A Django utility to scrub model field values on save.
This Django utility allows the definition of methods or functions to clean model object field values on save.
Since django-clean-fields is not a Django app, simply include it on your PYTHONPATH. The easiest way to do this is installing via pip:
pip install django-clean-fields
No changes to the project’s settings are necessary.
Two alternate implementation options are available: an extended model class that closely resembles conventions used by the Django forms API, and a decorator that registers a callable with the pre_save signal. Which approach to use is a decision left to the developer. The former option may provide the most familiarity with existing conventions; the latter offers more flexibility.
Any model inheriting from the abstract clean_fields.models.CleanFieldsModel will check for and run cleaner methods as its first action when saving. Such methods should match the conventions used by form field validators, namely:
- methods must be named clean_<field_name>
- methods must accept the current value of the field
- methods must return the “cleaned” value, ready to be written to the database
- method may raise an exception to interrupt saving
from django.core.exceptions import ValidationError from django.db import models from clean_fields.models import CleanFieldsModel class Article(CleanFieldsModel): title = models.CharField(max_length=30) def clean_title(self, unsaved_title): if "you'll never believe" in unsaved_title.lower(): raise ValidationError('Sensationalist Clickbait Not Allowed') return unsaved_title.title()
The clean_fields.decorators.cleans_field decorator can be applied to any callable, which will then be invoked when the pre_save signal is sent by the corresponding model. The decorator requires a single argument: a reference string identifying the field to clean, which must follow the pattern “app_name.ModelName.field_name”. Note that the full reference must be provided even if the callable is within the model class itself.
As above, any decorated callable must accept the current field value and return the “cleaned” value. The code below has the identical effect as the above example.
from django.core.exceptions import ValidationError from django.db import models from clean_fields.decorators import cleans_field class Article(models.Model): title = models.CharField(max_length=30) @cleans_field('your_app.Article.title') def ensure_title_case(self, unsaved_title): return unsaved_title.title() # Multiple cleaners can be defined for a single field. # Also, they needn't be instance methods on the model object. @cleans_field('your_app.Article.title') def validate_dignified_title(unsaved_title): if "you'll never believe" in unsaved_title.lower(): raise ValidationError('Sensationalist Clickbait Not Allowed') return unsaved_title
There is solid reasoning behind the omission of similar behavior in Django’s core. For one, it might create a feeling of false security. Validation runs on save, but that does not prevent “uncleaned” data from being committed to the database (for instance, via the ORM’s `bulk_create <https://docs.djangoproject.com/en/dev/ref/models/querysets/#bulk-create>`__ or `update <https://docs.djangoproject.com/en/dev/ref/models/querysets/#update>`__ methods, which circumvent save()). Furthermore, a lack of model-level validation encourages a separation between a user’s interaction with model objects and a developer’s interaction with model objects. This rigorous definition of user roles is usually a Good Thing, but it can impose an unnecessary burden on projects that don’t require user-driven interfaces. Be sure that this workflow benefits your project before installing it.
If in doubt, it’s worth noting some built-in alternative means to accomplish similar cleaning behavior. For instance:
The forms API
However, forms and their validation are intended to be used within the context of a web page. They loose much of their simplicity when handled entirely on the backend.
Model field constraints and validators
Model fields provide two ways avoid committing erroneous values to the database. The first are field options; passed as keyword arguments to your fields declarations, these will enforce value contraints on the database level (eg. CharField’s max_length). The second is the ability to define validators. These functions, more flexible in Python than at the database level, will raise errors if the values to be saved to not adhere to some defined pattern or convention.
While both these options keep the validation at the model level, their benefit is merely error prevention. Neither allow the ability to “massage” data into an acceptable format.
Django does provide some support custom model object validation via the `Model.clean() method <https://docs.djangoproject.com/en/dev/ref/models/instances/#django.db.models.Model.clean>`__. This allows modifying attributes, allows access to multiple fields, and will be called via the `Model.full_clean() method <https://docs.djangoproject.com/en/dev/ref/models/instances/#django.db.models.Model.full_clean>`__.
This sort of custom validation sounds ideal, except that it is not called when a model object is saved. full_clean() or clean() must be invoked manually by any object other than a ModelForm. Moreover, from a design perspective, it’s preferable to have methods with as narrow a focus as possible: a single method to clean a single aspect of a single field is better than clean(), which must handle all validation on all fields.
The cleans_field decorator already leverages built-in Django signals (specifically, the pre_save signal). It is possible to handle field scrubbing directly by defining your own signal handlers and connecting them to the appropriate signal.
The greatest shortcoming of this approach is that it encourages bad OO design: signal handlers of this nature would easily be defined apart from the models which they are meant to modify. Even implemented as staticmethods on the appropriate models, their method signature is obtuse, and therefore difficult to use outside of the context of signals.
This project intends to pick up the slack where the above built-in methods fall short, providing a simple interface to support streamlined model design. It’s not uncircumventable, so caveat emptor, but aims to make your life easier.