Type Checking With Sorbet

The majority of the code in Homebrew is written in Ruby which is a dynamic language. To avail the benefits of static type checking, we have set up Sorbet in our codebase which provides the benefits of static type checking to dynamic languages like Ruby.

The Sorbet Documentation is a good place to get started if you want to dive deeper into Sorbet and its abilities.

Sorbet in the Homebrew Codebase

Inline type annotations

The sig method is used to annotate method signatures. Here’s a simple example:

class MyClass
  sig { params(name: String).returns(String) }
  def my_method(name)
    "Hello, #{name}!"
  end
end

With params we specify that there is a parameter name which must be a String, and with returns we specify that this method always returns a String.

For more information on how to express more complex types, refer to the official documentation:

• Method Signatures
• Class Types
• Nilable Types
• Union Types

Ruby interface files (.rbi)

RBI files help Sorbet learn about constants, ancestors and methods defined in ways it doesn’t understand natively. We can also create an RBI file to help Sorbet understand dynamic definitions. Some of these files are automatically generated (see the next section) and some are manually written, e.g. extend/on_system.rbi.

There are also a very small number of files that Homebrew loads before sorbet-runtime, such as utils/gems.rb. Those files cannot have type signatures alongside the code itself, so RBI files are used there instead to retain static type checking.

The Library/Homebrew/sorbet directory

The rbi directory contains all Ruby Interface (.rbi) files auto-generated by running brew typecheck --update:

• gems: RBI files for all gems are generated using Tapioca.
• dsl: RBI files auto-generated by our Tapioca compilers.
• upstream.rbi: This file is manually written and contains temporary workarounds for upstream Sorbet issues. It is typically empty.

The tapioca directory contains configuration files and compilers for Tapioca, allowing Sorbet to type check the dynamically generated components of the codebase.

The config file is a newline-separated list of arguments to pass to srb tc, the same as if they’d been passed on the command line. Arguments in the config file are always passed first, followed by arguments provided on the command line. We use it to ignore e.g. gem directories which we do not wish to type check.

Using brew typecheck

Every Ruby file in the codebase has a magic # typed: <level> comment at the top, where <level> is one of Sorbet’s strictness levels, usually false, true or strict. The false files only report errors related to the syntax, constant resolution and correctness of the method signatures, but no type errors. Our long-term goal is to move all false files to true and start reporting type errors on those files as well. Therefore, when adding new files, you should ideally mark it with # typed: true and work out any resulting type errors.

When run without any arguments, brew typecheck will run considering the strictness levels set in each of the individual Ruby files in the core Homebrew codebase. However, when run on a specific file or directory, more errors may show up since Sorbet cannot resolve constants defined outside the scope of the specified file. These problems can be solved with RBI files. Currently brew typecheck provides --quiet, --file, --dir and --ignore options, but you can explore more options with srb tc --help and pass them with srb tc.

Resolving Type Errors

Sorbet reports type errors along with an error reference code, which can be used to look up more information on how to debug the error, or what causes the error in the Sorbet Documentation. Here’s how to debug some common type errors:

• Using T.reveal_type: in files which are true or higher, by wrapping a variable or method call in T.reveal_type, Sorbet will show us what type it thinks that variable has in the output of srb tc. This is particularly useful when writing method signatures and debugging. Make sure to remove this line from your code before committing your changes, since this is just a debugging tool.

• One of the most frequent errors that we’ve encountered is 7003: Method does not exist. Since Ruby is a very dynamic language, methods can be defined in ways Sorbet cannot see statically. In such cases, check if the method exists at runtime; if not, then Sorbet has caught a future bug! But, it is also possible that even though a method exists at runtime, Sorbet cannot see it. In such cases, we use .rbi files.

• Since Sorbet does not automatically assume that Kernel is to be included in modules, we may encounter many errors while trying to use methods like puts, ohai, odebug etc. There are generally two approaches to fixing this:

  • If you are using module_function but never run include ModuleName anywhere, remove the module_definition and convert all methods to class methods (prepend the name with self.).
  • If you do include the module elsewhere, add a requires_ancestor to the module defining what types of classes this module can be included in. This may be as simple as a requires_ancestor { Kernel }, which most classes are descended from.

• The tips above are very generic and apply to lots of cases. For some common gotchas when using Sorbet, refer to the Sorbet Error Reference and FAQ.