By Vanessa Nicole Naff for the GopherCon 2019 Liveblog on July 25, 2019
Presenter: Rebecca Stambler
Liveblogger: Vanessa Nicole Naff
The Go community has built many amazing tools to improve the Go developer experience. However, when a maintainer disappears or a new Go release wreaks havoc, the Go development experience becomes frustrating and complicated. This talk will cover the details behind a new tool to solve these issues: gopls (pronounced 'go please'). This tool is currently in development by the Go team and community, and it will ultimately serve as the backend for your Go editor.
When Rebecca first started writing code in Go, she was using Sublime, and she was fighting with her editor. Which extensions should she install to support her coding? Mystifying error messages were discouraging, and other editors were no less frustrating.
Now, after 3 years on the Go core team, she's figured it out. So how is the Go core team working to improve the experience for you?
First, we are asking:
Editors don't always have these features, which can slow us down. Rebecca learned a few years ago that even some engineers on the Go team at Google had never enabled autocomplete because it was so much overhead to figure out how.
So what are users doing and saying?
Rebecca and the core team categorized feedback responses from the Go community and identified some consistently-mentioned issues:
Overall, we're living with editors that don't fit our needs.
So let's deep-dive into some issues, but first:
Rebecca used VSCode as an example:
The Go extension is written in Typescript, so it's not doing direct analysis on our Go code.
It calls out to tools, like
godef. You could use CLI to do this yourself. It returns a position, and VSCode will move your cursor to that position.
But every tool has its own CLI, and return message format, and is maintained differently.
The VSCode extension uses 24 different Go tools.
We need it to be easier for all different editors to use all these tools.
Inefficiency is a major concern here. The VSCode Go extension doesn't set up a persistent server, so we create a new
godef process every time, starting from scratch. No work is shared for each request to
godef, even from a single file.
File is read -> Package found -> Dependency tree found -> All parsed and read
This works for a small package, but it can take multiple seconds to find all references for a large package.
We need a shared infrastructure between these tools to cache results and work faster!
image credit: KC Green
gocode as a real example here.
It was written in 2014 by the community, and is the default autocomplete tool for VSCode and vim.
But Go 1.10 added a build cache, which is incompatible with the way that
It had to be forked and rewritten to fix the broken user experience for Go version>=1.10.
Then it happened again with the introduction of go modules, which broke autocompletion and all tools, essentially.
The core team had to fork
gocode again and rewrite for modules support.
Now the VSCode extension has to choose between three forks of
gocode to provide autocomplete..
Key takeaway: The core team should invest effort not to break the engineer and maintainer experience with Go when releasing.
Rebecca explained that
gocode and all these other tools kept breaking because of the dependency on the behavior of
A tool should change when language changes, you'd think, but the core team had previously not managed the toolset directly, so the community had to catch up with the core team's release.
All the current stable Go developer tools mimic the go build system for parsing, typechecking, etc, and then provide helpful logic. Essentially, they are reverse-engineering the build process on the fly.
There is switching inside
go build so that it knows what to specifically to use for your build system! Are you using modules? dep? If your tool doesn't know where to look for your dependencies, it won't be able to typecheck anything.
We as a community want to stop tools from relying on reverse-engineering
Rebecca asserted that the Go core team needs to commit to supporting tools and the people that work with them.
The core team first took this on explicitly when they broke everyone's tools with the release of modules-- they began assisting the fixes for commonly-used community tools.
Rebecca here introduced the creation of a layer that will work with all types of dependency trees to avoid the dependency on
go build behavior.
It communicates with the Go packages driver, so we don't have to care about how dependencies are managed by the build. This reduces tool maintenance significantly, since now we can fix the server layer's interaction with the driver once instead of fixing 24+ tools if there's a breaking change from a language update.
What are the essential features for editor support; what should be supported? The core team had to ask these questions first.
Fortunately, there is a standard answer! The Language Server Protocol (LSP) already solves this problem in a standard way.
The LSP already defines core features to answer the questions the editor asks. A language server can then just route your request to the correct server when your language is identified, so if we have one for Go, we can support all the editors that already know how to use LSP. (That's mostly all of them.)
Its speed doesn't degrade with modules!
gopls will be easier to improve upon than other go toolsets: you just have to add a new function to add a new feature.
Rebecca also talked about a plan to add some other really cool support. Here's a sneak preview:
go vet, e.g.
It will also be reliable. You can hold the Go core team accountable for any issues simple by using the issue tracker.
Since the Go core team and the Go community are supporting and maintaining
gopls, you know it will be reliably worked. This tool wouldn't exist without the community!
The current development team is working to keep communication lines open:
Rebecca asked that you please send feature requests, feedback, and (if you want) contributions to
Get the most stable alpha and try it out if you wish:
go get golang.org/x/tools/[email protected]