Miscellaneous Utilities

Choices

Note

Django 3.0 adds enumeration types. These provide most of the same features as Choices.

Choices provides some conveniences for setting choices on a Django model field:

from model_utils import Choices

class Article(models.Model):
    STATUS = Choices('draft', 'published')
    status = models.CharField(choices=STATUS, default=STATUS.draft, max_length=20)

A Choices object is initialized with any number of choices. In the simplest case, each choice is a string; that string will be used both as the database representation of the choice, and the human-readable representation. Note that you can access options as attributes on the Choices object: STATUS.draft.

But you may want your human-readable versions translated, in which case you need to separate the human-readable version from the DB representation. In this case you can provide choices as two-tuples:

from model_utils import Choices

class Article(models.Model):
    STATUS = Choices(('draft', _('draft')), ('published', _('published')))
    status = models.CharField(choices=STATUS, default=STATUS.draft, max_length=20)

But what if your database representation of choices is constrained in a way that would hinder readability of your code? For instance, you may need to use an IntegerField rather than a CharField, or you may want the database to order the values in your field in some specific way. In this case, you can provide your choices as triples, where the first element is the database representation, the second is a valid Python identifier you will use in your code as a constant, and the third is the human-readable version:

from model_utils import Choices

class Article(models.Model):
    STATUS = Choices((0, 'draft', _('draft')), (1, 'published', _('published')))
    status = models.IntegerField(choices=STATUS, default=STATUS.draft)

You can index into a Choices instance to translate a database representation to its display name:

status_display = Article.STATUS[article.status]

Option groups can also be used with Choices; in that case each argument is a tuple consisting of the option group name and a list of options, where each option in the list is either a string, a two-tuple, or a triple as outlined above. For example:

from model_utils import Choices

class Article(models.Model):
STATUS = Choices(('Visible', ['new', 'archived']), ('Invisible', ['draft', 'deleted']))

Choices can be concatenated with the + operator, both to other Choices instances and other iterable objects that could be converted into Choices:

from model_utils import Choices

GENERIC_CHOICES = Choices((0, 'draft', _('draft')), (1, 'published', _('published')))

class Article(models.Model):
    STATUS = GENERIC_CHOICES + [(2, 'featured', _('featured'))]
    status = models.IntegerField(choices=STATUS, default=STATUS.draft)

Should you wish to provide a subset of choices for a field, for instance, you have a form class to set some model instance to a failed state, and only wish to show the user the failed outcomes from which to select, you can use the subset method:

from model_utils import Choices

OUTCOMES = Choices(
    (0, 'success', _('Successful')),
    (1, 'user_cancelled', _('Cancelled by the user')),
    (2, 'admin_cancelled', _('Cancelled by an admin')),
)
FAILED_OUTCOMES = OUTCOMES.subset('user_cancelled', 'admin_cancelled')

The choices attribute on the model field can then be set to FAILED_OUTCOMES, thus allowing the subset to be defined in close proximity to the definition of all the choices, and reused elsewhere as required.

Field Tracker

A FieldTracker can be added to a model to track changes in model fields. A FieldTracker allows querying for field changes since a model instance was last saved. An example of applying FieldTracker to a model:

from django.db import models
from model_utils import FieldTracker

class Post(models.Model):
    title = models.CharField(max_length=100)
    body = models.TextField()

    tracker = FieldTracker()

Note

django-model-utils 1.3.0 introduced the ModelTracker object for tracking changes to model field values. Unfortunately ModelTracker suffered from some serious flaws in its handling of ForeignKey fields, potentially resulting in many extra database queries if a ForeignKey field was tracked. In order to avoid breaking API backwards-compatibility, ModelTracker retains the previous behavior but is deprecated, and FieldTracker has been introduced to provide better ForeignKey handling. All uses of ModelTracker should be replaced by FieldTracker.

Summary of differences between ModelTracker and FieldTracker:

  • The previous value returned for a tracked ForeignKey field will now be the raw ID rather than the full object (avoiding extra database queries). (GH-43)
  • The changed() method no longer returns the empty dictionary for all unsaved instances; rather, None is considered to be the initial value of all fields if the model has never been saved, thus changed() on an unsaved instance will return a dictionary containing all fields whose current value is not None.
  • The has_changed() method no longer crashes after an object’s first save. (GH-53).

Accessing a field tracker

There are multiple methods available for checking for changes in model fields.

has_changed

Returns True if the given field has changed since the last save. The has_changed method expects a single field:

>>> a = Post.objects.create(title='First Post')
>>> a.title = 'Welcome'
>>> a.tracker.has_changed('title')
True
>>> a.tracker.has_changed('body')
False

The has_changed method relies on previous to determine whether a field’s values has changed.

If a field is deferred and has been assigned locally, calling has_changed() will load the previous value from the database to perform the comparison.

changed

Returns a dictionary of all fields that have been changed since the last save and the values of the fields during the last save:

>>> a = Post.objects.create(title='First Post')
>>> a.title = 'Welcome'
>>> a.body = 'First post!'
>>> a.tracker.changed()
{'title': 'First Post', 'body': ''}

The changed method relies on has_changed to determine which fields have changed.

Tracking specific fields

A fields parameter can be given to FieldTracker to limit tracking to specific fields:

from django.db import models
from model_utils import FieldTracker

class Post(models.Model):
    title = models.CharField(max_length=100)
    body = models.TextField()

    title_tracker = FieldTracker(fields=['title'])

An example using the model specified above:

>>> a = Post.objects.create(title='First Post')
>>> a.body = 'First post!'
>>> a.title_tracker.changed()
{'title': None}

Tracking Foreign Key Fields

It should be noted that a generic FieldTracker tracks Foreign Keys by db_column name, rather than model field name, and would be accessed as follows:

from django.db import models
from model_utils import FieldTracker

class Parent(models.Model):
    name = models.CharField(max_length=64)

class Child(models.Model):
    name = models.CharField(max_length=64)
    parent = models.ForeignKey(Parent)
    tracker = FieldTracker()
>>> p = Parent.objects.create(name='P')
>>> c = Child.objects.create(name='C', parent=p)
>>> c.tracker.has_changed('parent_id')

To find the db_column names of your model (using the above example):

>>> for field in Child._meta.fields:
        field.get_attname_column()
('id', 'id')
('name', 'name')
('parent_id', 'parent_id')

The model field name may be used when tracking with a specific tracker:

specific_tracker = FieldTracker(fields=['parent'])

But according to issue #195 this is not recommended for accessing Foreign Key Fields.

Checking changes using signals

The field tracker methods may also be used in pre_save and post_save signal handlers to identify field changes on model save.

Note

Due to the implementation of FieldTracker, post_save signal handlers relying on field tracker methods should only be registered after model creation.

FieldTracker implementation details

from django.db import models
from model_utils import FieldTracker, TimeStampedModel

class MyModel(TimeStampedModel):
    name = models.CharField(max_length=64)
    tracker = FieldTracker()

    def save(self, *args, **kwargs):
        """ Automatically add "modified" to update_fields."""
        update_fields = kwargs.get('update_fields')
        if update_fields is not None:
            kwargs['update_fields'] = set(update_fields) | {'modified'}
        super().save(*args, **kwargs)

# [...]

instance = MyModel.objects.first()
instance.name = 'new'
instance.save(update_fields={'name'})

This is how FieldTracker tracks field changes on instance.save call.

  1. In class_prepared handler FieldTracker patches save_base and refresh_from_db methods to reset initial state for tracked fields.
  2. In post_init handler FieldTracker saves initial values for tracked fields.
  3. MyModel.save changes update_fields in order to store auto updated modified timestamp. Complete list of saved fields is now known.
  4. Model.save does nothing interesting except calling save_base.
  5. Decorated save_base() method calls super().save_base and all fields that have values different to initial are considered as changed.
  6. Model.save_base sends pre_save signal, saves instance to database and sends post_save signal. All pre_save/post_save receivers can query instance.tracker for a set of changed fields etc.
  7. After Model.save_base return FieldTracker resets initial state for updated fields (if no update_fields passed - whole initial state is reset).
  8. instance.refresh_from_db() call causes initial state reset like for save_base().

When FieldTracker resets fields state

By the definition:

Note

  • Field value is changed if it differs from current database value.
  • Field value was changed if value has changed in database and field state didn’t reset.
instance = Tracked.objects.get(pk=1)
# name not changed
instance.name += '_changed'
# name is changed
instance.save()
# name is not changed again

Current implementation resets fields state after post_save signals emitting. This is convenient for “outer” code like in example above, but does not help when model save method is overridden.

class MyModel(models.Model)
    name = models.CharField(max_length=64)
    tracker = FieldTracker()

    def save(self):  # erroneous implementation
        self.name = self.name.replace(' ', '_')
        name_changed = self.tracker.has_changed('name')
        super().save()
        # changed state has been reset here, so we need to store previous state somewhere else
        if name_changed:
            do_something_about_it()

FieldTracker provides a context manager interface to postpone fields state reset in complicate situations.

  • Fields state resets after exiting from outer-most context
  • By default, all fields are reset, but field list can be provided
  • Fields are counted separately depending on field list passed to context managers
  • Tracker can be used as decorator
  • Different instances have their own context state
  • Different trackers in same instance have separate context state
class MyModel(models.Model)
    name = models.CharField(max_length=64)
    tracker = FieldTracker()

    def save(self):  # correct implementation
        self.name = self.name.replace(' ', '_')

        with self.tracker:
            super().save()
            # changed state reset is postponed
            if self.tracker.has_changed('name'):
                do_something_about_it()

    # Decorator example
    @tracker
    def save(self): ...

    # Restrict a set of fields to reset here
    @tracker(fields=('name'))
    def save(self): ...

    # Context manager with field list
    def save(self):
        with self.tracker('name'):
            ...