Datatypes

Native types

Python	Fortran
int	int
float	real(4)
float64	real(8)
complex	complex(16)
tuple	static array
list	dynamic array

• Default type for floating numbers is double precision. Hence, the following stamtement

  x = 1.
  

  will be converted to

  real(8) :: x
  
  x = 1.0d0
  

Slicing

• When assigning a slice of tuple, we must allocate memory before (tuples are considered as static arrays). Therefor, the following python code

  a = (1, 4, 9, 16)
  c = a[1:]
  

  will be converted to

  integer :: a (0:3)
  integer, allocatable :: c(:)
  
  a = (/ 1, 4, 9, 16 /)
  c = allocate(c(1,3))
  c = a(1 : )
  

Todo

memory allocation within the scope of definition

Dynamic vs Static typing

Since our aim is to generate code in a low-level language, which is in most cases of static typed, we will have to devise an alternative way to construct/find the appropriate type of a given variable. This can be done by including the concept of constructors or use specific headers to assist Pyccel in finding/infering the appropriate type.

Let’s explain this more precisely; we consider the following code

n = 5
x = 2.0 * n

In this example, n will be interprated as an integer while x will be a double number, so everything is fine.

The problem arises when using a function, like in the following example

def f(n):
  x = 2.0 * n
  return x

n = 5
x = f(n)

Now the question is what would be the signature of f if there was no call to it in the previous script?

To overcome this ambiguity, we rewrite our function as

#$ header f(int)
def f(n):
  x = 2.0 * n
  return x

Such an implementation still makes sens inside Python. As you can see, the type of x is infered by analysing our expressions.