If you're in the same process as the code that had its invariant fail, how can you ensure that your process is in a safe state afterwards? A failed invariant, by definition, means that you're in uncharted territory - any anticipated failures are (assuming proper design) reported via the usual error mechanisms.