Restart Server: "Stopping timed out" on first click when server is hung

[ekazakov14]

Extension Version

0.20260421.1

VS Code Version

1.116.0

Operating system Version

macOS 26.3.1 (arm64)

Steps to reproduce

1. Open a TypeScript workspace and let the language server become unresponsive (in my case the server got stuck inside a long-running diagnostics refresh under heavy memory pressure — see log excerpt below).
2. Run TypeScript Native Preview: Restart Server from the command palette.
3. Observe: error popup Stopping the server timed out. Server does not restart. No new diagnostics, no hover, no completions.
4. Run the command again.
5. Observe: server restarts normally, no popup.

Deterministic reproduction without waiting for an organic hang: add time.Sleep(5 * time.Second) at the start of handleShutdown in internal/lsp/server.go, build, and trigger Restart Server.

Issue

When the tsgo language server becomes unresponsive, TypeScript Native Preview: Restart Server fails on the first attempt with an error popup (Stopping the server timed out) and does nothing else. Only a second invocation actually restarts the server.

This is the worst possible UX for this command: it breaks in exactly the scenario it exists for — an unresponsive server.

Log excerpt

2026-04-21 12:00:10.653 [info] Delaying scheduled diagnostics refresh...
2026-04-21 12:00:10.655 [info] Delaying scheduled diagnostics refresh...
2026-04-21 12:00:10.655 [info] Delaying scheduled diagnostics refresh...
2026-04-21 12:00:10.656 [info] Delaying scheduled diagnostics refresh...
2026-04-21 12:00:11.158 [info] Running scheduled diagnostics refresh
2026-04-21 13:20:00.595 [info]  Restarting language server...
2026-04-21 13:20:02.598 [error] Stopping server timed out
2026-04-21 13:21:48.083 [info]  Restarting language server...          ← second click
2026-04-21 13:21:48.093 [info]  Resolved client capabilities: { ... }  ← new server up

The 2-second gap between Restarting language server... and Stopping server timed out is vscode-languageclient's hardcoded 2000 ms shutdown timeout. The 1:48 gap between attempts is just user reaction time — any delay above ~2 s would have worked identically.

Related closed issues

The same problem has been reported and closed without a code-side fix at least three times:

• #1010 — closed by reporter as "seems to be fixed".
• #1487 — describes the exact same two-click sequence ("I need to remember to click restart again… 2 restarts with a pause inbetween"). Closed when an unrelated server-side change reduced the frequency of hangs, but the client-side timeout-handling bug itself was never addressed.
• #2213 — same Stopping server timed out popup. Closed for lack of reproducible steps.