C++, because a copy of the template code is generated for each type parameter A. ML
compilers generally compile one copy of code that is used for all types A. (However, each
instance of the template expansion is smaller than the single copy generated by ML.)
ii. C++, because each copy of the code can be optimized for the specific type parameter A. In
ML, the code has to accept any type, so there are fewer opportunities for optimization,
and the code has to use pointers to values instead of values themselves so that all data
is the same size.