BooleanField does not return <type: 'bool'>

[Jeffrey Froman]

In some cases, a BooleanField returns an integer instead of a bool. The following example uses a MySQL backend:

# models.py
from django.db import models

class Simple(models.Model):
    b = models.!BooleanField()

---

$ ./manage.py syncdb
Creating table djest_simple
$ python
Python 2.5 (r25:51908, Sep 10 2007, 13:30:49)
[GCC 3.4.6 20060404 (Red Hat 3.4.6-8)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from djest.models import Simple
>>> simple = Simple.objects.create(b=True)
>>> simple.b
True
>>> simple_2 = Simple.objects.get(pk=simple.pk)
>>> simple_2.b
1

This may not be a problem for normal usage, but makes testing, particularly using doctest, much less elegant.

[Jeffrey Froman <django.tcijf@…>]

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

[James Bennett]

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.

[glassfordm]

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.

[anonymous]

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.

[Paul Kenjora]

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?

[bear330]

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.

[bear330]

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.

[Jacob]

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.

[(none)]

Milestone 1.0 maybe deleted

[stengleind]

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

[Jacob]

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.

[stengleind]

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

[to.roma.from.djbug@…]

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?

[James Bennett]

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.

[jkocherhans]

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

[George Vilches]

Re-opening this per Alex Gaynor's request. The current solution does not work for MySQL, MySQL continues to return 0 and 1. The attached patch (with more complete tests on models!) fixes this.

[Russell Keith-Magee]

(In [12900]) Fixed #7190 -- Corrected a problem with Boolean value handling on the MySQL backend. Thanks to George Vilches for the initial patch.

[Jacob]

Milestone 1.2 deleted

[Ramiro Morales]

In [17588]:

Fixed #15169 -- Corrected handling of model boolean fields in MySQL spatial backend.

Thanks goes to zmsmith and others for reporting the issue and to Justin Bronn for the fix.

Refs #7190, r12578, r12900, #13293, r12939.

[kirkm@…]

MySQL 5.0.3 changed bit fields from being tinyints to being some other thing (see: ​http://dev.mysql.com/doc/refman/5.0/en/bit-field-literals.html). Django now sees them as hex strings like '\x01' for b'1'. As a result, BooleanFields in models no longer work as expected. Setting a single-bit BooleanField using True and False in django works (the database gets '\x01' and '\x00' respectively), but reading these back does not get you a Python bool, but rather the raw string, like '\x00'. I think we need to tweak the Django code again to handle this case.

[Ramiro Morales]

Replying to kirkm@…:

MySQL 5.0.3 changed bit fields from being tinyints to being some other thing (see: ​http://dev.mysql.com/doc/refman/5.0/en/bit-field-literals.html). Django now sees them as hex strings like '\x01' for b'1'. As a result, BooleanFields in models no longer work as expected. Setting a single-bit BooleanField using True and False in django works (the database gets '\x01' and '\x00' respectively), but reading these back does not get you a Python bool, but rather the raw string, like '\x00'. I think we need to tweak the Django code again to handle this case.

Hmm this would mean the tests added in r1290 when fixing #7190: ​https://github.com/django/django/blob/stable/1.4.x/tests/regressiontests/model_fields/tests.py#L178 should fail when run against MySQL >= 5.0.3 which I don't think is the case.

We will need more information about the exact conditions under which you are seeing this. Bonus points if you can express them in a test case.

But please open another ticket instead of reopening this one.

[anonymous]

Replying to ramiro:

Hmm this would mean the tests added in r1290 when fixing #7190: ​https://github.com/django/django/blob/stable/1.4.x/tests/regressiontests/model_fields/tests.py#L178 should fail when run against MySQL >= 5.0.3 which I don't think is the case.

Ah, I'm using django 1.3.1 which I guess explains it.

[anonymous]

Replying to anonymous:

Replying to ramiro:

Hmm this would mean the tests added in r1290 when fixing #7190: ​https://github.com/django/django/blob/stable/1.4.x/tests/regressiontests/model_fields/tests.py#L178 should fail when run against MySQL >= 5.0.3 which I don't think is the case.

Ah, I'm using django 1.3.1 which I guess explains it.

That didn't explain it, but I did find the problem. When I added that column, I used the "bit" datatype instead of tinyint which is what Django wants to use. Once I switched it to a tinyint, everything started behaving properly. Sorry for the alarm!