Home

Analyse and test C with Python

16 August 2017

I’ve recently been implementing data structures in C. Writing C is fun and educational, because it’s lower level than most other common languages. It’s fast, and writing it teaches you to understand what’s happening at a machine level.

However, C is also relatively difficult to write, making it harder to analyse and test. It would be helpful to be able to do this with a higher level language, such as Python. Analysis and testing don’t affect performance of the actual data structure, so using a slower but easier and more productive language for this seems reasonable.

In this article, we walk through a simple example of doing this with a built-in Python library for interfacing with C called ctypes.

Example C code

For this example, I’ve written a simple (but unsafe) stack data structure, which defines the following API:

// stack.h
typedef struct {
    int stk[100];
    int top;
} stack;

void stack_push(stack* s, int x);
int stack_pop(stack* s);

The stack’s source code is included below1.

Shared library

ctypes interacts with shared libraries. We can compile our stack to a shared library with:

$ clang -shared -o stack.so -fPIC stack.c

Python wrapper

Next, we need to define a wrapper module2, which creates the Python objects used to interact with the shared library.

# stack_wrapper.py
import ctypes

class stack(ctypes.Structure):
    _fields_ = [
        ("stk", ctypes.c_int*100),
        ("top", ctypes.c_int),
    ]

_lib = ctypes.cdll.LoadLibrary("stack.so")

new = _lib.stack_new
new.restype = ctypes.POINTER(stack)

push = _lib.stack_push
push.argtypes = [ctypes.POINTER(stack), ctypes.c_int]

pop = _lib.stack_pop
pop.restype = ctypes.c_int
pop.argtypes = [ctypes.POINTER(stack)]

Note: It seems idiomatic from examples in the Python documentation to use the global import from ctypes import * to reduce code verbosity. I haven’t done this here to make it clear what objects come from ctypes.

Use

It’s now simple to import and use this module:

>>> import stack_wrapper as stack
>>> s = stack.new()
>>> stack.push(s, 5)
>>> stack.push(s, 15)
>>> stack.pop(s)
15
>>> stack.pop(s)
5

We’re now fully in Python land, with a set of objects which let us call into our C code. These objects are native Python, and can be used in any way that native Python objects can be used. We can use them with:


  1. Warning: This stack implementation was written to be as simple as possible, and isn’t really safe to use.

    // stack.c
    #include <stdlib.h>
    #include "stack.h"
    
    stack* stack_new() {
        stack* s = malloc(sizeof(stack));
        s->top = -1;
        return s;
    }
    
    void stack_push(stack* s, int x) {
        s->top++;
        s->stk[s->top] = x;
    }
    
    int stack_pop(stack* s) {
        int x = s->stk[s->top];
        s->top--;
        return x;
    }
    

  2. When using ctypes, we end up writing a Python wrapper module, which is basically a Python translation of the C header file. Other libraries for calling into C try to remove this repetition. One example is cffi, written by the PyPy team.