v2: recovery default handler changes

[amenzhinsky]

Hi there.

I'd like to discus the default panic handler in recovery package.

From my experience we always have a top level error handling interceptor that catches all non-grpc errors, sends them to a service like sentry and hides the details from users with status.Error(codes.Internal, "internal error"), so at least it should be fmt.Errorf("panic caught: %v", p).

The second thing is panic stack traces which are really important to track the root cause. The standard http server does it this way.

I do understand that this can be achieved using a custom panic handler but I suppose many people would like this changes and I would stop copy-pasting the following code to every project:

func recoverPanic(p interface{}) error {
	stack := make([]byte, 64<<10)
	stack = stack[:runtime.Stack(stack, false)]
	return &PanicError{Panic: p, Stack: stack}
}

type PanicError struct {
	Panic interface{}
	Stack []byte
}

func (e *PanicError) Error() string {
	return fmt.Sprintf("panic caught: %v\n\n%s", e.Panic, e.Stack)
}

[johanbrandhorst]

I think that behavior makes sense as the default behavior. Would you like to contribute this change?