Gradual typing of Ruby at Scale

Date: July 16, 2019
Speaker: Dmitry Petrashko

Abstract

Stripe maintains an extremely large and growing Ruby code base in which ~3/4 of Stripe’s engineers do most of their work. Continuing to scale development in that code base is one of the most critical tasks to maintaining product velocity and the productivity of Stripe engineering.

Based on a wide range of experiences, we believe that adding static types for a significant subset of that codebase helps developers understand code better, write code with more confidence, and detect+prevent significant classes of bugs.

This talk shares experience of Stripe successfully been building a typechecker for internal use, including core design decisions made in early days of the project and how they withstood reality of production use:

The choice of building a gradual type system, allowing different teams to adopt it independently on different pace, at cost of reduced guarantees;
The choice of using manifest types and leaning towards explicitness, to make it easier to read and maintain code, at cost of more verbose definitions;
The choice of using a nominal type system, to encourage development of interfaces where every type has to be declare all the interfaces upfront, at cost of ease of adoption in a duck-typed language;
The choice to build a very custom control flow dependent type checking to infer truthiness and organically support smart casts and pattern matching.