LeanCheck
LeanCheck is a simple enumerative property-based testing library. Properties
are defined as Haskell functions returning a boolean value which should be
True
for all possible choices of argument values. LeanCheck applies
enumerated argument values to these properties in search for a counterexample.
Properties can be viewed as parameterized unit tests.
LeanCheck works by producing tiers of test values: a possibly infinite list
of finite sublists of same-and-increasingly-sized values. This enumeration is
similar to Feat's. However, the ranking and ordering of values are defined
differently. The interface is also different.
Throughout this README lines that begin with the symbol >
indicate a line
entered into an interactive interpreter (ghci
). The result of evaluating the
expression is then printed on the following line.
LeanCheck implementation is easy to understand.
LeanCheck's core is under 190 lines of code.
Installing
To install the latest LeanCheck version from Hackage using Cabal v3.0 or later, just run:
$ cabal v1-update
$ cabal v1-install leancheck
If you are using a cabal earlier than v3.0, just run:
$ cabal update
$ cabal install leancheck
With Cabal v3.0 or later,
using v1-install
is preferred for the time being.
If all fails, you can start over by resetting your user's cabal installation
with:
rm -rf ~/.cabal/{bin,lib,logs,share,store} ~/.ghc/*/
WARNING: the above command will erase all user-local packages.
LeanCheck has (official) packages available on
Stackage,
OpenSUSE,
Gentoo,
Arch Linux and
NixOS.
Checking if properties are True
To check if properties are True,
just use the function holds
:: Testable a => Int -> a -> Bool
.
It takes two arguments:
the number of values to test
and a property (function returning Bool),
then, it returns a boolean indicating whether the property holds.
See (ghci):
> import Test.LeanCheck
> import Data.List
> holds 100 $ \xs -> sort (sort xs) == sort (xs::[Int])
True
> holds 100 $ \xs -> [] `union` xs == (xs::[Int])
False
As a rule-of-thumb, you should run holds for 500, 1 000, or 10 000 tests.
With more than that you may run out-of-memory depending on the types being
tested.
Finding counter examples
To find counter examples to properties,
you can use the function counterExample
:: Testable a => Int -> a -> Maybe [String]
.
It takes two arguments:
the number of values to test
and a property (function returning Bool).
Then, it returns Nothing if no results are found or Just a list of Strings
representing the offending arguments to the property.
See (ghci):
> import Test.LeanCheck
> import Data.List
> counterExample 100 $ \xs -> sort (sort xs) == sort (xs::[Int])
Nothing
> counterExample 100 $ \xs -> [] `union` xs == (xs::[Int])
Just ["[0,0]"]
> counterExample 100 $ \xs ys -> xs `union` ys == ys `union` (xs::[Int])
Just ["[]","[0,0]"]
Checking properties like in SmallCheck/QuickCheck
To "check" properties like in SmallCheck and QuickCheck
automatically printing results on standard output,
you can use the function check
:: Testable a => a -> IO ()
.
> import Test.LeanCheck
> import Data.List
> check $ \xs -> sort (sort xs) == sort (xs::[Int])
+++ OK, passed 200 tests.
> check $ \xs ys -> xs `union` ys == ys `union` (xs::[Int])
*** Failed! Falsifiable (after 4 tests):
[] [0,0]
The function check
tests for a maximum of 200 tests.
To check for a maximum of n
tests, use checkFor
n
.
To get a boolean result wrapped in IO
, use checkResult
or checkResultFor
.
There is no "quiet" option, just use holds
or counterExample
in that case.
Testing user-defined types
LeanCheck works on properties with Listable
argument types.
Listable
instances are declared similarly to SmallCheck:
data MyType = MyConsA
| MyConsB Int
| MyConsC Int Char
| MyConsD String
instance Listable MyType where
tiers = cons0 MyConsA
\/ cons1 MyConsB
\/ cons2 MyConsC
\/ cons1 MyConsD
For types that do not have a constraning data invariant, instances can be
automatically derived with Template Haskell by using deriveListable
like
so:
deriveListable ''MyType
The tiers
function return a potentially infinite list of finite sub-lists
(tiers). Each successive tier has values of increasing size.
tiers :: Listable a => [[a]]
For convenience, the function list
returns a potentially infinite list
of values of the bound type:
list :: Listable a => [a]
So, for example:
> take 5 (list :: [(Int,Int)])
[(0,0),(0,1),(1,0),(0,-1),(1,1)]
The list
function can be used to debug your custom instances.
Listable
class instances are more customizable than what is described here:
check source comments or haddock documentation for details.
Standard Listable Instances
LeanCheck comes out-of-the-box with Listable
instances for all types in the
Haskell 2010 Language Report with the intentional exception of a few types.
The leancheck-instances package aims to support types in the
Haskell Platform -- $ cabal install leancheck-instances
.
Providers for Tasty, test-framework and Hspec
The following providers allow including LeanCheck properties into
Tasty, test-framework or Hspec test suites.
Memory usage
Due to the way it is implemented (using lists of lists), LeanCheck can be quite
memory intensive if we set the maximum number of tests of a property to
millions of values (YMMV).
For the default maximum number of tests (200) you should be safe on most cases.
If you use 1 000 or 10 000 as the maximum number of tests for a property you're
also generally safe. More than that, it is in a hit or miss basis.
For more details, see LeanCheck memory usage.
Beginner friendliness
LeanCheck strives to be beginner/student friendly both in the interface and its
implementation. For instance, to understand LeanCheck's core, one does not
need to understand Monads as they aren't used at all there.
In the name of keeping the implementation easy to understand,
a compromise were made in terms of performance
(cf. LeanCheck memory usage).
LeanCheck is mostly Haskell 98 compliant and almost Haskell 2010 compliant.
With the exception of Listable
derivation modules (TH and Generics),
the only extension used by LeanCheck is CPP. This is to maintain
compatibility with different compilers. LeanCheck even compiles and runs on
Hugs98 from September 2006.
LeanCheck has 100% Haddock coverage with most functions having examples.
Further reading
For a detailed documentation of each function, see
LeanCheck's Haddock documentation.
For an introduction to property-based testing
and a step-by-step guide to LeanCheck, see the
tutorial on property-based testing with LeanCheck
(doc/tutorial.md
in the source repository).
LeanCheck is subject to a chapter in a PhD Thesis (2017).
LeanCheck has a list of frequently asked questions and answers.