Yes if a process dies any message pipes connected to it in other processes
will eventually signal peer closure.

I'm a bit concerned about the "eventually" part. If say we call mojo::ReportBadMessage and subsequently drop the callback will we get any errors?

When you report a bad message for a request that expects a response, you
should also either invoke the response callback anyway, or close the
binding endpoint which received the request. If the endpoint is closed,
it's safe to drop callbacks.

rockot@, is it possible for ReportBadMesage to just close the message pipe in question when this happens? Or does mojo not have enough information on the stack to make that determination?

There isn't enough information in the stack, and it would also be weird
since the pipe is ultimately owned by a single ScopedMessagePipeHandle,
which is in turn owned at the bindings layer. Bad message reporting happens
below the bindings layer. We could always have ReportBadMessage take a
ScopedMessagePipeHandle and require the underlying C API to take a handle
as well. But then it's easy for users to unintentionally pass the wrong
pipe.