Tuesday 11 April 2023
Chris Kipp and the Scala Center team
Tooling is a pivotal part of the developer experience. Often when someone asks about a new language or how another developer enjoys working in a certain language, it’s not long before the topic of tooling comes up. Questions like “what’s the build tool like?” and “how’s the editor support?” are often some of the first questions asked, after talking about language features.
Earlier this year when the Center published the 2023 Scala Center Roadmap we mentioned our plans for a tooling summit this year. In the last few years a global pandemic prohibited us from meeting in person and an all-consuming Scala 3 release effort that required extra effort on tooling identified a deep need for something like this summit — to touch base, to map out current issues and opportunities in the tooling ecosystem, and to re-lay the groundwork for ongoing collaboration.
With all this in mind, we’d like to share some updates from the recent summit hosted by the Scala Center at EPFL in Lausanne, Switzerland.
Who was there
What originally started out as a plan to better sync between those working on the compiler and core tools at the Scala Center with those working on IntelliJ IDEA at JetBrains eventually blossomed into an event with over 40 attendees from around the world representing all sorts of companies and organization actively involved in the development and maintenance of core tooling in the Scala ecosystem. Below you will find a list of some of those companies and orgs:
- Netflix (Netflix Algorithms Engineering)
- Scala Center
- Xebia Functional
What we talked about
The Tooling Summit was spread over three full days of sessions around teamwork, de-duplication efforts, stabilization, and discussing the future of Scala tooling. Much of this time was spent brainstorming, discussing, and even hacking on various ideas that originated from a set of topics scheduled throughout the summit. You can find a list of these topics below:
- Ensure tooling can use structured diagnostics
- JetBrains / Scala Center education collaboration
- Scope and syntax of scala-cli using directives
- Debugging in Scala 3
- Bazel and Scala
- Standardizing Scala worksheet implementations
- The state of editor support for scala-cli
- A discussion around BSP
- Making tooling more robust
- Limiting the ways to import a project into your IDE
- Porting Scalameta to Scala 3
- IDE support in “polyglot” projects that contain Scala
- Tasty Query and its use-cases
- Planning for sbt 2
- Bringing the Compiler Academy format to tooling
- Shared publishing implementations for build tools
- Merge Bloop mainline and scala-cli’s fork of Bloop
Many of these topics resulted in issues being created, PRs being started, and new discussions and ideas popping up all around GitHub. Here’s a small overview and summary of the topics and their outcomes:
Ensure tooling can use structured diagnostics
We had a session on ensuring that the tooling around the ecosystem could properly consume and utilize structured diagnostics that come straight from the compiler. There was a talk on this subject during the past ScalaIO conference by Chris Kipp called The Journey of a Dotty diagnostic. While scala-cli already has some actionable diagnostics that allow you to update an outdated using directive, during the summit we talked about various “quick fixes” that the compiler could produce and even had a rough prototype to illustrate this using IntelliJ by the end of the summit!
The idea would be that fixes like this can come directly from the compiler ensuring that all tooling that interacts with it can benefit. You can follow the progress on this work in this Dotty mega issue on structured diagnostics.
JetBrains / Scala Center education collaboration
Both the Scala Center and JetBrains have educational efforts around helping newcomers to Scala get familiar with the language and the ecosystem. They’re not alone in this effort as multiple other companies like Xebia Functional also have courses and exercises to that end. As a result of this conversation there is a more concrete focus on communicating around the ecosystem about educational initiatives and also efforts to look more into utilizing tools like the JetBrains Educations Plugin.
Scope and syntax of scala-cli using directives
As scala-cli has is well on it’s way to becoming the new default
discussions around using directives continually come up. Questions such as what
should or shouldn’t be allowed, what format they should take, and what their
scope is are commonplace. Here’s some links that were referenced around this
topic and some that were a result of this discussion:
- SIP 46 scala-cli as the default Scala command
- The using_directives library
- Syntax issue around quoted strings
- Camel case issue
- Target directive renaming issue
Debugging in Scala 3
Debugging in Scala is a complex domain. This makes it all the more important to ensure we’re de-duplicating efforts where we can. The emergence of the scala-debug-adapter is a great example of de-duplicating the efforts around debugging for various tools. As a result of this conversation there will be more work towards utilizing the scala-debug-adapter in other tools that aren’t yet using it and also potentially inlining things like the Scala 3 expression evaluator right into the Dotty codebase. Here are a handful of other items that either came out of the discussion t the summit or were referenced:
- Better expose and publish the Scala 3 step filter
- Explore an interpreting evaluator
- Start the debugger in Metals instead of the build server
- A collection formatter
- goto definition for functions
Bazel and Scala
Many larger companies that use Scala are increasing their usage of Bazel for their polyglot mono-repo codebases. Many of these companies have specific needs unique to very large codebases that often result in bespoke solutions. The goal of this sessions was to discuss a few different aspects of this by focusing on the following areas:
- Why use Bazel?
- Competing implementations of rules_scala
- Difficult IDE support compared to other tools like sbt or Gradle
- Deployment strategies
While no concrete action has been taken yet, the conversation has started and will be sure to continue moving forward.
Standardizing Scala worksheet implementations
Currently in the Scala ecosystem we have 3 different approaches to worksheets:
- IntelliJ worksheets (plain and interactive mode) which re-use functionality from the Scala REPL
- Metals worksheets using mdoc as an engine
- Scastie which uses a custom macro implementation
This presents another great opportunity to de-duplicate the efforts here and provide a solution that is closer to the compiler that can be re-used by other tools. Here are a few outputs of the discussion:
- A thread on youtrack has been created about modern worksheets in IntelliJ.
- There will also be some research going into what a shared interface or worksheets could look like inside of the Scala 3 compiler.
- There will be some work to try and unify the way dependencies and scala options are included in worksheets to better align with using directives for example.
- The VS Code Notebook API will be explored to see if it can produce better output than the current Metals implementation with VS Code.
The state of editor support for scala-cli
Since scala-cli is still quite new, much of the conversation during this topic was getting all parties involved on the same page about how scala-cli works, what limitations currently exist, and how we can collectively tackle them. Here are a few action items and links from the discussion:
- scala-cli support in IntelliJ Scala Plugin
- We’d like to ensure syntax highlighting works for using directives in all the normal places you’d expect them to
- We’d like to ensure there is solid single-file support in various editors when using scala-cli. This is being looked into both in IntelliJ and Metals
A discussion around BSP
The Build Server Protocol (BSP) was envisioned as a unified way to compile, run, test, etc your code in a way that doesn’t make build clients or servers re-implement integrations for every tool similar to how the Language Server Protocol (LSP) works with language servers and your favorite text editor. While this started small with a single client (IntelliJ) and server (Bloop), we now have multiple client and server implementations for sbt, Mill, Bazel, Bloop, and more even extending to other languages outside of the JVM. In order to ensure a healthy future for BSP we’ve gone ahead and formalized the team structure a bit, are in the process of migrating to a Smithy as a source of truth, and are actively exploring a smoother BSP Discovery process.
Making tooling more robust
Having your tooling not work as expected is frustrating. Not knowing why it’s not working is even more frustrating. This conversation surrounds ideas and topics like:
- Better logging
- Fault tolerance
- Stress testing
As a result of these conversations the Metals team will be exploring usage of Sentry to better track when things are going wrong. The idea here would be to provide an experience similar to IntelliJ’s Exception Analyzer. LSP also has a telemetry event that could be utilized, but more research needs to be done here.
Limiting the ways to import a project into your IDE
Both Metals and IntelliJ users have multiple ways to import projects. This could be simply choosing your build server in Metals, or choosing to use BSP at all in IntelliJ. There are ongoing discussions and work going on in both of these tools to make the getting started experience on a project more seamless no matter the tool or project. For example you can see the discussion choosing a default build server for metals here.
Porting Scalameta to Scala 3
Scalameta is a core part of the tooling ecosystem being utilized by tools like Metals, Scalafmt, Scalafix, and mdoc. While Scalameta can handle parsing Scala 3 code, it can’t be used as a library in Scala 3 projects. The main hiccup in this effort is the extensive use of macros, which aren’t easily migrated. The main outcome of this discussion was simply better understanding the challenges involved, so that if someone is willing to put in the time here, we have a starting point and some research done.
IDE support in “polyglot” projects that contain Scala
Traditionally IntelliJ has been the leader in this front offering great support for mixed JVM-language projects. However this is made trickier with tools that may not always be well-supported in this context like sbt or Mill. There is a new upcoming BSP plugin for IntelliJ this may offer a better level of support for using Mill or sbt server for multi-language projects. On the Metals side, this is made even more trickier as LSP isn’t traditionally used for mixed multi-language projects. There will be some more research on seeing if it’s possible to better integrate with the Java language servers. There is an ongoing thread about this that can be followed here.
Tasty Query and its use-cases
Tasty Query is a newer tool in the ecosystem that is a compiler-independent way to semantically analyze TASTy files. Some concrete actions items that came out of this session to be worked on are below:
- Add a way to render trees and types as Scala code
- Identify/document which APIs are safe to use without a full classpath
Planning for sbt 2
sbt is the most widely used build tool for Scala, spanning small to giant projects. Many Scala developers daily interact with this tool and its plethora of plugins. Due to this, a lot of effort has been going into ensuring a smooth transition to sbt 2 which will boast a handful of new features focused on simplicity, interactivity, and stability. Eugene Yokota has put a ton of effort into this so far, so here’s a dump of relevant blogs and discussion around this that may interest you.
- sbt 2.0 ideas blog post
- sbt 2.0 ideas discussion
- RFC-1: sbt cache ideas blog post
- sbt cache ideas discussion
- RFC-2: sbt 2.0 RFC process blog post
- RFC process discussion
- RFC-3: drop custom config blog post
- drop custom config discussion
Bringing the Compiler Academy format to tooling
The Compiler Academy has been an ongoing effort to help onboard new contributors to the Scala 3 compiler. While the format has been a success, it’s been difficult to get the amount of compiler maintainers necessary to participate in the sprees. The discussion at the summit focused on how to continue the effort to get more active participants in the Scala 3 compiler development and also to look into taking the same format and applying it to other tools in the Scala ecosystem.
Shared publishing implementations for build tools
Across the ecosystem many tools have their own implementation to publish an artifact. For example, sbt, Mill, and scala-cli all use their own implementations. During this topic coursier/publish was mentioned as a possible library that could be utilized across the ecosystem. There were multiple questions that would need to be answered first about the reality of this possibility that are being tracked here.
Merge Bloop mainline and scala-cli’s fork of Bloop
For most users Bloop should just be an implementation detail. However, in the ecosystem we’ve frequently seen questions and comments about how it works, and questions about why it works a certain way. Some of the answers to these questions has led scala-cli to fork Bloop resulting in 2 bloop servers often running on a user’s machine if they are a Metals user. In order to combat this and provide a smoother Bloop experience for all users, we discussed what would need to happen in order to merge the fork of Bloop back inline. You can follow a thread on this effort here.
What comes next?
Overall, we consider the summit a success. With positive feedback from participants and movement already visible on the topics above, we feel confident that these efforts will result in a more stable, feature-rich, and unified tooling ecosystem. The Scala Center will be continuing to facilitate and plan further collaboration related to these topics and efforts.
We’re also looking ahead to another summit that will coincide with the upcoming Scala Days in Madrid. We’ll follow up on the discussion topics above and also dive into new ones.
There will be a discussion surrounding the content of this post on the contributors forum, so please engage with questions or comments.