defer() uses the same cached value for all models

[jbronn]

While trying to get defer() to work with GeoQuerySets I ran into some strange issues. I eventually figured out that that problem was not isolated to GeoDjango. Here a toy model that demonstrates the problem:

from django.db import models

class City(models.Model):
    name = models.CharField(max_length=30)
    def __unicode__(self): 
        return self.name

Creating two cities, and showing what happens when the QuerySet is evaluated:

>>> c = City.objects.create(name='Pueblo')
>>> c = City.objects.create(name='Houston')
>>> City.objects.all()
[<City: Pueblo>, <City: Houston>]

However, when evaluating a QuerySet with the name field deferred, this is what's returned:

>>> City.objects.defer('name')
[<City_Deferred_name: Pueblo>, <City_Deferred_name: Pueblo>]

Notice how it shows "Pueblo" for both records, when the second one should be "Houston." The same thing happens with slicing:

>>> qs = City.objects.defer('name')
>>> qs[0].name
u'Pueblo'
>>> qs[1].name
u'Pueblo'

My gut tells me that when DeferredAttribute is cached it is somehow cached for all of the other attributes. However, I cannot pinpoint how and/or where this is happening.

[Alex Gaynor]

An alternate approach would be to put the pk in the class name, however that would mean creating O(n) classes, which strikes me as silly and wasteful of memory, given that this approach works.

[Alex Gaynor]

add tests and code to make it work properly with foreing keys

[Malcolm Tredinnick]

I'm not convinced about this patch yet (Alex: work on getting one patch right, instead of submitting 14 really quick ones). All that instance.__dict__ stuff looks a bit ugly. But I see where you're going. Will work on it a bit over the weekend.

[Malcolm Tredinnick]

Fixed in r10382.

[Jacob]

Milestone 1.1 deleted