[GSoC 2025] Improve the display of documentation during code completion in SourceKit-LSP

Development GSoC
1

Hello everyone!

I'm Ahmed Elrefaey, a third-year computer science student interested in contributing to Swift through GSoC 2025 on the project "Improve the display of documentation during code completion in SourceKit-LSP."

I've checked the Language Server Protocol's signature help specification and tried it with TypeScript on VS Code to get a feel of how it will be for Swift. I also used SourceKit LSP through the VS Code Swift extension to experience the first line only documentation status.

I've also set up the Swift and SourceKit LSP repositories and built Swift locally following the Getting Started guide and I've been exploring the codebase and trying to understand the different components.

My understanding is that SourceKit LSP implements the Language Server Protocol integrating with clangd and sourcekitd which lives in the Swift repository. Looking into it I found that CodeCompletion.cpp is mainly responsible for the code completion logic in collaboration with the parser which triggers the code completion callbacks when it reaches a tok::code_complete token with information from the surrounding context.

I still don't understand the whole flow though so I'd like to ask a few questions:

  1. How do I build sourcekitd locally and integrate it with SourceKit LSP (e.g. in the Swift VS Code extension)?

  2. How are code completion results generated? I came across CompletionLookup but I still don't understand how it's connected to other components.

  3. How and when is documentation for declarations collected and stored? I found DocComment.cpp but I'm not sure how it's used.

I found a potential improvement in SourceKit that I thought might help me better understand the codebase:

Adding code completion for unowned(|, where | is the cursor position, suggesting unsafe and safe. Similarly for nonisolated(unsafe) and access-modifier(set) like private(set). Would this be useful as a first contribution? If so, I'd appreciate any pointers on how to approach it.

Additionally, I found this GitHub issue [Swift 6] No diagnostic when explicitly capturing a non-sendable value which I think is interesting, would it be a good first contribution, or would you recommend something else to start with?

@ahoppen I'd love to hear your thoughts on things to focus more on at this stage to better understand the project and prepare a good proposal. Thanks in advance! :folded_hands:

2 Likes
2

We don’t really have an easy way of doing that right now. I would suggest using test cases in test/SourceKit and test/IDE to debug sourcekitd. The easiest way to test changes in VS Code would probably to open a PR, let CI build a toolchain and integrate that into SourceKit-LSP as described here.

For member and global completions, we use the compiler’s name lookup logic –search for lookup in CompletionLookup.cpp for the entry points.

If I remember correctly, we call something like getBriefComment in code completion right now. There should be an equivalent method to get the full comment.

Yes, that would be a good contribution. You can probably use [Completion] Completions for `inout`, `borrowing`, `consuming`, `isolated`, `consume`, and `copy` by hamishknight · Pull Request #74331 · swiftlang/swift · GitHub as a blueprint for adding such completions.

My guess would be that this is some malfunction inside the type checker, which probably isn’t the easiest first contribution.

1 Like