Currently we expect to find suitable operation for any source we may choose. This works well when we only try to mutate code which was build from scratch by the fuzzer itself. However if we try to mutate pre-existing llvm ir we may encounter any possible operation.
I believe it's not practical to demand support for all of them. Instead we can apply same logic as we did when running deleter on empty function. By knowing that mutation attempt will be repeated many times we can bail from the single run and try to do it next time by choosing different operation source.