(added on 8/20) Summary for archaeologists from the future: a couple of the changes to package:vm_service made it such that any pending requests are completed with RPCErrors when the service is suddenly shut down (e.g. the user shut down their device during flutter run). Such errors should be captured by the tool and either discarded entirely or be captured and replaced by a helpful error message where appropriate.

(added on 11/22): Per #153471 (comment), there probably are legitimate bug(s) here, but I am not sure if they still exist in the 3.27 beta.

Tool crashes with messages of the form RPCError: <RPC method name>: (-32000) Service connection disposed have come to represent a significant number of tool crashes. On 3.24.0 to-date (8/24), 1870 crashes have been reported by 410-517 clients.

This exception comes with a few stack traces, hence this creating this umbrella issue.

• RPCError: readyToResume: (-32000) Service connection disposed (no stack trace) #153472 (affected 410 clients, but no stack trace)
  • This one has no stack trace. I wonder if this means this must come from VMService::dispose since that callsite does not provide a stack trace in its Completer::completeError call.
• RPCError: setLibraryDebuggable: (-32000) Service connection disposed #153473 (only affected 7 clients)
• RPCError: _flutter.listViews: (-32000) Service connection disposed #153474 (affected 100 clients)
  • Originates from VMService::_call. I wonder if the tool is making vm service calls after the vm service has been disposed and what might be causing this.

Edit: some more instances since we now get stack traces on these:

• RPCError: readyToResume: (-32000) Service connection disposed #154905
• RPCError: getIsolate: (-32000) Service connection disposed #154906
• RPCError: getVM: (-32000) Service connection disposed #154903