A declarative library for describing dependencies between data

Version on this page:
LTS Haskell 22.26:
Stackage Nightly 2024-06-22:
Latest on Hackage:

See all snapshots graphula appears in

MIT licensed
Maintained by Freckle Education
This version can be pinned in stack with:graphula-,2580

Module documentation for

Graphula Core

Graphula is a simple interface for generating persistent data and linking its dependencies. We use this interface to generate fixtures for automated testing. The interface is extensible and supports pluggable front-ends.

Arbitrary Data

Graphula utilizes QuickCheck to generate random data. We need to declare Arbitrary instances for our types.

share [mkPersist sqlSettings, mkMigrate "migrateAll"] [persistLowerCase|
  a String
  b Int
  deriving Show Eq Generic

  a AId
  b String
  deriving Show Eq Generic

  a AId
  b BId
  c String
  deriving Show Eq Generic

  Id UUIDKey
  a Int
  b String
  deriving Show Eq Generic

  Id DId sqltype=uuid
  a String
  deriving Show Eq Generic

  a Bool
  UniqueFA a
  deriving Show Eq Generic

instance Arbitrary A where
  arbitrary = A <$> arbitrary <*> arbitrary

instance Arbitrary B where
  arbitrary = B <$> arbitrary <*> arbitrary

instance Arbitrary C where
  arbitrary = C <$> arbitrary <*> arbitrary <*> arbitrary

instance Arbitrary D where
  arbitrary = D <$> arbitrary <*> arbitrary

instance Arbitrary E where
  arbitrary = E <$> arbitrary

instance Arbitrary F where
  arbitrary = F <$> arbitrary


We declare dependencies via the HasDependencies typeclass and its associated type Dependencies.

By default a type does not have any dependencies. We only need to declare an empty instance.

instance HasDependencies A
instance HasDependencies F

For single dependencies we use the Only type.

instance HasDependencies B where
  type Dependencies B = Only AId

Groups of dependencies use tuples. Declare these dependencies in the order they appear in the type. HasDependencies leverages generic programming to inject dependencies for you.

instance HasDependencies C where
  type Dependencies C = (AId, BId)

Non Sequential Keys

Graphula supports non-sequential keys with the KeySource associated type. To generate a key using its Arbitrary instance, use 'SourceArbitrary. Non-serial keys will need to also derive an overlapping Arbitrary instance.

instance HasDependencies D where
  type KeySource D = 'SourceArbitrary

deriving newtype instance {-# OVERLAPPING #-} Arbitrary (Key D)

You can also elect to always specify an external key using 'SourceExternal. This means that this type cannot be constructed with node; use nodeKeyed instead.

instance HasDependencies E where
  type KeySource E = 'SourceExternal

By default, HasDependencies instances use type KeySource _ = 'SourceDefault, which means that graphula will expect the database to provide a key.


Graphula allows logging of graphs via runGraphulaLogged. Graphula dumps graphs to a temp file on test failure.

loggingSpec :: IO ()
loggingSpec = do
    logFile :: FilePath
    logFile = "test.graphula"

    -- We'd typically use `runGraphulaLogged` which utilizes a temp file.
    failingGraph :: IO ()
    failingGraph = runGraphulaT Nothing runDB . runGraphulaLoggedWithFileT logFile $ do
      Entity _ a <- node @A () $ edit $ \n ->
        n {aA = "success"}
      liftIO $ aA a `shouldBe` "failed"

    `shouldThrow` anyException

  n <- lines <$> readFile "test.graphula"
  n `shouldSatisfy` (not . null)

Running It

simpleSpec :: IO ()
simpleSpec =
  runGraphulaT Nothing runDB $ do
    -- Type application is not necessary, but recommended for clarity.
    Entity aId _ <- node @A () mempty
    Entity bId b <- node @B (only aId) mempty
    Entity _ c <- node @C (aId, bId) $ edit $ \n -> n { cC = "edited" }
    Entity dId _ <- node @D () mempty
    Entity eId _ <- nodeKeyed @E (EKey dId) () mempty

    -- Do something with your data
    liftIO $ do
      cC c `shouldBe` "edited"
      cA c `shouldBe` bA b
      unEKey eId `shouldBe` dId

runGraphulaT carries frontend instructions. If we’d like to override them we need to declare our own frontend.

For example, a front-end that always fails to insert.

newtype GraphulaFailT m a = GraphulaFailT { runGraphulaFailT :: m a }
  deriving newtype (Functor, Applicative, Monad, MonadIO, MonadGraphulaBackend)

instance MonadGraphulaFrontend (GraphulaFailT m) where
  insert _ _ = pure Nothing
  remove = const (pure ())

insertionFailureSpec :: IO ()
insertionFailureSpec = do
    failingGraph :: IO ()
    failingGraph =  runGraphulaT Nothing runDB . runGraphulaFailT $ do
      Entity _ _ <- node @A () mempty
      pure ()
    `shouldThrow` (== (GenerationFailureMaxAttemptsToInsert (typeRep $ Proxy @A)))

Note that graphula can fail naturally if we define a graph that violates unique constraints in the database:

constraintFailureSpec :: IO ()
constraintFailureSpec = do
    failingGraph :: IO ()
    failingGraph =  runGraphulaT Nothing runDB $
      replicateM_ 3 $ node @F () mempty
    `shouldThrow` (== (GenerationFailureMaxAttemptsToInsert (typeRep $ Proxy @F)))

or if we define a graph with an unsatisfiable predicates:

ensureFailureSpec :: IO ()
ensureFailureSpec = do
    failingGraph :: IO ()
    failingGraph =  runGraphulaT Nothing runDB $ do
      Entity _ _ <- node @A () $ ensure $ \a -> a /= a
      pure ()
    `shouldThrow` (== (GenerationFailureMaxAttemptsToConstrain (typeRep $ Proxy @A)))





  • Release without dependencies upper bounds


  • Relax persistent packages’ upper bounds


  • Relax dependencies upper bounds


  • Support GHC-8.10 and unliftio-core 2.0.0


First tagged release.