Write a function in R (that involves loops)

Have it take a vector of doubles and a length.

Compile it to LLVM

Then pass this to Python

If python is embedded in R or vice versa with Rpy2, then 
call the routine directly.  Pass reference to routine or module
and then allow Python call that.

If don't have two interpreters in same process, export the compile
routine to a file and allow Python to import it. Then call.

Work with NumPy strides, etc.




Lazy evaluation:
Example: 

f = function(x, y)
{
  g()
}

g = function(z, a)
{
  z
  a
}