Generic library design

[Marius Amado-Alves <amado.alves@gmail.com>]

Hello hive mind. Designing a few generic libraries involving real and complex number types and arrays thereof. Undecided as to the type of generic parameter(s). Namely:

When should the parameter(s) be the base real type like (examples drawn from the standard, please extrapolate)

   generic
      type Real is digits <>;
   package Ada.Numerics.Generic_Real_Arrays is
   ...

with the necessary sidekicks e.g. Generic_Elementary_Functions instantiated internally,

when should it be one or more packages like 

   generic
      with package Real_Arrays   is new
         Ada.Numerics.Generic_Real_Arrays   (<>);
      use Real_Arrays;
      with package Complex_Types is new
         Ada.Numerics.Generic_Complex_Types (Real);
      use Complex_Types;
   package Ada.Numerics.Generic_Complex_Arrays is
   ...

with the set or a subset of the necessary entities "pre-instantiated" by the user? Is there a design "rule"?

Consider the running example: why is Ada.Numerics.Generic_Complex_Arrays parametrized by packages, instead of just a Real type and then instantiate the packages intarnally?

And are compilers smart enough to merge identical instances w.r.t. to their generic arguments? Maybe with help of qualification pragmas telling there is no state?

No need to answer to all questions at once:-)

Thanks a lot.