perf(lsp): cancellable TS diagnostics #13565
Conversation
| @@ -217,6 +217,7 @@ impl DiagnosticsServer { | |||
| snapshot.clone(), | |||
| &config, | |||
| &ts_server, | |||
| token.clone(), | |||
There was a problem hiding this comment.
With this change we might want to consider removing or reducing the 200ms delay on line 208.
Edit: I'm just trying out the big file benchmark to see.
There was a problem hiding this comment.
there is also antidotal evidence too by using a debug build and editing a large project in vscode... I can try to take it for a spin later... though if we drop that, we should do it as a follow up PR, I suspect.
There was a problem hiding this comment.
That's a good idea! Though I just tested it now and removing the delay is still much slower ☹
Before (this PR):
Starting Deno benchmark
-> Start benchmarking lsp
- Simple Startup/Shutdown
(10 runs, mean: 508ms)
- Big Document/Several Edits
(5 runs, mean: 1879ms)
- Find/Replace
(10 runs, mean: 78ms)
- Code Lens
(10 runs, mean: 612ms)
<- End benchmarking lsp
After (debounce removed):
Starting Deno benchmark
-> Start benchmarking lsp
- Simple Startup/Shutdown
(10 runs, mean: 295ms)
- Big Document/Several Edits
(5 runs, mean: 4668ms)
- Find/Replace
(10 runs, mean: 78ms)
- Code Lens
(10 runs, mean: 402ms)
<- End benchmarking lsp
There was a problem hiding this comment.
The test might not be triggering it, because I don't think it is sending and setting cancellation tokens. I think we would need to modify the test, based on observed messages to and from the server.
Adds the ability to cancel generating TypeScript diagnostics while they're being generated. I'm not sure how to test this to see what kind of improvement this may bring, but I assume it will be helpful.