Blog
❮ Back to index

Using Haskell's QuickCheck for Python

19 Mar 2015 — Hacks
Writing proposals? Try Proppy, our new product!

Everyone I know who has ever used QuickCheck in anger cannot live without it.

The basic idea behind QuickCheck is that we write a proposition, e.g.: Reversing a list twice returns the same as the original list. In code:

reverse(reverse(x)) == x

QuickCheck will try to find a counterexample to that proposition by generating random values for x. And it is very good at finding mean values:

Shrinking

The really cool part is that when QuickCheck finds a counterexample it will attempt to make the example smaller. Lets say we have a broken reverse function that sorts the list instead. So for the input:

[2, 0, -1, 3, 9, 12]

our broken reverse returns:

[-1, 0, 2, 3, 9, 12]

instead of the correct:

[12, 9, 3, -1, 0, 2]

QuickCheck will shrink the input to [2, 0, -1] which still breaks but is much easier to debug. Neat!

Calling Python from Haskell

There is a relatively recent package on Hackage called pyfi. It calls Python through it's C FFI and passes values by converting them to and from JSON. It comes with a great tutorial on its github page.

QuickCheck

We start with a simple python module:

# moremath.py
def square(x):
    return x * x

In Haskell every call to Python happens in the IO monad so we have to use monadic QuickCheck:

-- squarecheck.hs
import Python
import Test.QuickCheck
import Test.QuickCheck.Monadic

square :: Int -> IO Int
square = defVV "from moremath import square as export"

main = do
    quickCheck $ monadicIO $ do
    v <- pick arbitrary
    r <- run $ square v
    assert $ r == v * v

Now we can run the tests like so:

$ runhaskell squarecheck.hs
+++ OK, passed 100 tests.

That's all there is to it!

Edit: Someone sent me a link to hypothesis which looks like a really solid Python version of QuickCheck. The API has been adjusted to fit into Python. Most importantly, it has a good shrinker!

Written by:

Tom Hunger