Occasional huge non-GC stutters when saving files with rust-analyzer

[snyball]

Thank you for the bug report

• I am using the latest version of lsp-mode related packages.
• I checked FAQ and Troubleshooting sections
• You may also try reproduce the issue using clean environment using the following command: M-x lsp-start-plain

Bug description

I've nailed it down to lsp--text-document-did-save, which seems to be blocking occasionally, sometimes it takes as long as a couple of seconds. I've confirmed that it's not caused by the GC kicking in, by testing with (setq garbage-collection-messages t)

Steps to reproduce

• M-x lsp-start-plain
• Open a Rust file in a medium-sized (>10000LOC) project with lsp-mode
• Repeatedly make small changes and save
• You may have to repeat these steps tens of times before seeing a stutter

The stutters vary in length, from hundreds to thousands of milliseconds.

Expected behavior

I expected lsp--text-document-did-save to not block.

My hunch is that rust-analyzer occasionally takes a long time to process the didChange message, and will block when processing other input during this time. If a blocking lsp-mode call comes shortly after the non-blocking lsp--text-document-did-save, then that other synchronous call will be blocking the UI, when it would usually be quick. This explains the inconsistent behavior, as it would depend on when the various timers are triggered.

I don't observe this behavior with e.g Zig zls.

If that's the case, one simple hack would be to just stagger those timers after lsp--text-document-did-save, but this is obviously both not robust, and introduces an arbitrary delay to LSP feedback whenever a file is saved (perhaps unnecessarily if rust-analyzer processed the didChange request quickly.)

Which Language Server did you use?

rust-analyzer

OS

Linux

Error callstack

No response

Anything else?

No response