The problem with strong typing

In the previous section, we learned that Fortran’s strong typing discipline prohibits passing arguments of incompatible data types to procedures. This means that when you write a function that expects a real number as input, you can’t simply pass an integer as an input argument. This would trigger a compile-time error. You can see this for yourself right now. Start with a basic function to compute the average of a one-dimensional real array:

pure real function average(x) result(res)
  real, intent(in) :: x(:)                 ❶
  res = sum(x) / size(x)                   ❷
end function average

❶ Real 1-D array

❷ Sum of all elements divided by the number of elements

Attempting to invoke average with an array of integers–say, average([1, 6, 4])–the compiler will report an error:

gfortran mod_average_incompatible.f90
mod_average_incompatible.f90:20:10:              ❶
 
  print *, average([1, 6, 4])                    ❷
          1
Error: Type mismatch in argument ‘x’ at (1);     ❸
       passed INTEGER(4) to REAL(4)              ❸

❶ File name, row, and column where the error occurred

❷ The source code that triggered the error

❸ The error message

In this case, the error message is quite helpful. The compiler tells us that there’s a type mismatch for argument x, as well as which data type was passed (INTEGER(4)) and which was expected (REAL(4)). Here, the number 4 corresponds to the default type kind of 4 bytes–int32 and real32 literal constants from the iso_fortran_env module. We could do due diligence and make sure that we pass an argument with a matching type to every procedure. However, being able to pass data to a function without having to worry about the type is convenient and will help you write shorter and more correct code.

To implement a generic function to compute an average of arrays of different data types, we’ll go through the following steps:

1. Write the specific functions for each data type; these functions must have unique names.
2. Write the interface (generic procedure) that points to the specific functions.
3. Make the interface publicly available in the module.
4. Apply the generic procedures to the data.