title | author | published |
---|---|---|
Test your JavaScript with QuickCheck |
Phil Freeman |
2015-07-16 |
QuickCheck is a property-based testing library which was originally written in Haskell, but which has been ported to a number of other languages.
QuickCheck works by generating random data with which to test your properties. This allows us to gain confidence that the properties hold, as more and more randomly-generated tests are run.
purescript-quickcheck
is a port of QuickCheck to PureScript, which preserves the syntax and types of the original Haskell code.
purescript-quickcheck
can be used to test code which is written in PureScript, but can be used to test Javascript functions as well. In this post, I'll demonstrate each of these use cases.
QuickCheck defines the quickCheck
function, which will print test results to the console, or fail with an exception in the event of a test failure.
Create a new project using Pulp, install purescript-quickcheck
, and open PSCi:
pulp init
bower install purescript-quickcheck
pulp repl
Start by importing the QuickCheck library:
> import Prelude
> import Test.QuickCheck
The quickCheck
function takes one argument: the property you would like to test. Let's try a very simple property:
> quickCheck \n -> n + 1 == 1 + n
If everything worked, you should see the following result:
100/100 test(s) passed.
unit
This indicates 100 successful random test runs.
Let's see what happens when we try testing a broken property:
> quickCheck \n -> n + 1 == n
You should see an exception printed to the console:
Error: Test 1 failed:
Failed: Test returned false
That's not a very helpful error, so let's improve it:
> quickCheck \n -> n + 1 == n <?> "Test failed for input " <> show n
This time you should see the following failure message:
Error: Test 1 failed:
Test failed for input -654791
Alternatively, we could use the ===
operator, which provides a better error message:
> quickCheck \n -> n + 1 === n
Error: Test 1 failed:
-663820 /= -663821
Let's write an implementation of the greatest common divisor function in PSCi.
This function is easiest to enter in the multi-line input mode, so switch to
that by entering :paste
at the prompt:
> :paste
… gcd 0 n = n
… gcd n 0 = n
… gcd n m | n < 0 = gcd (-n) m
… gcd n m | m < 0 = gcd n (-m)
… gcd n m | n > m = gcd (n - m) m
… gcd n m = gcd n (m - n)
…
After typing that in, press Control-D to finish the input.
Now let's assert some basic properties that we expect to hold of the gcd
function.
> quickCheck \n -> gcd n 1 === 1
This test should pass, but might take a while, because the standard random generator for integers which comes bundled with purescript-quickcheck
generates integers in the range -1000000 to 1000000.
We can modify our test to only consider small integers:
> quickCheck \n -> gcd (n / 1000) 1 === 1
This time, the test should complete quickly. However, we've coupled the generation of our data (/ 1000
) with the property we're testing, which is against the spirit of QuickCheck. A better approach is to define a newtype
which can be used to generate small integers.
Create a new file src/SmallInt.purs
and paste the following code:
module SmallInt where
import Prelude
import Test.QuickCheck
import Test.QuickCheck.Arbitrary
data SmallInt = SmallInt Int
runInt :: SmallInt -> Int
runInt (SmallInt i) = i
instance arbSmallInt :: Arbitrary SmallInt where
arbitrary = map (SmallInt <<< (_ / 1000)) arbitrary
Back in PSCi, we can now test properties without having to explicitly define how to generate our random data:
> quickCheck \(SmallInt n) (SmallInt m) -> gcd n m == gcd m n
The idea is that the particular scheme that is chosen to generate data should be indicated by the types of our function arguments, so newtype
s can be quite useful when defining multiple data generation schemes for a single type.
QuickCheck can also be used to test higher-order functions, by randomly generating functions.
Let's test that the map
function on arrays satisfies the functor laws.
For these two tests, I will write the test function using a let binding to avoid having to write type signatures in properties.
The first functor law says that if you map a function which does not modify its argument (the identity function) over a structure, then the structure should not be modified either.
> import Data.Array
> :paste
… firstFunctorLaw :: Array Int -> Boolean
… firstFunctorLaw arr = map id arr == arr
…
> quickCheck firstFunctorLaw
100/100 test(s) passed.
unit
The second functor law says that mapping two functions over a structure one-by-one is equivalent to mapping their composition over the structure:
> :paste
… secondFunctorLaw :: (Int -> Int) -> (Int -> Int) -> Array Int -> Boolean
… secondFunctorLaw f g arr = map f (map g arr) == map (f <<< g) arr
…
> quickCheck secondFunctorLaw
100/100 test(s) passed.
unit
Now let's try an example of testing a function written in Javascript.
This file contains a set of FFI bindings for some of the functions defined by the UnderscoreJS library. It is a nice example of a set of pure functions written in Javascript which we can test with QuickCheck.
Copy the contents of that file into src/UnderscoreFFI.purs
, and reload PSCi with that module loaded:
> import UnderscoreFFI
The UnderscoreFFI
module defines a wrapper for the sortBy
function. Let's test that the function is idempotent:
> :paste
… sortIsIdempotent :: Array Int -> Boolean
… sortIsIdempotent arr = sortBy id (sortBy id arr) == sortBy id arr
…
> quickCheck sortIsIdempotent
100/100 test(s) passed.
unit
In fact, we don't need to sort by the identity function. Since QuickCheck supports higher-order functions, we can test with a randomly-generated sorting function:
> :paste
… sortIsIdempotent' :: (Int -> Int) -> Array Int -> Boolean
… sortIsIdempotent' f arr = sortBy f (sortBy f arr) == sortBy f arr
…
> quickCheck sortIsIdempotent
100/100 test(s) passed.
unit
Have a look through the UnderscoreFFI
module, and see what other properties you can define.
Hopefully I've shown that QuickCheck can be a useful tool, whether you write your code in PureScript or not. Its strength is in its type-directed approach to data generation, which allows you to say what you want to test directly, rather than how to generate test data.