all tests pass. The trick I used was to iterate through the list of values and updated value by value. Again, I'm not sure if that's the best approach, but at least it seems to work.

[Julien Phalip]

2nd attempt. All tests pass.

[Jacob]

(In [8705]) Fixed #8278: fixed QueryDict.update(QueryDict). Thanks, julien.

[Jacob]

Milestone 1.0 deleted