When is Manhattan distance potentially preferable to Euclidean distance?

A. When all features are class labels B. When coordinate-wise absolute differences are more appropriate or outlier sensitivity should be reduced C. When no features are numeric D. When association support is required