We Keep Coding

Why overriden save() method is bypassed?

In a Django model, I want to avoid doubles so I wrote this:
class Points(TimeStampedModel):
....
def save(self, *args, **kwargs):
if self.pk:
super(Points, self).save(*args, **kwargs)
else: # save() is a creation here, not an update
if Points.objects.filter(benef_card=self.benef_card,
spendable_at=self.spendable_at).exists():
pass
else:
super(Points, self).save(*args, **kwargs)
I was very surprised to find this result in my database:
I suppose there is something wrong with my code, but I'd like to know how these doubles could exist inspite of the protection I wrote in my save() method?

I think what you want instead is:
class Points(TimeStampedModel):
# ...
class Meta:
unique_together = ('benef_card', 'spendable_at')
Then you don't need to override save -- the uniqueness will be handled by a DB constraint and it is generally the way to go. This approach is better because save is not always called (example: bulk operations) so you might get different behavior across your app.
You might also want to check out update_or_create which just returns an object with attributes you need, creating it if it doesn't exist.

You could use Django signals instead to check before save.

How to create a django User using DRF's ModelSerializer

In django, creating a User has a different and unique flow from the usual Model instance creation. You need to call create_user() which is a method of BaseUserManager.
Since django REST framework's flow is to do restore_object() and then save_object(), it's not possible to simply create Users using a ModelSerializer in a generic create API endpoint, without hacking you way through.
What would be a clean way to solve this? or at least get it working using django's built-in piping?
Edit:
Important to note that what's specifically not working is that once you try to authenticate the created user instance using django.contrib.auth.authenticate it fails if the instance was simply created using User.objects.create() and not .create_user().

Eventually I've overridden the serializer's restore_object method and made sure that the password being sent is then processes using instance.set_password(password), like so:
def restore_object(self, attrs, instance=None):
if not instance:
instance = super(RegisterationSerializer, self).restore_object(attrs, instance)
instance.set_password(attrs.get('password'))
return instance
Thanks everyone for help!

Another way to fix this is to overwrite pre_save(self, obj) method in your extension of viewsets.GenericViewSet like so:
def pre_save(self, obj):
""" We have to encode the password in the user object that will be
saved before saving it.
"""
viewsets.GenericViewSet.pre_save(self, obj)
# Password is raw right now, so set it properly (encoded password will
# overwrite the raw one then).
obj.user.set_password(obj.user.password)
Edit:
Note that the obj in the code above contains the instance of User class. If you use Django's user model class directly, replace obj.user with obj in the code (the last line in 2 places).

I'm working with DRF. And here is how I create users:
I have a Serializer with overrided save method:
def save(self, **kwargs ):
try:
user = create_new_user(self.init_data)
except UserDataValidationError as e:
raise FormValidationFailed(e.form)
self.object = user.user_profile
return self.object
create_new_user is just my function for user creation and in the view, I just have:
def post(self, request, *args, **kwargs):
return self.create(request, *args, **kwargs)

It seems like you should be overriding restore_object() in your serializer, not save(). This will allow you to create your object correctly.
However, it looks like you are trying to abuse the framework -- you are trying to make a single create() create two objects (the user and the profile). I am no DRF expert, but I suspect this may cause some problems.
You would probably do better by using a custom user model (which would also include the profile in the same object).

passing custom save argument through a ModelForm

I have a model with some customizations in the save method.
def SomeModel(models.Model):
def save(self, *args, **kwargs):
if not kwargs.pop('skip_expensive_processing', False):
do_expensice_processing()
return super(SomeModel, self).save(*args, **kwargs)
Basically, whenever the save method gets called, I want some expensive process to be executed
But when doing a bunch a saves together (a mass import), I don't want to do the expensive processing on each save. I want to do the expensive process once after all the objects are saved.
In in case of a mass save, the objects are being created through a ModelForm. I need to find some way to modify the form so that when the form calls the save method on SomeModel, it pases on that skip_expensive_processing keyword arg. How do I do this?
I loked through te source of the ModelForm.save() method, but it doesn't seem to be caling the model save method in a too straight forward manner...

You probably don't need to override the modelform's save method. You should just be able to pass commit=True, and then the model save won't be called at all.

When is the "post_save" signal in django activated/called?

I've to update some database tables after saving a particular model. I've used the #receiver(post_save decorator for this. But when in this decorator function, the values are still not saved in the database. I've one to many relation but when I get the current instance that is being saved using kwargs['instance'], it doesn't have child objects. But after saving when I check from shell, it does have child objects. Following is the code that I'm using:
#receiver(post_save, sender=Test)
def do_something(sender, **kwargs):
test = kwargs['instance']
users = User.objects.filter(tags__in=test.tags.values_list('id',flat=True))
for user in users:
other_model = OtherModel(user=user, test=test, is_new=True)
other_model.save()

post_save is sent at the end of Model.save_base(), which is itself called by Model.save((). This means that if you override your model's save() method, post_save is sent when you call on super(YourModel, self).save(*args, **kw).
If Tag has a ForeignKey on Test and the Test instance was just created, you can't expect to have any Tag instance related to your Test instance at this stage, since the Tag instances obviously need to know the Test instance's pk first so they can be saved too.

The post_save for the parent instance is called when the parent instance is saved. If the children are added after that, then they won't exist at the time the parent post_save is called.

Why doesn't django's model.save() call full_clean()?

I'm just curious if anyone knows if there's good reason why django's orm doesn't call 'full_clean' on a model unless it is being saved as part of a model form.
Note that full_clean() will not be called automatically when you call your model’s save() method. You’ll need to call it manually when you want to run one-step model validation for your own manually created models.
django's full clean doc
(NOTE: quote updated for Django 1.6... previous django docs had a caveat about ModelForms as well.)
Are there good reasons why people wouldn't want this behavior? I'd think if you took the time to add validation to a model, you'd want that validation run every time the model is saved.
I know how to get everything to work properly, I'm just looking for an explanation.

AFAIK, this is because of backwards compatibility. There are also problems with ModelForms with excluded fields, models with default values, pre_save() signals, etc.
Sources you might be intrested in:
http://code.djangoproject.com/ticket/13100
http://groups.google.com/group/django-developers/browse_frm/thread/b888734b05878f87

Because of the compatibility considering, the auto clean on save is not enabled in django kernel.
If we are starting a new project and want the default save method on Model could clean automatically, we can use the following signal to do clean before every model was saved.
from django.dispatch import receiver
from django.db.models.signals import pre_save, post_save
#receiver(pre_save)
def pre_save_handler(sender, instance, *args, **kwargs):
instance.full_clean()

The simplest way to call the full_clean method is just to override the save method in your model:
class YourModel(models.Model):
...
def save(self, *args, **kwargs):
self.full_clean()
return super(YourModel, self).save(*args, **kwargs)

Commenting on #Alfred Huang's answer and coments on it. One might lock the pre_save hook down to an app by defining a list of classes in the current module (models.py) and checking against it in the pre_save hook:
CUSTOM_CLASSES = [obj for name, obj in
inspect.getmembers(sys.modules[__name__])
if inspect.isclass(obj)]
#receiver(pre_save)
def pre_save_handler(sender, instance, **kwargs):
if type(instance) in CUSTOM_CLASSES:
instance.full_clean()

If you have a model that you want to ensure has at least one FK relationship, and you don't want to use null=False because that requires setting a default FK (which would be garbage data), the best way I've come up with is to add custom .clean() and .save() methods. .clean() raises the validation error, and .save() calls the clean. This way the integrity is enforced both from forms and from other calling code, the command line, and tests. Without this, there is (AFAICT) no way to write a test that ensures that a model has a FK relation to a specifically chosen (not default) other model.
class Payer(models.Model):
name = models.CharField(blank=True, max_length=100)
# Nullable, but will enforce FK in clean/save:
payer_group = models.ForeignKey(PayerGroup, null=True, blank=True,)
def clean(self):
# Ensure every Payer is in a PayerGroup (but only via forms)
if not self.payer_group:
raise ValidationError(
{'payer_group': 'Each Payer must belong to a PayerGroup.'})
def save(self, *args, **kwargs):
self.full_clean()
return super().save(*args, **kwargs)
def __str__(self):
return self.name

Instead of inserting a piece of code that declares a receiver, we can use an app as INSTALLED_APPS section in settings.py
INSTALLED_APPS = [
# ...
'django_fullclean',
# your apps here,
]
Before that, you may need to install django-fullclean using PyPI:
pip install django-fullclean

A global pre_save signal can work well if you want to always ensure model validation. However it will run into issues with Django's auth in current versions (3.1.x) and could cause issues with models from other apps you are using.
Elaborating on #Peter Shannon's answer, this version will only validate models inside the module you execute it in, skips validation with "raw" saves and adds a dispatch_uid to avoid duplicate signals.
from django.db.models.signals import pre_save
import inspect
import sys
MODELS = [obj for name, obj in
inspect.getmembers(sys.modules[__name__], inspect.isclass)]
def validate_model(sender, instance, **kwargs):
if 'raw' in kwargs and not kwargs['raw']:
if type(instance) in MODELS:
instance.full_clean()
pre_save.connect(validate_model, dispatch_uid='validate_models')