Cython

Chapter 10. Cython, NumPy, and Typed Memoryviews

All problems in computer science can be solved by another level of indirection, except, of course, for the problem of too many indirections.

— D. Wheeler

Two great qualities of Cython are its breadth and maturity: it compiles nearly all Python code (and whatever it cannot handle is typically straightforward to address); it brings the power and control of C’s type system to Python; and it integrates with external C and C++ code with ease. The task for this chapter is to round out Cython’s capabilities and cover Cython’s array features—which include support for NumPy arrays—in depth.

We have seen how well Cython supports built-in containers like list, tuple, dict, and set. These container objects are very easy to use, can contain any type of Python object, and are highly optimized for object lookup, assignment, and retrieval. The way the list type implements storage and retrieval is very different from dict, but from an implementation perspective, containers all have one thing in common: they all store references to Python objects. If we have a Python list of one million ints, every element in that list, at the C level, is a pointer to a boxed-up PyObject. Converting such a list to a C array of C ints is expensive, requiring us to iterate through the list and convert each PyObject to a C int, all the while doing proper error checking.

For homogeneous containers (e.g., a list containing nothing but floats), we can do much better in terms ...