catalogue: Super lightweight function registries for your library

catalogue is a tiny, zero-dependencies library that makes it easy to add function (or object) registries to your code. Function registries are helpful when you have objects that need to be both easily serializable and fully customizable. Instead of passing a function into your object, you pass in an identifier name, which the object can use to lookup the function from the registry. This makes the object easy to serialize, because the name is a simple string. If you instead saved the function, you'd have to use Pickle for serialization, which has many drawbacks.

⏳ Installation

pip install catalogue

conda install -c conda-forge catalogue

> ⚠️ Important note: catalogue v2.0+ is only compatible with Python 3.6+. > For Python 2.7+ compatibility, use catalogue v1.x.

👩‍💻 Usage

Let's imagine you're developing a Python package that needs to load data somewhere. You've already implemented some loader functions for the most common data types, but you want to allow the user to easily add their own. Using catalogue.create you can create a new registry under the namespace your_package → loaders.

# YOUR PACKAGE
import catalogue

loaders = catalogue.create("your_package", "loaders")

This gives you a loaders.register decorator that your users can import and decorate their custom loader functions with.

# USER CODE
from your_package import loaders

@loaders.register("custom_loader")
def custom_loader(data):
    # Load something here...
    return data

The decorated function will be registered automatically and in your package, you'll be able to access all loaders by calling loaders.get_all.

# YOUR PACKAGE
def load_data(data, loader_id):
    print("All loaders:", loaders.get_all()) # {"custom_loader": }
    loader = loaders.get(loader_id)
    return loader(data)

The user can now refer to their custom loader using only its string name ("custom_loader") and your application will know what to do and will use their custom function.

# USER CODE
from your_package import load_data

load_data(data, loader_id="custom_loader")

❓ FAQ

But can't the user just pass in the custom_loader function directly?

Sure, that's the more classic callback approach. Instead of a string ID, load_data could also take a function, in which case you wouldn't need a package like this. catalogue helps you when you need to produce a serializable record of which functions were passed in. For instance, you might want to write a log message, or save a config to load back your object later. With catalogue, your functions can be parameterized by strings, so logging and serialization remains easy – while still giving you full extensibility.

How do I make sure all of the registration decorators have run?

Decorators normally run when modules are imported. Relying on this side-effect can sometimes lead to confusion, especially if there's no other reason the module would be imported. One solution is to use entry points.

For instance, in spaCy we're starting to use function registries to make the pipeline components much more customizable. Let's say one user, Jo, develops a better tagging model using new machine learning research. End-users of Jo's package should be able to write spacy.load("jo_tagging_model"). They shouldn't need to remember to write import jos_tagged_model first, just to run the function registries as a side-effect. With entry points, the registration happens at install time – so you don't need to rely on the import side-effects.

🎛 API

function catalogue.create

Create a new registry for a given namespace. Returns a setter function that can be used as a decorator or called with a name and func keyword argument. If entry_points=True is set, the registry will check for Python entry points advertised for the given namespace, e.g. the entry point group spacy_architectures for the namespace "spacy", "architectures", in Registry.get and Registry.get_all. This allows other packages to auto-register functions.

architectures = catalogue.create("spacy", "architectures")

# Use as decorator
@architectures.register("custom_architecture")
def custom_architecture():
    pass

# Use as regular function
architectures.register("custom_architecture", func=custom_architecture)

class Registry

The registry object that can be used to register and retrieve functions. It's usually created internally when you call catalogue.create.

method Registry.__init__

Initialize a new registry. If entry_points=True is set, the registry will check for Python entry points advertised for the given namespace, e.g. the entry point group spacy_architectures for the namespace "spacy", "architectures", in Registry.get and Registry.get_all.

# User-facing API
architectures = catalogue.create("spacy", "architectures")
# Internal API
architectures = Registry(("spacy", "architectures"))

method Registry.__contains__

Check whether a name is in the registry.

architectures = catalogue.create("spacy", "architectures")

@architectures.register("custom_architecture")
def custom_architecture():
    pass

assert "custom_architecture" in architectures

method Registry.__call__

Register a function in the registry's namespace. Can be used as a decorator or called as a function with the func keyword argument supplying the function to register. Delegates to Registry.register.

method Registry.register

Register a function in the registry's namespace. Can be used as a decorator or called as a function with the func keyword argument supplying the function to register.

architectures = catalogue.create("spacy", "architectures")

# Use as decorator
@architectures.register("custom_architecture")
def custom_architecture():
    pass

# Use as regular function
architectures.register("custom_architecture", func=custom_architecture)

method Registry.get

Get a function registered in the namespace.

custom_architecture = architectures.get("custom_architecture")

method Registry.get_all

Get all functions in the registry's namespace.

all_architectures = architectures.get_all()
# {"custom_architecture": }

method Registry.get_entry_points

Get registered entry points from other packages for this namespace. The name of the entry point group is the namespace joined by _.

architectures = catalogue.create("spacy", "architectures", entry_points=True)
# Will get all entry points of the group "spacy_architectures"
all_entry_points = architectures.get_entry_points()

method Registry.get_entry_point

Check if registered entry point is available for a given name in the namespace and load it. Otherwise, return the default value.

architectures = catalogue.create("spacy", "architectures", entry_points=True)
# Will get entry point "custom_architecture" of the group "spacy_architectures"
custom_architecture = architectures.get_entry_point("custom_architecture")

method Registry.find

Find the information about a registered function, including the module and path to the file it's defined in, the line number and the docstring, if available.

import catalogue

architectures = catalogue.create("spacy", "architectures", entry_points=True)

@architectures("my_architecture")
def my_architecture():
    """This is an architecture"""
    pass

info = architectures.find("my_architecture")
# {'module': 'your_package.architectures',
#  'file': '/path/to/your_package/architectures.py',
#  'line_no': 5,
#  'docstring': 'This is an architecture'}

function catalogue.check_exists

Check if a namespace exists.