fortran-src-0.15.0: src/Language/Fortran/Common/Array.hs
{-# LANGUAGE UndecidableInstances #-} -- required due to instance design
module Language.Fortran.Common.Array where
import Control.DeepSeq ( NFData )
import GHC.Generics ( Generic )
import Data.Data ( Data, Typeable )
import Data.Binary ( Binary )
import Text.PrettyPrint.GenericPretty ( Out(..) )
import qualified Text.PrettyPrint as Pretty
import qualified Language.Fortran.PrettyPrint as F
-- | A single array dimension with bounds of type @a@.
--
-- * @'Num' a => 'Dim' a@ is a static, known-size dimension.
-- * @'Dim' ('Language.Fortran.AST.Expression' '()')@ is a dimension with
-- unevaluated bounds expressions. Note that these bounds may be constant
-- expressions, or refer to dummy variables, or be invalid.
-- * @'Num' a => 'Dim' ('Maybe' a)@ is a dimension where some bounds are
-- known, and others are not. This may be useful to record some information
-- about dynamic explicit-shape arrays.
data Dim a = Dim
{ dimLower :: a -- ^ Dimension lower bound.
, dimUpper :: a -- ^ Dimension upper bound.
} deriving stock (Show, Generic, Data, Eq)
deriving stock (Functor, Foldable, Traversable)
deriving anyclass (NFData, Binary)
-- | This instance is purely for convenience. No definition of ordering is
-- provided, and the implementation may change at any time.
deriving stock Ord
-- | Fortran syntax uses @lower:upper@, so only provide an 'Out' instance for
-- that style.
instance Out a => Out (Dim a) where
doc (Dim lb ub) = doc lb <> Pretty.char ':' <> doc ub
instance Out (Dim a) => F.Pretty (Dim a) where
pprint' _ = doc
-- | Fortran array dimensions, defined by a list of 'Dim's storing lower and
-- upper bounds.
--
-- You select the list type @t@ (which should be 'Functor', 'Foldable' and
-- 'Traversable') and the bound type @a@ (e.g. 'Int').
--
-- Using a non-empty list type such as 'Data.List.NonEmpty.NonEmpty' will
-- disallow representing zero-dimension arrays, providing extra soundness.
--
-- Note the following excerpt from the F2018 standard (8.5.8.2 Explicit-shape
-- array):
--
-- > If the upper bound is less than the lower bound, the range is empty, the
-- > extent in that dimension is zero, and the array is of zero size.
--
-- Note that the 'Foldable' instance does not provide "dimension-like" access to
-- this type. That is, @'length' (a :: 'Dims' t a)@ will _not_ tell you how many
-- dimensions 'a' represents. Use 'dimsLength' for that.
data Dims t a
-- | Explicit-shape array. All dimensions are known.
= DimsExplicitShape
(t (Dim a)) -- ^ list of all dimensions
-- | Assumed-size array. The final dimension has no upper bound (it is
-- obtained from its effective argument). Earlier dimensions may be defined
-- like explicit-shape arrays.
| DimsAssumedSize
(Maybe (t (Dim a))) -- ^ list of all dimensions except last
a -- ^ lower bound of last dimension
-- | Assumed-shape array. Shape is taken from effective argument. We store the
-- lower bound for each dimension, and thus also the rank (via list length).
| DimsAssumedShape
(t a) -- ^ list of lower bounds
deriving stock (Generic)
deriving stock (Functor, Foldable, Traversable)
-- We have to standalone derive most instances due to the @t@ list-like.
deriving stock instance (Show a, Show (t a), Show (t (Dim a)))
=> Show (Dims t a)
deriving anyclass instance (NFData a, NFData (t a), NFData (t (Dim a)))
=> NFData (Dims t a)
deriving stock instance (Data a, Data (t a), Data (t (Dim a)), Typeable t)
=> Data (Dims t a)
deriving stock instance (Eq a, Eq (t a), Eq (t (Dim a)))
=> Eq (Dims t a)
deriving anyclass instance (Binary a, Binary (t a), Binary (t (Dim a)))
=> Binary (Dims t a)
-- | This instance is purely for convenience. No definition of ordering is
-- provided, and the implementation may change at any time.
deriving stock instance (Ord a, Ord (t a), Ord (t (Dim a)))
=> Ord (Dims t a)
instance (Foldable t, Functor t, Out (Dim a), Out a)
=> Out (Dims t a) where
docPrec _ = doc
doc = Pretty.parens . \case
DimsExplicitShape ds ->
prettyIntersperse dimSep $ fmap doc ds
DimsAssumedShape ss ->
prettyIntersperse dimSep $ fmap go ss
where
go s = doc s <> Pretty.char ':'
DimsAssumedSize mds d ->
-- A bit fragile, but hopefully won't explode on empty 'Just's.
case mds of
Nothing -> prettyLast
Just ds -> prettyAfter dimSep (fmap doc ds) <> prettyLast
where
prettyLast = doc d <> Pretty.text ":*"
where
dimSep = Pretty.text ", "
instance Out (Dims t a) => F.Pretty (Dims t a) where
pprint' _ = doc
-- Faster is possible for non @[]@ list-likes, but this is OK for the general
-- case.
prettyIntersperse :: Foldable t => Pretty.Doc -> t Pretty.Doc -> Pretty.Doc
prettyIntersperse dBetween ds =
case foldMap (\d -> [dBetween, d]) ds of
[] -> mempty
_:ds' -> mconcat ds'
prettyAfter :: Foldable t => Pretty.Doc -> t Pretty.Doc -> Pretty.Doc
prettyAfter dAfter = foldMap (\d -> d <> dAfter)
-- | Traverse over the functor in a 'Dims' value with a functor bound type.
--
-- For example, to turn a @'Dims' t ('Maybe' a)@ into a @'Maybe' ('Dims' t a)@.
dimsTraverse :: (Traversable t, Applicative f) => Dims t (f a) -> f (Dims t a)
dimsTraverse = traverse id
-- TODO provide a SPECIALIZE clause for the above Maybe case. performance! :)
-- | How many dimensions does the given 'Dims' represent?
dimsLength :: Foldable t => Dims t a -> Int
dimsLength = \case
DimsExplicitShape ds -> length ds
DimsAssumedShape ss -> length ss
DimsAssumedSize mds _d ->
case mds of
Nothing -> 1
Just ds -> length ds + 1