Constant sourcekitd timeouts in neovim with inlay hints enabled

[stancl]

Swift version

swift-driver version: 1.115.1 Apple Swift version 6.0.3 (swiftlang-6.0.3.1.10 clang-1600.0.30.1) Target: arm64-apple-macosx15.0

Platform

macOS 15.3.1

Editor

Neovim

Description

Hi, my sourcekit LSP keeps crashing/timing out (interrupting typing) in neovim. I ran sourcekit-lsp diagnose and noticed a bunch of "failed" checks relating to inlay hints tests. I tried turning them off using vim.lsp.inlay_hint.enable(false) and that makes the error go away. I'll say there still is some input lag when using this LSP but that's a minor issue in comparison. With inlay hints enabled an error message interrupts typing on essentially every 5 or so keystrokes.

The specific error message is: sourcekit: -32001: sourcekitd request timed out

I'm using just the default config for sourcekit. I also recorded a video reproduction so you can see what this looks like on my end https://asciinema.org/a/n7jOSeE0Cs9J8FUEksm7zruT3

Steps to Reproduce

Even just a hello world — project created by Xcode (seems like some project files are necessary to trigger the LSP) and opened in neovim with the LSP set up with default config.

Logging

log.txt

[ahoppen]

Synced to Apple’s issue tracker as rdar://145871554

[ahoppen]

Thanks for filing the issue. The lines that look most suspicious to me are Could not set up a build system at but it’s hard to tell what’s going wrong with the redacted logging. Would you be willing to enable extended logging, as described in https://github.com/swiftlang/sourcekit-lsp/blob/25af3dd/Documentation/Diagnose%20Bundle.md#enable-extended-logging, reproduce the issue and attache a new sourcekit-lsp diagnose bundle?

[stancl]

This should be it: log.txt

Let me know if I did it right, the file still seems free of any specifics though it did go up in size a lot.

[agemocui]

Looks like the issue is caused by the code at line 176 in SourceKitD.swift.

sourcekit-lsp/Sources/SourceKitD/SourceKitD.swift

Lines 176 to 178 in 7882228

	if sourcekitdResponse.error == .requestCancelled && !Task.isCancelled {
	throw SKDError.timedOut
	}

The intent seems to throw timedOut error when the request timed out but the task is not cancelled.

Should change to

if sourcekitdResponse.error == .timedOut && !Task.isCancelled {
  throw SKDError.timedOut
}

[agemocui]

Made a PR: https://github.com/swiftlang/sourcekit-lsp/pull/2026/files
Please help review.
Thanks.

[ahoppen]

Let me know if I did it right, the file still seems free of any specifics though it did go up in size a lot.

Looks like it didn’t pick up the extended logging. With extended logging, the recent log lines should no longer contain any <private> entries.

[agemocui]

@ahoppen

What I observed for this issue is:

This issue occurs when 2 (same) requests for inlayHint are sent to sourcekitd concurrently as follows:

1. Request A is sent to sourcekitd
2. Request B is sent to sourcekitd concurrently before response for request A returns.
3. sourcekitd returns .requestCancelled immediately for request A, and reaches to the code throwing timedOut. That’s why I suspected that ‘if’ check.
4. sourcekitd returns for request B without error

The error seen in neovim is due to that timedOut error when request A is cancelled by sourcekitd.

Sorry, just started learning swift for one week.

[ahoppen]

Thanks for that detailed analysis 🙏🏽 With that, it was obvious to me what’s happening: We implicitly cancel the sourcekitd request that’s backing inlay type hints when a new request comes in (https://github.com/swiftlang/swift/blob/b96e9f452987444c004be14693d4d31113cedee5/tools/SourceKit/lib/SwiftLang/SwiftSourceDocInfo.cpp#L2998) but we should disable that implicit cancellation here, similar to how we set cancelOnSubsequentRequest: 0 on other requests. I’ll work on a fix.

[carlocaione]

Hello, any news about this? This is hitting me hard :)

[ahoppen]

No, I didn’t get to it yet.

[Pierre-Aoki]

Same issue here !