anmonteiro Code ramblings

Reason and the Language Server Protocol: The Landscape

The state of editor tooling for programming languages has seen great change in recent memory thanks to Microsoft's specification of a Language Server Protocol (LSP).

While some editors bundle LSP out of the box, making it a breeze to get up and running, others take a little more tweaking. In a series of posts, I will detail the state of LSP editor tooling in Reason / OCaml, as well as how to get set up quickly in a variety of editors.

Language Server Protocol Implementations

There are currently two alternative LSP server implementations for the OCaml ecosystem. Despite what their names might suggest, all of them support both OCaml and Reason. Their goals and tradeoffs are described below.

1. ocaml-language-server

ocaml-language-server is the first implementation of an LSP server to have appeared for OCaml / Reason and made a few compromises in the name of shipping quickly and iterating fast:

  • It is implemented in TypeScript, and makes use of the upstream LSP implementation library by Microsoft for Node.js
  • It makes extensive use of existing tools for functionality such as code formatting, autocompletion and type information reporting.

    • These include, but are not limited to:

      • Merlin1 for code diagnostics, completion and type information reporting;
      • ocp-indent and refmt for OCaml and Reason code formatting, respectively;
      • Calls to the underlying build systems and package managers, including BuckleScript, the OCaml / Reason to JavaScript compiler.

2. reason-language-server

reason-language-server is a new implementation of the Language Server Protocol that is implemented in Reason and compiled to native code. It makes another set of trade-offs, which are detailed below, and is the implementation that will be covered in this series of posts.

  • It reduces the reliance on external tools such as Merlin – reason-language-server implements its own type information reporting and source code diagnostics.
  • Given it is written in Reason and compiled to native code, reason-language-server has implemented its own LSP protocol communication infrastructure because there wasn't one yet in the OCaml ecosystem.
  • It also includes support for bsb-native, a fork of BuckleScript that can build OCaml and Reason projects to OCaml bytecode and native assembly.

Setting up Reason in Visual Studio Code

The first editor that we are going to cover in this series is also the simplest one to set up. The following instructions assume that you have a working Visual Studio Code installation.

  1. Open VS Code and select the extensions panel.

  2. Look for the reason-vscode package by Jared Forsyth as seen in the following image2.

  3. Click "Install", wait until it finishes and click the "Reload" button to reload the workspace and apply your changes.

The above should be all you need! From now on, whenever you open a Reason or OCaml file you'll see the Reason Language Server in action: beyond the syntax highlighting provided by the extension, you'll see type hinting and error information if that is the case. Please open an issue in Reason Language Server in case anything has gone wrong.

Parting Thoughts and Next Posts

This post briefly covered the landscape of Language Server Protocol implementations in OCaml and Reason, as well as the tradeoffs each one embraces and the differences between them.

Next time we'll be covering how to set up the Reason LSP integration for Emacs. Please tweet @_anmonteiro with any questions or feedback about this post.

Happy hacking!


1 Merlin is an editor service that implements modern IDE features for OCaml and Reason. It solves the same problem as LSP, however there needs to be a specific implementation for every editor that intends to integrate with the Merlin protocol. LSP, however, only needs one implementation per editor in order to support every language that provides an LSP server. ↩
2 Alternatively, you can install the second package shown in the picture, which will install `ocaml-language-server`. That won't, however, be covered in this post. ↩