Is there a bug in splitapply with a group of nans

The splitapply code calls localapply for missing data or data marked with nan. It seems nonsensical! If I mark data with nan, then it may be invalid and should not be called by local apply. In fact, splitapply calls localapply and then removes the results corresponding to data expost. It doesn't make sense, here is the code,

splitData = localsplit(dataVars,gnums,sgnums,gdim);

varargout = localapply(fun,splitData,gdim,nargout);

% Clean up NaN Groups if any(emptyIdx) for ii = 1:length(varargout) varargout{ii}(emptyGroup) = []; end end