Seems sensible to me. I wonder if we should regard this as a breaking or minor-level change though, as it's possible someone was relying on the more specialized type for type inference. Admittedly that possibility seems a bit remote.

I think it should probably be a hard version change rather than minor; because the type signature clearly changes. However, anyone using the function would be restricted to site-based usage, in which the generator perturbed would be of the same type as the indexed value - such that, if replaced by my more general version, it should still compile a-ok.