queryset.count() does not use _result_cache

[webograph@…]

when using .count() on an already executed queryset, the query is evaluated again instead of just checking the length of the _result_cache.

it can be worked around by using len() on queries known to have been executed, in which case len() performs better then .count().

[Chris Beaven]

Bump back if you think it needs tests, but it's working fine for me and the code is pretty straight-forward.

[Benjamin Slavin]

-0 from me.

If you want to know the number of objects, I'd recommend sticking to len(QS). Otherwise, there's no guarantee that the result from .count() is the number of objects in the QS you have.

There are ways to fix this, but I think that the current method is preferable.

Here's an example of the problem:

>>> a = Foo.objects.all()
>>> b = Foo.objects.all()
>>> a
[<Foo1>, <Foo2>]
>>> a.count()
2L
>>> b.count()
2L
......Something is added to the DB......
>>> b
[<Foo1>, <Foo2>, <Foo3>]
>>> b.count()
2L
>>> len(b)
3

I'm changing this to 'design decision needed'.

[Chris Beaven]

You're right that caching the count is probably overly aggressive. Here's a patch which doesn't cache but simply does what the original ticket suggests.

b.count() should always match len(_result_cache) now.

[Chris Beaven]

Behaviour now:

>>> a = Foo.objects.all()
>>> b = Foo.objects.all()
>>> a
[<Foo1>, <Foo2>]
>>> a.count()
2L
>>> b.count()
2L

Something is added to the DB.
>>> b
[<Foo1>, <Foo2>, <Foo3>]
>>> b.count()
3
>>> len(b)
3

There is a change from current functionality with this patch: a is still cached so a.count() will now be 2 whereas previously it would use another SQL query and return 3L.

[Benjamin Slavin]

I think that this is better than the original patch. I still personally prefer just using len(QS), but I change my vote to +0.

My only remaining concern if this patch is introduced is the varied data types of the return value... It may be worth putting a typecast in there so the return type is predictably 'long'.

[Chris Beaven]

Supposedly, the type .count() returns depends on the database so it's not predictable anyway.

[Jacob]

(In [4561]) Fixed #3541: queryset.count() now respects the queryset cache.