Algebraic Data Types (ADTs) in Haskell can automatically become instances of some typeclasses (like Show, Eq) by deriving from them.
data Maybe a = Nothing | Just a
deriving (Eq, Ord)
My question is, how does this deriving work, i.e. how does Haskell know how to implement the functions of the derived typeclass for the deriving ADT?
Also, why is deriving restricted to certain typeclasses only? Why can't I write my own typeclass which can be derived?
The short answer is, magic :-). This is to say that automatic deriving is baked into the Haskell spec, and every compiler can choose to implement it in its own way. There's lots of work on how to make it extensible however.
Derive is a tool for Haskell to let you write your own deriving mechanisms.
GHC used to provide a derivable type class extension called Generic Classes, but it was rarely used, as it was somewhat weak. That has now been taken out, and work is ongoing to integrate a new generic deriving mechanism as described in this paper: http://www.dreixel.net/research/pdf/gdmh.pdf
For more on this, see:
From the Haskell 98 report:
The only classes in the Prelude for which derived instances are allowed are Eq, Ord, Enum, Bounded, Show, and Read...
Here's the description of how to derive these type classes: http://www.haskell.org/onlinereport/derived.html#derived-appendix
It is possible to use Template Haskell to generate instance declarations in a similar way to deriving-clauses.
The following example is shamelessly stolen from the Haskell Wiki:
In this example we use the following Haskell code
$(gen_render ''Body)to produce the following instance:
instance TH_Render Body where render (NormalB exp) = build 'normalB exp render (GuardedB guards) = build 'guardedB guardsThe function
gen_renderabove is defined as follows. (Note that this code must be in separate module from the above usage).-- Generate an intance of the class TH_Render for the type typName gen_render :: Name -> Q [Dec] gen_render typName = do (TyConI d) <- reify typName -- Get all the information on the type (type_name,_,_,constructors) <- typeInfo (return d) -- extract name and constructors i_dec <- gen_instance (mkName "TH_Render") (conT type_name) constructors -- generation function for method "render" [(mkName "render", gen_render)] return [i_dec] -- return the instance declaration -- function to generation the function body for a particular function -- and constructor where gen_render (conName, components) vars -- function name is based on constructor name = let funcName = makeName $ unCapalize $ nameBase conName -- choose the correct builder function headFunc = case vars of [] -> "func_out" otherwise -> "build" -- build 'funcName parm1 parm2 parm3 ... in appsE $ (varE $ mkName headFunc):funcName:vars -- put it all together -- equivalent to 'funcStr where funcStr CONTAINS the name to be returned makeName funcStr = (appE (varE (mkName "mkName")) (litE $ StringL funcStr))Which uses the following functions and types.
First some type synonyms to make the code more readable.
type Constructor = (Name, [(Maybe Name, Type)]) -- the list of constructors type Cons_vars = [ExpQ] -- A list of variables that bind in the constructor type Function_body = ExpQ type Gen_func = Constructor -> Cons_vars -> Function_body type Func_name = Name -- The name of the instance function we will be creating -- For each function in the instance we provide a generator function -- to generate the function body (the body is generated for each constructor) type Funcs = [(Func_name, Gen_func)]The main reusable function. We pass it the list of functions to generate the functions of the instance.
-- construct an instance of class class_name for type for_type -- funcs is a list of instance method names with a corresponding -- function to build the method body gen_instance :: Name -> TypeQ -> [Constructor] -> Funcs -> DecQ gen_instance class_name for_type constructors funcs = instanceD (cxt []) (appT (conT class_name) for_type) (map func_def funcs) where func_def (func_name, gen_func) = funD func_name -- method name -- generate function body for each constructor (map (gen_clause gen_func) constructors)A helper function of the above.
-- Generate the pattern match and function body for a given method and -- a given constructor. func_body is a function that generations the -- function body gen_clause :: (Constructor -> [ExpQ] -> ExpQ) -> Constructor -> ClauseQ gen_clause func_body data_con@(con_name, components) = -- create a parameter for each component of the constructor do vars <- mapM var components -- function (unnamed) that pattern matches the constructor -- mapping each component to a value. (clause [(conP con_name (map varP vars))] (normalB (func_body data_con (map varE vars))) []) -- create a unique name for each component. where var (_, typ) = newName $ case typ of (ConT name) -> toL $ nameBase name otherwise -> "parm" where toL (x:y) = (toLower x):y unCapalize :: [Char] -> [Char] unCapalize (x:y) = (toLower x):yAnd some borrowed helper code taken from Syb III / replib 0.2.
typeInfo :: DecQ -> Q (Name, [Name], [(Name, Int)], [(Name, [(Maybe Name, Type)])]) typeInfo m = do d <- m case d of d@(DataD _ _ _ _ _) -> return $ (simpleName $ name d, paramsA d, consA d, termsA d) d@(NewtypeD _ _ _ _ _) -> return $ (simpleName $ name d, paramsA d, consA d, termsA d) _ -> error ("derive: not a data type declaration: " ++ show d) where consA (DataD _ _ _ cs _) = map conA cs consA (NewtypeD _ _ _ c _) = [ conA c ] {- This part no longer works on 7.6.3 paramsA (DataD _ _ ps _ _) = ps paramsA (NewtypeD _ _ ps _ _) = ps -} -- Use this on more recent GHC rather than the above paramsA (DataD _ _ ps _ _) = map nameFromTyVar ps paramsA (NewtypeD _ _ ps _ _) = map nameFromTyVar ps nameFromTyVar (PlainTV a) = a nameFromTyVar (KindedTV a _) = a termsA (DataD _ _ _ cs _) = map termA cs termsA (NewtypeD _ _ _ c _) = [ termA c ] termA (NormalC c xs) = (c, map (\x -> (Nothing, snd x)) xs) termA (RecC c xs) = (c, map (\(n, _, t) -> (Just $ simpleName n, t)) xs) termA (InfixC t1 c t2) = (c, [(Nothing, snd t1), (Nothing, snd t2)]) conA (NormalC c xs) = (simpleName c, length xs) conA (RecC c xs) = (simpleName c, length xs) conA (InfixC _ c _) = (simpleName c, 2) name (DataD _ n _ _ _) = n name (NewtypeD _ n _ _ _) = n name d = error $ show d simpleName :: Name -> Name simpleName nm = let s = nameBase nm in case dropWhile (/=':') s of [] -> mkName s _:[] -> mkName s _:t -> mkName t
