Juliacon 2024

Julia lacks a built-in method for defining strict interfaces. Instead, we use an ad-hoc combination of inheritance, traits, documentation, interface tests, and usually, run-time errors when something is not implemented!

At the same time we rely heavily on shared interfaces for high levels of package interop, and this combination makes better interfaces a common request from Julia developers.

There are a number of efforts underway exploring how interfaces could work better in Julia. This is just one of those, and certainly not the most sophisticated one!

The general assumptions of Interfaces.jl are that:

• Interfaces can enforce both run-time and compile-time constraints, e.g. it is reasonable to insist that a method must return a positive valued Int
• Interfaces often need optional components, such as mutability in AbstractArrays
• Interfaces can be tested for package types only if given a specific object with that type
• Interfaces definitions should provide traits other packages can check
• Interface implementations should be verified during tests or package precompilation
• The easiest way to define an interface is to simply provide a named list of functions that test all the mandatory and optional ways an object should behave.

We get these properties by defining a simple NamedTuples of functions, and a mechanism to make some of those names optional in the @interface and @implements declarations.

By providing data during @implements declarations we can also programmatically test interfaces implementations in any object, package or multiple packages with a single command.

Slides: https://rafaqz.github.io/JuliaCon2024-Interfaces/
Video: https://www.youtube.com/live/Bo3k4oD1Avg?feature=shared&t=27085