{-# Language FlexibleContexts, FlexibleInstances,
MultiParamTypeClasses, RankNTypes, StandaloneDeriving,
TypeFamilies, TypeOperators, UndecidableInstances #-}
-- | An attribute grammar is a particular kind of 'Transformation' that assigns attributes to nodes in a
-- tree. Different node types may have different types of attributes, so the transformation is not natural. All
-- attributes are divided into 'Inherited' and 'Synthesized' attributes.
module Transformation.AG where
import qualified Rank2
import Transformation (Domain, Codomain)
import qualified Transformation.Deep as Deep
-- | Type family that maps a node type to the type of its attributes, indexed per type constructor.
type family Atts (f :: * -> *) a
-- | Type constructor wrapping the inherited attributes for the given transformation.
newtype Inherited t a = Inherited{inh :: Atts (Inherited t) a}
-- | Type constructor wrapping the synthesized attributes for the given transformation.
newtype Synthesized t a = Synthesized{syn :: Atts (Synthesized t) a}
deriving instance (Show (Atts (Inherited t) a)) => Show (Inherited t a)
deriving instance (Show (Atts (Synthesized t) a)) => Show (Synthesized t a)
-- | A node's 'Semantics' is a natural tranformation from the node's inherited attributes to its synthesized
-- attributes.
type Semantics t = Inherited t Rank2.~> Synthesized t
-- | An attribution rule maps a node's inherited attributes and its child nodes' synthesized attributes to the node's
-- synthesized attributes and the children nodes' inherited attributes.
type Rule t g = forall sem . sem ~ Semantics t
=> (Inherited t (g sem sem), g sem (Synthesized t))
-> (Synthesized t (g sem sem), g sem (Inherited t))
-- | The core function to tie the recursive knot, turning a 'Rule' for a node into its 'Semantics'.
knit :: (Rank2.Apply (g sem), sem ~ Semantics t) => Rule t g -> g sem sem -> sem (g sem sem)
knit r chSem = Rank2.Arrow knit'
where knit' inherited = synthesized
where (synthesized, chInh) = r (inherited, chSyn)
chSyn = chSem Rank2.<*> chInh
-- | The core type class for defining the attribute grammar. The instances of this class typically have a form like
--
-- > instance Attribution MyAttGrammar MyNode (Semantics MyAttGrammar) Identity where
-- > attribution MyAttGrammar{} (Identity MyNode{})
-- > (Inherited fromParent,
-- > Synthesized MyNode{firstChild= fromFirstChild, ...})
-- > = (Synthesized _forMyself,
-- > Inherited MyNode{firstChild= _forFirstChild, ...})
--
-- If you prefer to separate the calculation of different attributes, you can split the above instance into two
-- instances of the 'Transformation.AG.Generics.Bequether' and 'Transformation.AG.Generics.Synthesizer' classes
-- instead. If you derive 'GHC.Generics.Generic' instances for your attributes, you can even define each synthesized
-- attribute individually with a 'Transformation.AG.Generics.SynthesizedField' instance.
class Attribution t g deep shallow where
-- | The attribution rule for a given transformation and node.
attribution :: t -> shallow (g deep deep) -> Rule t g
-- | Drop-in implementation of 'Transformation.$'
applyDefault :: (q ~ Semantics t, x ~ g q q, Rank2.Apply (g q), Attribution t g q p)
=> (forall a. p a -> a) -> t -> p x -> q x
applyDefault extract t x = knit (attribution t x) (extract x)
{-# INLINE applyDefault #-}
-- | Drop-in implementation of 'Transformation.Full.<$>'
fullMapDefault :: (p ~ Domain t, q ~ Semantics t, q ~ Codomain t, x ~ g q q, Rank2.Apply (g q),
Deep.Functor t g, Attribution t g p p)
=> (forall a. p a -> a) -> t -> p (g p p) -> q (g q q)
fullMapDefault extract t local = knit (attribution t local) (t Deep.<$> extract local)
{-# INLINE fullMapDefault #-}