Collect telemetry on recovered notification panics and stderr on unrecovered panics

[andrewbranch]

Adds crash telemetry in two cases we were missing:

1. Recovered panics were only sent as LSP telemetry notifications during requests, not during notifications
2. Unrecovered panics were tracked with languageServer.connectionClosed, but lacked any error message. We now capture the last 40 lines of STDERR from the output channel and include that in the event.

[Copilot]

Pull request overview

Adds missing crash telemetry coverage for the TypeScript-Go language server, both for recovered panics (now including notifications) and for unrecovered crashes (now including recent stderr content from the VS Code output channel).

Changes:

• Always emit TelemetryEventInfo for recovered panics, regardless of whether the LSP message is a request or notification.
• Capture and report the last buffered stderr output when the language server connection closes unexpectedly.
• Extend the extension telemetry typing to include lastStderr for the languageServer.connectionClosed event.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
internal/lsp/server.go	Moves panic telemetry emission outside the request-only branch so notifications also send crash telemetry.
_extension/src/client.ts	Hooks output-channel error output to buffer stderr and includes it in languageServer.connectionClosed telemetry.
_extension/src/telemetryReporting.ts	Updates telemetry event typing to include lastStderr for connection-closed events.

[jakebailey]

I'm not sure we can send the stderr raw; is there specific text we think we are going to find that we can look for?

[andrewbranch]

That is not what @DanielRosenwasser told me in chat, unless I misunderstood

[DanielRosenwasser]

If we can pretty confidently validate that we have a stack trace, we can do some client-side sanitization like we used to. We have some stuff on the endpoint to get rid of PII just in case.

[andrewbranch]

Can I just collect everything between

and

?

What kind of client-side sanitization would you like me to do?

[jakebailey]

I don't know what kind of sanitization would be safe; for stack traces we know the exact format and can parse it to remove info. I don't know off the top of my head what's possible to go over stderr. We fall back to stderr when the client isn't available for logging yet, and I don't know what happens if say, the process OOMs and logs some info, or otherwise hits a big crash?

[andrewbranch]

7f439ae made it so we only process lines between panic: and Server exited and ports the server-side stack sanitization.

[jakebailey]

right, but I'm more thinking about if there's a panic followed by more text that some other goroutine is logging, interleaved. Maybe it just works?

[andrewbranch]

The sanitizer redacts lines that don't match a recognized pattern.