json-autotype-1.0.12: Data/Aeson/AutoType/Format.hs
{-# LANGUAGE CPP #-}
{-# LANGUAGE TemplateHaskell #-}
{-# LANGuaGE ScopedTypeVariables #-}
{-# LANGUAGE ViewPatterns #-}
{-# LANGUAGE OverloadedStrings #-}
{-# LANGuaGE DeriveGeneric #-}
{-# LANGuaGE FlexibleContexts #-}
-- | Formatting type declarations and class instances for inferred types.
module Data.Aeson.AutoType.Format(
displaySplitTypes, splitTypeByLabel, unificationCandidates,
unifyCandidates,
normalizeTypeName
) where
import Control.Arrow ((&&&))
import Control.Applicative ((<$>), (<*>))
import Control.Lens.TH
import Control.Lens
import Control.Monad (forM)
import Control.Exception(assert)
import qualified Data.HashMap.Strict as Map
import Data.Monoid
import qualified Data.Set as Set
import qualified Data.Text as Text
import Data.Text (Text)
import Data.Set (Set )
import Data.List (foldl1')
import Data.Char (isAlpha, isDigit)
import Control.Monad.State.Class
import Control.Monad.State.Strict(State, runState)
import qualified Data.Graph as Graph
import GHC.Generics (Generic)
import Data.Aeson.AutoType.Type
import Data.Aeson.AutoType.Extract
import Data.Aeson.AutoType.Util ()
--import Debug.Trace -- DEBUG
trace _ x = x
fst3 :: (t, t1, t2) -> t
fst3 (a, _b, _c) = a
data DeclState = DeclState { _decls :: [Text]
, _counter :: Int
}
deriving (Eq, Show, Ord, Generic)
makeLenses ''DeclState
type DeclM = State DeclState
type Map k v = Map.HashMap k v
stepM :: DeclM Int
stepM = counter %%= (\i -> (i, i+1))
tShow :: (Show a) => a -> Text
tShow = Text.pack . show
-- | Wrap a type alias.
wrapAlias :: Text -> Text -> Text
wrapAlias identifier contents = Text.unwords ["type", identifier, "=", contents]
-- | Wrap a data type declaration
wrapDecl :: Text -> Text -> Text
wrapDecl identifier contents = Text.unlines [header, contents, " } deriving (Show,Eq,Generic)"]
--,"\nderiveJSON defaultOptions ''" `Text.append` identifier]
where
header = Text.concat ["data ", identifier, " = ", identifier, " { "]
-- | Explanatory type alias for making declarations
-- First element of the triple is original JSON identifier,
-- second element of the triple is the mapped identifier name in Haskell.
-- third element of the triple shows the type in a formatted way
type MappedKey = (Text, Text, Text, Bool)
-- | Make ToJSON declaration, given identifier (object name in Haskell) and mapping of its keys
-- from JSON to Haskell identifiers *in the same order* as in *data type declaration*.
makeFromJSON :: Text -> [MappedKey] -> Text
makeFromJSON identifier contents =
Text.unlines [
Text.unwords ["instance FromJSON", identifier, "where"]
, Text.unwords [" parseJSON (Object v) =", makeParser identifier contents]
, " parseJSON _ = mzero" ]
where
makeParser identifier [] = Text.unwords ["return ", identifier]
makeParser identifier _ = Text.unwords [identifier, "<$>", inner]
inner = " <*> " `Text.intercalate`
map takeValue contents
takeValue (jsonId, _, ty, True ) = Text.concat ["v .:?? \"", jsonId, "\""] -- nullable types
takeValue (jsonId, _, _ , False) = Text.concat ["v .: \"", jsonId, "\""]
-- Contents example for wrapFromJSON:
-- " <$>
--" v .: "hexValue" <*>
--" v .: "colorName\""
-- | Make ToJSON declaration, given identifier (object name in Haskell) and mapping of its keys
-- from JSON to Haskell identifiers in the same order as in declaration
makeToJSON :: Text -> [MappedKey] -> Text
makeToJSON identifier contents =
Text.unlines [
Text.concat ["instance ToJSON ", identifier, " where"]
, Text.concat [" toJSON (", identifier, " {", wildcard, "}) = object [", inner ", ", "]"]
#if MIN_VERSION_aeson(0,11,0)
, maybeToEncoding
#endif
]
where
maybeToEncoding | null contents = ""
| otherwise =
Text.concat [" toEncoding (", identifier, " {", wildcard, "}) = pairs (", inner "<>", ")"]
wildcard | null contents = ""
| otherwise = ".."
inner separator = separator `Text.intercalate`
map putValue contents
putValue (jsonId, haskellId, _typeText, _nullable) = Text.unwords [escapeText jsonId, ".=", haskellId]
escapeText = Text.pack . show . Text.unpack
-- Contents example for wrapToJSON
--"hexValue" .= hexValue
-- ,"colorName" .= colorName]
-- | Makes a generic identifier name.
genericIdentifier :: DeclM Text
genericIdentifier = do
i <- stepM
return $! "Obj" `Text.append` tShow i
-- * Printing a single data type declaration
newDecl :: Text -> [(Text, Type)] -> DeclM Text
newDecl identifier kvs = do attrs <- forM kvs $ \(k, v) -> do
formatted <- formatType v
return (k, normalizeFieldName identifier k, formatted, isNullable v)
let decl = Text.unlines [wrapDecl identifier $ fieldDecls attrs
,""
,makeFromJSON identifier attrs
,""
,makeToJSON identifier attrs]
addDecl decl
return identifier
where
fieldDecls attrList = Text.intercalate ",\n" $ map fieldDecl attrList
fieldDecl :: (Text, Text, Text, Bool) -> Text
fieldDecl (_jsonName, haskellName, fType, _nullable) = Text.concat [
" ", haskellName, " :: ", fType]
addDecl decl = decls %%= (\ds -> ((), decl:ds))
-- | Add new type alias for Array type
newAlias :: Text -> Type -> DeclM Text
newAlias identifier content = do formatted <- formatType content
addDecl $ Text.unlines [wrapAlias identifier formatted]
return identifier
-- | Convert a JSON key name given by second argument,
-- from within a dictionary keyed with first argument,
-- into a name of Haskell record field (hopefully distinct from other such selectors.)
normalizeFieldName :: Text -> Text -> Text
normalizeFieldName identifier = escapeKeywords .
uncapitalize .
(normalizeTypeName identifier `Text.append`) .
normalizeTypeName
keywords :: Set Text
keywords = Set.fromList ["type", "data", "module", "class", "where", "let", "do"]
escapeKeywords :: Text -> Text
escapeKeywords k | k `Set.member` keywords = k `Text.append` "_"
escapeKeywords k = k
-- | Format the type within DeclM monad, that records
-- the separate declarations on which this one is dependent.
formatType :: Type -> DeclM Text
formatType TString = return "Text"
formatType TNum = return "Double"
formatType TBool = return "Bool"
formatType (TLabel l) = return $ normalizeTypeName l
formatType (TUnion u) = wrap <$> case length nonNull of
0 -> return emptyTypeRepr
1 -> formatType $ head nonNull
_ -> Text.intercalate ":|:" <$> mapM formatType nonNull
where
nonNull = Set.toList $ Set.filter (TNull /=) u
wrap :: Text -> Text
wrap inner | TNull `Set.member` u = Text.concat ["(Maybe (", inner, "))"]
| otherwise = inner
formatType (TArray a) = do inner <- formatType a
return $ Text.concat ["[", inner, "]"]
formatType (TObj o) = do ident <- genericIdentifier
newDecl ident d
where
d = Map.toList $ unDict o
formatType e | e `Set.member` emptySetLikes = return emptyTypeRepr
formatType t = return $ "ERROR: Don't know how to handle: " `Text.append` tShow t
emptyTypeRepr :: Text
emptyTypeRepr = "(Maybe Value)" -- default, accepts future extension where we found no data
runDecl :: DeclM a -> Text
runDecl decl = Text.unlines $ finalState ^. decls
where
initialState = DeclState [] 1
(_, finalState) = runState decl initialState
-- * Splitting object types by label for unification.
type TypeTree = Map Text [Type]
type TypeTreeM a = State TypeTree a
addType :: Text -> Type -> TypeTreeM ()
addType label typ = modify $ Map.insertWith (++) label [typ]
splitTypeByLabel' :: Text -> Type -> TypeTreeM Type
splitTypeByLabel' _ TString = return TString
splitTypeByLabel' _ TNum = return TNum
splitTypeByLabel' _ TBool = return TBool
splitTypeByLabel' _ TNull = return TNull
splitTypeByLabel' _ (TLabel r) = assert False $ return $ TLabel r -- unnecessary?
splitTypeByLabel' l (TUnion u) = do m <- mapM (splitTypeByLabel' l) $ Set.toList u
return $! TUnion $! Set.fromList m
splitTypeByLabel' l (TArray a) = do m <- splitTypeByLabel' (l `Text.append` "Elt") a
return $! TArray m
splitTypeByLabel' l (TObj o) = do kvs <- forM (Map.toList $ unDict o) $ \(k, v) -> do
component <- splitTypeByLabel' k v
return (k, component)
addType l (TObj $ Dict $ Map.fromList kvs)
return $! TLabel l
-- | Splits initial type with a given label, into a mapping of object type names and object type structures.
splitTypeByLabel :: Text -> Type -> Map Text Type
splitTypeByLabel topLabel t = Map.map (foldl1' unifyTypes) finalState
where
finalize (TLabel l) = assert (l == topLabel) $ return ()
finalize topLevel = addType topLabel topLevel
initialState = Map.empty
(_, finalState) = runState (splitTypeByLabel' topLabel t >>= finalize) initialState
formatObjectType :: Text -> Type -> DeclM Text
formatObjectType identifier (TObj o) = newDecl identifier d
where
d = Map.toList $ unDict o
formatObjectType identifier other = newAlias identifier other
-- | Display an environment of types split by name.
displaySplitTypes :: Map Text Type -> Text
displaySplitTypes dict = trace ("displaySplitTypes: " ++ show (toposort dict)) $ runDecl declarations
where
declarations =
forM (toposort dict) $ \(name, typ) ->
formatObjectType (normalizeTypeName name) typ
-- | Normalize type name by:
-- 1. Treating all characters that are not acceptable in Haskell variable name as end of word.
-- 2. Capitalizing each word, but a first (camelCase).
-- 3. Adding underscore if first character is non-alphabetic.
-- 4. Escaping Haskell keywords if the whole identifier is such keyword.
-- 5. If identifier is empty, then substituting "JsonEmptyKey" for its name.
normalizeTypeName :: Text -> Text
normalizeTypeName s = ifEmpty "JsonEmptyKey" .
escapeKeywords .
escapeFirstNonAlpha .
Text.concat .
map capitalize .
filter (not . Text.null) .
Text.split (not . acceptableInVariable) $ s
where
ifEmpty x "" = x
ifEmpty _ nonEmpty = nonEmpty
acceptableInVariable c = isAlpha c || isDigit c
escapeFirstNonAlpha cs | Text.null cs = cs
escapeFirstNonAlpha cs@(Text.head -> c) | isAlpha c = cs
escapeFirstNonAlpha cs = "_" `Text.append` cs
capitalize :: Text -> Text
capitalize word = Text.toUpper first `Text.append` rest
where
(first, rest) = Text.splitAt 1 word
uncapitalize :: Text -> Text
uncapitalize word = Text.toLower first `Text.append` rest
where
(first, rest) = Text.splitAt 1 word
-- | Topological sorting of splitted types so that it is accepted declaration order.
toposort :: Map Text Type -> [(Text, Type)]
toposort splitted = map ((id &&& (splitted Map.!)) . fst3 . graphKey) $ Graph.topSort graph
where
(graph, graphKey) = Graph.graphFromEdges' $ map makeEntry $ Map.toList splitted
makeEntry (k, v) = (k, k, allLabels v)
-- | Computes all type labels referenced by a given type.
allLabels :: Type -> [Text]
allLabels = flip go []
where
go (TLabel l) ls = l:ls
go (TArray t) ls = go t ls
go (TUnion u) ls = Set.foldr go ls u
go (TObj o) ls = Map.foldr go ls $ unDict o
go _other ls = ls
-- * Finding candidates for extra unifications
-- | For a given splitted types, it returns candidates for extra
-- unifications.
unificationCandidates :: Map.HashMap t Type -> [[t]]
unificationCandidates = Map.elems .
Map.filter candidates .
Map.fromListWith (++) .
concatMap entry .
Map.toList
where
-- | Candidate entry has to have at least two candidates, so that unification makes sense
candidates [ ] = False
candidates [_] = False
candidates _ = True
-- | Make a candidate entry for each object type, which points from its keys to its label.
entry (k, TObj o) = [(Set.fromList $ Map.keys $ unDict o, [k])]
entry _ = [] -- ignore array elements and toplevel type if it is Array
-- | Unifies candidates on a give input list.
unifyCandidates :: [[Text]] -> Map Text Type -> Map Text Type
unifyCandidates candidates splitted = Map.map (remapLabels labelMapping) $ replacements splitted
where
unifiedType :: [Text] -> Type
unifiedType cset = foldr1 unifyTypes $
map (splitted Map.!) cset
replace :: [Text] -> Map Text Type -> Map Text Type
replace cset@(c:_) s = Map.insert c (unifiedType cset) (foldr Map.delete s cset)
replace [] _ = error "Empty candidate set in replace"
replacements :: Map Text Type -> Map Text Type
replacements s = foldr replace s candidates
labelMapping :: Map Text Text
labelMapping = Map.fromList $ concatMap mapEntry candidates
mapEntry cset@(c:_) = [(x, c) | x <- cset]
mapEntry [] = error "Empty candidate set in mapEntry"
-- | Remaps type labels according to a `Map`.
remapLabels :: Map Text Text -> Type -> Type
remapLabels ls (TObj o) = TObj $ Dict $ Map.map (remapLabels ls) $ unDict o
remapLabels ls (TArray t) = TArray $ remapLabels ls t
remapLabels ls (TUnion u) = TUnion $ Set.map (remapLabels ls) u
remapLabels ls (TLabel l) = TLabel $ Map.lookupDefault l l ls
remapLabels _ other = other