Podcast: Thinking Elixir 161 - Lexical LS and Elixir Dev UX

Episode 161 of Thinking Elixir. Language Servers underpin the language specific support we rely on in modern code editors. Lately, there have been new efforts in this area in the Elixir community. We talk with Steve Cohen about his project Lexical LS to learn about his new Elixir Language Server, how long he’s been at it, and what it can do today. We learn about some of the technical challenges when creating a tool like this, and why there’s still room for new projects in this area. Steve explains how Lexical LS is architected and that it is a goal to be easy to contribute to and work on the Lexical project itself. Time to give it a try!


This project just popped up on github as a recommendation for me. It looks interesting and I’ll definitely check out the podcast. I’m curious if anyone has used it with vim and if so, how the setup went.


I am currently struggling to get this installed myself. Keeps crashing on initial load on do_boot.


Adhering to the “lowest common denominator” caveats in the README of Elixir and Erlang versions fixed this for me.


I found it simplest to build my own binary like headmiser says, other wise it wouldn’t boot and swapping project versions backwards seemed more fraught than just building it. Building it was no drama, though the readme says build/... instead of _build/... for the output dir.

It does not support textDocument/hover (ie “show docs under cursor”) which I use quite a lot. I had a look at adding support but it seems you need to do more than add the defrequest "textDocument/hover" to Lexical.Protocol.Requests to let clients know the server capabilities, then I had to put it down and go back to work :slight_smile:

The completions seemed to be about as quick as ElixirLS.

Function picking from the completion menu are a bit different. ELS shows the full argument spec – including pattern matches (eg: fill(%S{} = s) plus the function @doc, where as LEX only shows the variable names (fill(s), no @doc, no patterns). However when selecting the function, ELS inserts fill(|) while LEX inserts fill(|s|) which is nice.

I have nvim setup to only show LSP/etc when I exit insert mode so I dont find the “real time” nature distracting, others might not enjoy it.

I think the formatter seemed to respect .formatter.exs, meaning it didn’t stick parens around ecto model fields which I think not possible with ELS last time I checked.

Inline credo sounds cool but it failed to build (when built by Lexical on start up), possibly because I self built Lexical.

Will probably check back in on it in a few months, though I have no real complaints with ElixirLS beyond the formatting thing, but I just run !mix format % instead.

1 Like