BooleanField does not return <type: 'bool'>

patch for django/db/models/fields/init.py

This is kind of tricky, because under the hood in Python bool is a subclass of int that only ever has two instances (which test equal to 0 and 1 for False and True, respectively). Not all DBs actually store a boolean value, either; some store a 0 or a 1, and return that. Given that, and Python's general bent toward duck typing, I'm not sure whether we should strictly ensure that it always returns a bool instance.

Note: the approach in the mysql.patch attachment handles some cases that the approach in the boolfield.diff attachment does not: for example, the former fixes results returned by calling Model.objects.value(), and so fixes the doctests in django/tests/regressiontests/model_inheritance_regress/models.py; the latter does not fix these doctests.

Also, a note from the django developer's mailing list:

Pim Van Heuven wrote:

The lack of boolean type coercion is more serious than it looks like at first glance. (starting from the example at http://code.djangoproject.com/ticket/7190) In [1]: import django.newforms as forms In [2]: from django.newforms.models import model_to_dict In [3]: from simple.models import Simple In [4]: simple_false = Simple.objects.create(b=False) In [5]: simple_false_2 = Simple.objects.get(pk=simple_false.pk) In [6]: class SimpleForm?(forms.ModelForm?): ...: b = forms.BooleanField?(widget=forms.HiddenInput?) ...: class Meta: ...: model = Simple ...: In [7]: SimpleForm?(data = model_to_dict(simple_false)).as_p() Out[7]: u'<input type="hidden" name="b" value="False" id="id_b" />' In [8]: SimpleForm?(data = model_to_dict(simple_false_2)).as_p() Out[8]: u'<input type="hidden" name="b" value="0" id="id_b" />' When you POST the Out[8] form the value becomes bool("0") = True. So when you save simple_false_2 based on the form the value is inverted from False to True. Seems like a critical error... The mysql.patch solves this issue too. Pim.

Replying to ubernostrum:

This is kind of tricky, because under the hood in Python bool is a subclass of int that only ever has two instances (which test equal to 0 and 1 for False and True, respectively). Not all DBs actually store a boolean value, either; some store a 0 or a 1, and return that. Given that, and Python's general bent toward duck typing, I'm not sure whether we should strictly ensure that it always returns a bool instance.

I dont think we can get away with duck typing here since it causes problems with the tests, Model.objects.value() and when posting a form with a hidden input for a boolean field (see example below). After a quick grep it seems that the old mysql, sqlite and the oracle driver all have some kind of boolean type coercion. Besides the mysql.path confines the changes to the mysql backend.

I applied the patch to 0.97-pre-SVN-5 with no effect. Please advise if anyone else has applied the patch with success, maybe missing some code on the attachments?

Replying to ubernostrum:

This is kind of tricky, because under the hood in Python bool is a subclass of int that only ever has two instances (which test equal to 0 and 1 for False and True, respectively). Not all DBs actually store a boolean value, either; some store a 0 or a 1, and return that. Given that, and Python's general bent toward duck typing, I'm not sure whether we should strictly ensure that it always returns a bool instance.

Yes, 1 means true in python, but not at this situation:

>>> 1 is True
False

I also encounter this problem in my code. I think it should alway return True or False for Boolean field. If I use:

if field is True

that will fail. but if I use:

if field

That will be true if field is any non false value. (list, tuple, string...)

To make BooleanField? always return True or False, it only need to modify BooleanField?'s to_python method.

Thanks.

Replying to bear330:

To make BooleanField? always return True or False, it only need to modify BooleanField?'s to_python method.

if value in (True, False): return value

=>

if value in (True, False): return value == True

Thanks.

This basically comes down to a limitation of MySQL and MySQLdb. The cast function technically would work, but it's silly to cast any tinyint to bool.

Milestone 1.0 maybe deleted

By the way, this killed some code of mine that depended on True being True. My functional test had an assertEqual(True, someBooleanField) and it silently broke when I switched backends from sqlite to MySQL.

If it is not going to be fixed, it needs to be very clearly documented (which it is currently not).

-Dave

Please don't reopen tickets closed by a core developer, and especially not if you're going to change the subject. If you disagree with the wontfix call, bring it up on django-dev; if you'd like something else done (more docs are almost always a good idea), open a new ticket.

Ticket done (http://code.djangoproject.com/ticket/8802).

I will bring this up on the mailing list because:

1 == True True

1 is True False

-Dave

Could you please elaborate what exactly prevents you from applying one of the patches? I searched the mailing list archives and the only obstacle I found was inconsistency with qs.values() which would return 0/1 rather than True/False, but to my mind it’s values()’s behavior that should be changed, not the other way round (#10412). You call the int->bool cast “silly”, and maybe it is, but what are your actual concerns—performance hit, inconsistency, something else?

Reopening after discussion in #django-dev, targeting for 1.2.

Note that the eventual fix for this can't be backported into 1.1 since we documented the behavior.

(In [12578]) Fixed #7190. Boolean fields now return bool values instead of 1 or 0. Thanks, Alex Gaynor.