Registry & Autodiscover

Django and many third-party apps use the autodiscover pattern for loading code across apps. The idea is to choose a module to check for across in each app and load it for some reason.

Avocado uses this pattern as well for:

  • formatters
  • translators
  • exporters
  • interfaces

In each of these cases, the goal is the same: to register classes or instances to be used as choices in the admin. This bridges the gap between code and data by referencing bits of code by name. This of course can be fragile as there are no checks or hard constraints if the underlying class names change.

To make use of this pattern simply do the following:

# foo/
from avocado.core import loader

# If the discovered modules need to register something, initialize
# the registry first
registry = loader.Registry()

# Kick off the autodiscovery. This module itself must be imported
# into a module that is loaded early on by Django such as

A module that would be discovered would look like this:

# bar/
from foo.mymodule import registry

class SomeClass(object):


To set a default value for the registry (or override an existing one), pass the default=True argument. This will set the name of the entry to 'Default'.

Buffered Paginator

Django provides a solid pagination API, but the Paginator relies on a list of objects to be passed in.

BufferedPaginator is a subclass of the Django Paginator class that allows for explicitly setting the _count attribute i.e. the total length of object_list. This removes the need for the Paginator to perform this calculation on object_list which enables reusing a pre-cached value.

from avocado.core.paginator import BufferedPaginator

# Buffered set of objects
buff_list = range(10)

# Total count of 100, 2 per page, offset of 40
paginator = BufferedPaginator(100, buff_list, offset=40, per_page=2)
paginator.num_page # 50
paginator.cached_page_indices() # 21, 26
paginator.cache_pages() # 5 # False # True # False

The object_list (named buff_list in the example) that is passed in as an argument represents a buffered part of the whole object_list. This works in conjunction with the BufferedPage which can test whether or not a particular page is actually stored in the BufferedPaginator object.

buf_size represents the size of the buffer, i.e. number of rows that will be available at any given time and in most cases the size of object_list assuming count is greater than buf_size.

If object_list is not passed in, all calculations will still be available, but any operations that act on the object_list will throw an error.

The use of the class is primarily for large data sets in which it is impractical to store the entire object_list into a paginator or if computing the count or slicing the list (even for QuerySets) is too costly.