prepared-properties 1.0.3

Declarative annotations and prefetches for Django models

Django Prepared Properties

Declarative annotations and prefetches for Django models.

We often find ourselves writing a lot of quite complex annotations and prefetches. While these are used across the project, the logic that defines them is kind of hidden in a queryset method (or more of them, if the annotations rely on eachother). If there's a default/naive implementation as a property, that logic is held in a completely different place. Using these annotations requires you to call each of those queryset methods, which quickly becomes a bit of a mess.

This package attempts to solve this by allowing you to define annotations and prefetches as 'prepared' properties, so you can add them to you querysets by referencing the class attribute. The queryset prepare method resolves annotation dependencies and builds the correct queryset for you.

Requires python 3.6 and django 2.2. It probably works on django 3, but I haven't tried it (PRs welcome <3)

Annotated Properties

To add an annotated property, you simply need to pass in the annotation into either the AnnotatedProperty class or annotated_property decorator. You can then prepare it by calling prepare(Model.propery_name) on the queryset.

To avoid cyclic references to models in the annotation (eg. by referring to the model the property is on in the annotation), you can also pass the annoation as a lambda, which will be evaluated when the queryset is annotated.

When you use the decorator, the body of the method you decorate will be used when the annotation is not present. Using it will emit a warning advising you to use the prepare querset method.

from django.db.models import Model, Sum, OuterRef, Manager
from prepared_properties import annotated_property, AnnotatedProperty, PropertiedQuerySet

class Book(Model):
    page_number = models.PositiveIntegerField()

class Author(Model):
    objects = Manager.from_queryset(PropertiedQuerySet)()

    pages_written = AnnotatedProperty(
        Subquery(
            Book.objects.filter(author=OuterRef("pk"))
            .values("author")
            .annotate(pages_written=Sum("page_number"))
            .values("page_number")
        )
    )

    # ... or with a default getter:

    @annotated_property(
        Subquery(
            Book.objects.filter(author=OuterRef("pk"))
            .values("author")
            .annotate(pages_written=Sum("page_number"))
            .values("page_number")
        )
    )
    def pages_written(self):
        # a warning is emitted before this is run.
        return self.book_set.aggregate(pages_written=Sum("page_number"))[
            "pages_written"
        ]

for author in Author.objects.prepare(Author.pages_written):
    print(author.pages_written)

Dependent Annotated Properties

Often, annotations might depend on other annotations being present. If you pass an array of property names into the property constructor or decorator, all dependent annotatations will also be added to the queryset when you prepare the property using the other one:

class Author(Model):
    objects = Manager.from_queryset(PropertiedQuerySet)()

    pages_written = AnnotatedProperty(
        Subquery(
            Book.objects.filter(author=OuterRef("pk"))
            .values("author")
            .annotate(pages_written=Sum("page_number"))
            .values("page_number")
        )
    )

    twice_the_pages_written = AnnotatedProperty(
        F("pages_written") * Value(2), depends_on=["pages_written"]
    )

Calling Book.objects.prepare(Author.twice_the_pages_written) will now also annotate Book.pages_written. This allows you to change the underlying implementation of the annotations without changing queryset definitions everywhere. Neato.

Prefetched properties

A similar, yet less feature-complete thing can be done for prefetches:

class Author(Model):
    objects = Manager.from_queryset(PropertiedQuerySet)()
    short_books = PrefetchedProperty(
        "book_set", Book.objects.filter(page_number__lt=100)
    )


for author in Author.objects.prepare(Author.pages_written):
    print(author.short_books)

Since prefetches can't depend on eachother the depends_on kwarg is not supported for prefetches. The default getter is also not supported for now (django checks wether the attribute is present using hasattr before doing the prefetch, which would always execute the naive getter.)

As you can see, the prepare method doesn't care wether you pass it prefetches or annotations, so a property can change from annotation to prefetch or vice versa without changing the queryset definition or the model interface!