{-# LANGUAGE Trustworthy #-} {-# LANGUAGE CPP, NoImplicitPrelude, StandaloneDeriving, ScopedTypeVariables #-} {-# OPTIONS_HADDOCK not-home #-} ----------------------------------------------------------------------------- -- | -- Module : GHC.Read -- Copyright : (c) The University of Glasgow, 1994-2002 -- License : see libraries/base/LICENSE -- -- Maintainer : cvs-ghc@haskell.org -- Stability : internal -- Portability : non-portable (GHC Extensions) -- -- The 'Read' class and instances for basic data types. -- ----------------------------------------------------------------------------- module GHC.Read ( Read(..) -- class -- ReadS type , ReadS -- H2010 compatibility , lex , lexLitChar , readLitChar , lexDigits -- defining readers , lexP, expectP , paren , parens , list , choose , readListDefault, readListPrecDefault , readNumber , readField , readFieldHash , readSymField -- Temporary , readParen ) where #include "MachDeps.h" import qualified Text.ParserCombinators.ReadP as P import Text.ParserCombinators.ReadP ( ReadS , readP_to_S ) import qualified Text.Read.Lex as L -- Lex exports 'lex', which is also defined here, -- hence the qualified import. -- We can't import *anything* unqualified, because that -- confuses Haddock. import Text.ParserCombinators.ReadPrec import Data.Maybe import GHC.Unicode import GHC.Num import GHC.Real import GHC.Float import GHC.Show import GHC.Base import GHC.Arr import GHC.Word import GHC.List (filter) import GHC.Tuple (Solo (..)) -- | @'readParen' 'True' p@ parses what @p@ parses, but surrounded with -- parentheses. -- -- @'readParen' 'False' p@ parses what @p@ parses, but optionally -- surrounded with parentheses. readParen :: Bool -> ReadS a -> ReadS a -- A Haskell 2010 function readParen b g = if b then mandatory else optional where optional r = g r ++ mandatory r mandatory r = do ("(",s) <- lex r (x,t) <- optional s (")",u) <- lex t return (x,u) -- | Parsing of 'String's, producing values. -- -- Derived instances of 'Read' make the following assumptions, which -- derived instances of 'Text.Show.Show' obey: -- -- * If the constructor is defined to be an infix operator, then the -- derived 'Read' instance will parse only infix applications of -- the constructor (not the prefix form). -- -- * Associativity is not used to reduce the occurrence of parentheses, -- although precedence may be. -- -- * If the constructor is defined using record syntax, the derived 'Read' -- will parse only the record-syntax form, and furthermore, the fields -- must be given in the same order as the original declaration. -- -- * The derived 'Read' instance allows arbitrary Haskell whitespace -- between tokens of the input string. Extra parentheses are also -- allowed. -- -- For example, given the declarations -- -- > infixr 5 :^: -- > data Tree a = Leaf a | Tree a :^: Tree a -- -- the derived instance of 'Read' in Haskell 2010 is equivalent to -- -- > instance (Read a) => Read (Tree a) where -- > -- > readsPrec d r = readParen (d > app_prec) -- > (\r -> [(Leaf m,t) | -- > ("Leaf",s) <- lex r, -- > (m,t) <- readsPrec (app_prec+1) s]) r -- > -- > ++ readParen (d > up_prec) -- > (\r -> [(u:^:v,w) | -- > (u,s) <- readsPrec (up_prec+1) r, -- > (":^:",t) <- lex s, -- > (v,w) <- readsPrec (up_prec+1) t]) r -- > -- > where app_prec = 10 -- > up_prec = 5 -- -- Note that right-associativity of @:^:@ is unused. -- -- The derived instance in GHC is equivalent to -- -- > instance (Read a) => Read (Tree a) where -- > -- > readPrec = parens $ (prec app_prec $ do -- > Ident "Leaf" <- lexP -- > m <- step readPrec -- > return (Leaf m)) -- > -- > +++ (prec up_prec $ do -- > u <- step readPrec -- > Symbol ":^:" <- lexP -- > v <- step readPrec -- > return (u :^: v)) -- > -- > where app_prec = 10 -- > up_prec = 5 -- > -- > readListPrec = readListPrecDefault -- -- Why do both 'readsPrec' and 'readPrec' exist, and why does GHC opt to -- implement 'readPrec' in derived 'Read' instances instead of 'readsPrec'? -- The reason is that 'readsPrec' is based on the 'ReadS' type, and although -- 'ReadS' is mentioned in the Haskell 2010 Report, it is not a very efficient -- parser data structure. -- -- 'readPrec', on the other hand, is based on a much more efficient 'ReadPrec' -- datatype (a.k.a \"new-style parsers\"), but its definition relies on the use -- of the @RankNTypes@ language extension. Therefore, 'readPrec' (and its -- cousin, 'readListPrec') are marked as GHC-only. Nevertheless, it is -- recommended to use 'readPrec' instead of 'readsPrec' whenever possible -- for the efficiency improvements it brings. -- -- As mentioned above, derived 'Read' instances in GHC will implement -- 'readPrec' instead of 'readsPrec'. The default implementations of -- 'readsPrec' (and its cousin, 'readList') will simply use 'readPrec' under -- the hood. If you are writing a 'Read' instance by hand, it is recommended -- to write it like so: -- -- @ -- instance 'Read' T where -- 'readPrec' = ... -- 'readListPrec' = 'readListPrecDefault' -- @ class Read a where {-# MINIMAL readsPrec | readPrec #-} -- | attempts to parse a value from the front of the string, returning -- a list of (parsed value, remaining string) pairs. If there is no -- successful parse, the returned list is empty. -- -- Derived instances of 'Read' and 'Text.Show.Show' satisfy the following: -- -- * @(x,\"\")@ is an element of -- @('readsPrec' d ('Text.Show.showsPrec' d x \"\"))@. -- -- That is, 'readsPrec' parses the string produced by -- 'Text.Show.showsPrec', and delivers the value that -- 'Text.Show.showsPrec' started with. readsPrec :: Int -- ^ the operator precedence of the enclosing -- context (a number from @0@ to @11@). -- Function application has precedence @10@. -> ReadS a -- | The method 'readList' is provided to allow the programmer to -- give a specialised way of parsing lists of values. -- For example, this is used by the predefined 'Read' instance of -- the 'Char' type, where values of type 'String' should be are -- expected to use double quotes, rather than square brackets. readList :: ReadS [a] -- | Proposed replacement for 'readsPrec' using new-style parsers (GHC only). readPrec :: ReadPrec a -- | Proposed replacement for 'readList' using new-style parsers (GHC only). -- The default definition uses 'readList'. Instances that define 'readPrec' -- should also define 'readListPrec' as 'readListPrecDefault'. readListPrec :: ReadPrec [a] -- default definitions readsPrec = readPrec_to_S readPrec readList = readPrec_to_S (list readPrec) 0 readPrec = readS_to_Prec readsPrec readListPrec = readS_to_Prec (\_ -> readList) readListDefault :: Read a => ReadS [a] -- ^ A possible replacement definition for the 'readList' method (GHC only). -- This is only needed for GHC, and even then only for 'Read' instances -- where 'readListPrec' isn't defined as 'readListPrecDefault'. readListDefault = readPrec_to_S readListPrec 0 readListPrecDefault :: Read a => ReadPrec [a] -- ^ A possible replacement definition for the 'readListPrec' method, -- defined using 'readPrec' (GHC only). readListPrecDefault = list readPrec ------------------------------------------------------------------------ -- H2010 compatibility -- | The 'lex' function reads a single lexeme from the input, discarding -- initial white space, and returning the characters that constitute the -- lexeme. If the input string contains only white space, 'lex' returns a -- single successful \`lexeme\' consisting of the empty string. (Thus -- @'lex' \"\" = [(\"\",\"\")]@.) If there is no legal lexeme at the -- beginning of the input string, 'lex' fails (i.e. returns @[]@). -- -- This lexer is not completely faithful to the Haskell lexical syntax -- in the following respects: -- -- * Qualified names are not handled properly -- -- * Octal and hexadecimal numerics are not recognized as a single token -- -- * Comments are not treated properly lex :: ReadS String -- As defined by H2010 lex s = readP_to_S L.hsLex s -- | Read a string representation of a character, using Haskell -- source-language escape conventions. For example: -- -- > lexLitChar "\\nHello" = [("\\n", "Hello")] -- lexLitChar :: ReadS String -- As defined by H2010 lexLitChar = readP_to_S (do { (s, _) <- P.gather L.lexChar ; let s' = removeNulls s in return s' }) where -- remove nulls from end of the character if they exist removeNulls [] = [] removeNulls ('\\':'&':xs) = removeNulls xs removeNulls (first:rest) = first : removeNulls rest -- There was a skipSpaces before the P.gather L.lexChar, -- but that seems inconsistent with readLitChar -- | Read a string representation of a character, using Haskell -- source-language escape conventions, and convert it to the character -- that it encodes. For example: -- -- > readLitChar "\\nHello" = [('\n', "Hello")] -- readLitChar :: ReadS Char -- As defined by H2010 readLitChar = readP_to_S L.lexChar -- | Reads a non-empty string of decimal digits. lexDigits :: ReadS String lexDigits = readP_to_S (P.munch1 isDigit) ------------------------------------------------------------------------ -- utility parsers lexP :: ReadPrec L.Lexeme -- ^ Parse a single lexeme lexP = lift L.lex expectP :: L.Lexeme -> ReadPrec () expectP lexeme = lift (L.expect lexeme) expectCharP :: Char -> ReadPrec a -> ReadPrec a expectCharP c a = do q <- get if q == c then a else pfail {-# INLINE expectCharP #-} skipSpacesThenP :: ReadPrec a -> ReadPrec a skipSpacesThenP m = do s <- look skip s where skip (c:s) | isSpace c = get *> skip s skip _ = m paren :: ReadPrec a -> ReadPrec a -- ^ @(paren p)@ parses \"(P0)\" -- where @p@ parses \"P0\" in precedence context zero paren p = skipSpacesThenP (paren' p) paren' :: ReadPrec a -> ReadPrec a paren' p = expectCharP '(' $ reset p >>= \x -> skipSpacesThenP (expectCharP ')' (pure x)) parens :: ReadPrec a -> ReadPrec a -- ^ @(parens p)@ parses \"P\", \"(P0)\", \"((P0))\", etc, -- where @p@ parses \"P\" in the current precedence context -- and parses \"P0\" in precedence context zero parens p = optional where optional = skipSpacesThenP (p +++ mandatory) mandatory = paren' optional list :: ReadPrec a -> ReadPrec [a] -- ^ @(list p)@ parses a list of things parsed by @p@, -- using the usual square-bracket syntax. list readx = parens ( do expectP (L.Punc "[") (listRest False +++ listNext) ) where listRest started = do L.Punc c <- lexP case c of "]" -> return [] "," | started -> listNext _ -> pfail listNext = do x <- reset readx xs <- listRest True return (x:xs) choose :: [(String, ReadPrec a)] -> ReadPrec a -- ^ Parse the specified lexeme and continue as specified. -- Esp useful for nullary constructors; e.g. -- @choose [(\"A\", return A), (\"B\", return B)]@ -- We match both Ident and Symbol because the constructor -- might be an operator eg @(:~:)@ choose sps = foldr ((+++) . try_one) pfail sps where try_one (s,p) = do { token <- lexP ; case token of L.Ident s' | s==s' -> p L.Symbol s' | s==s' -> p _other -> pfail } -- See Note [Why readField] -- | 'Read' parser for a record field, of the form @fieldName=value@. The -- @fieldName@ must be an alphanumeric identifier; for symbols (operator-style) -- field names, e.g. @(#)@, use 'readSymField'). The second argument is a -- parser for the field value. readField :: String -> ReadPrec a -> ReadPrec a readField fieldName readVal = do expectP (L.Ident fieldName) expectP (L.Punc "=") readVal {-# NOINLINE readField #-} -- See Note [Why readField] -- | 'Read' parser for a record field, of the form @fieldName#=value@. That is, -- an alphanumeric identifier @fieldName@ followed by the symbol @#@. The -- second argument is a parser for the field value. -- -- Note that 'readField' does not suffice for this purpose due to -- . readFieldHash :: String -> ReadPrec a -> ReadPrec a readFieldHash fieldName readVal = do expectP (L.Ident fieldName) expectP (L.Symbol "#") expectP (L.Punc "=") readVal {-# NOINLINE readFieldHash #-} -- See Note [Why readField] -- | 'Read' parser for a symbol record field, of the form @(###)=value@ (where -- @###@ is the field name). The field name must be a symbol (operator-style), -- e.g. @(#)@. For regular (alphanumeric) field names, use 'readField'. The -- second argument is a parser for the field value. readSymField :: String -> ReadPrec a -> ReadPrec a readSymField fieldName readVal = do expectP (L.Punc "(") expectP (L.Symbol fieldName) expectP (L.Punc ")") expectP (L.Punc "=") readVal {-# NOINLINE readSymField #-} -- Note [Why readField] -- -- Previously, the code for automatically deriving Read instance (in -- typecheck/GHC.Tc.Deriv.Generate.hs) would generate inline code for parsing fields; -- this, however, turned out to produce massive amounts of intermediate code, -- and produced a considerable performance hit in the code generator. -- Since Read instances are not generally supposed to be performance critical, -- the readField and readSymField functions have been factored out, and the -- code generator now just generates calls rather than manually inlining the -- parsers. For large record types (e.g. 500 fields), this produces a -- significant performance boost. -- -- See also #14364. -------------------------------------------------------------- -- Simple instances of Read -------------------------------------------------------------- -- | @since 2.01 deriving instance Read GeneralCategory -- | @since 2.01 instance Read Char where readPrec = parens ( do L.Char c <- lexP return c ) readListPrec = parens ( do L.String s <- lexP -- Looks for "foo" return s +++ readListPrecDefault -- Looks for ['f','o','o'] ) -- (more generous than H2010 spec) readList = readListDefault -- | @since 2.01 instance Read Bool where readPrec = parens ( do L.Ident s <- lexP case s of "True" -> return True "False" -> return False _ -> pfail ) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance Read Ordering where readPrec = parens ( do L.Ident s <- lexP case s of "LT" -> return LT "EQ" -> return EQ "GT" -> return GT _ -> pfail ) readListPrec = readListPrecDefault readList = readListDefault -- | @since 4.11.0.0 deriving instance Read a => Read (NonEmpty a) -------------------------------------------------------------- -- Structure instances of Read: Maybe, List etc -------------------------------------------------------------- {- For structured instances of Read we start using the precedences. The idea is then that 'parens (prec k p)' will fail immediately when trying to parse it in a context with a higher precedence level than k. But if there is one parenthesis parsed, then the required precedence level drops to 0 again, and parsing inside p may succeed. 'appPrec' is just the precedence level of function application. So, if we are parsing function application, we'd better require the precedence level to be at least 'appPrec'. Otherwise, we have to put parentheses around it. 'step' is used to increase the precedence levels inside a parser, and can be used to express left- or right- associativity. For example, % is defined to be left associative, so we only increase precedence on the right hand side. Note how step is used in for example the Maybe parser to increase the precedence beyond appPrec, so that basically only literals and parenthesis-like objects such as (...) and [...] can be an argument to 'Just'. -} -- | @since 2.01 instance Read a => Read (Maybe a) where readPrec = parens (do expectP (L.Ident "Nothing") return Nothing +++ prec appPrec ( do expectP (L.Ident "Just") x <- step readPrec return (Just x)) ) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance Read a => Read [a] where {-# SPECIALISE instance Read [String] #-} {-# SPECIALISE instance Read [Char] #-} {-# SPECIALISE instance Read [Int] #-} readPrec = readListPrec readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Ix a, Read a, Read b) => Read (Array a b) where readPrec = parens $ prec appPrec $ do expectP (L.Ident "array") theBounds <- step readPrec vals <- step readPrec return (array theBounds vals) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance Read L.Lexeme where readPrec = lexP readListPrec = readListPrecDefault readList = readListDefault -------------------------------------------------------------- -- Numeric instances of Read -------------------------------------------------------------- readNumber :: Num a => (L.Lexeme -> ReadPrec a) -> ReadPrec a -- Read a signed number readNumber convert = parens ( do x <- lexP case x of L.Symbol "-" -> do y <- lexP n <- convert y return (negate n) _ -> convert x ) convertInt :: Num a => L.Lexeme -> ReadPrec a convertInt (L.Number n) | Just i <- L.numberToInteger n = return (fromInteger i) convertInt _ = pfail convertFrac :: forall a . RealFloat a => L.Lexeme -> ReadPrec a convertFrac (L.Ident "NaN") = return (0 / 0) convertFrac (L.Ident "Infinity") = return (1 / 0) convertFrac (L.Number n) = let resRange = floatRange (undefined :: a) in case L.numberToRangedRational resRange n of Nothing -> return (1 / 0) Just rat -> return $ fromRational rat convertFrac _ = pfail -- | @since 2.01 instance Read Int where readPrec = readNumber convertInt readListPrec = readListPrecDefault readList = readListDefault -- | @since 4.5.0.0 instance Read Word where readsPrec p s = [(fromInteger x, r) | (x, r) <- readsPrec p s] -- | @since 2.01 instance Read Word8 where readsPrec p s = [(fromIntegral (x::Int), r) | (x, r) <- readsPrec p s] -- | @since 2.01 instance Read Word16 where readsPrec p s = [(fromIntegral (x::Int), r) | (x, r) <- readsPrec p s] -- | @since 2.01 instance Read Word32 where #if WORD_SIZE_IN_BITS < 33 readsPrec p s = [(fromInteger x, r) | (x, r) <- readsPrec p s] #else readsPrec p s = [(fromIntegral (x::Int), r) | (x, r) <- readsPrec p s] #endif -- | @since 2.01 instance Read Word64 where readsPrec p s = [(fromInteger x, r) | (x, r) <- readsPrec p s] -- | @since 2.01 instance Read Integer where readPrec = readNumber convertInt readListPrec = readListPrecDefault readList = readListDefault -- | @since 4.8.0.0 instance Read Natural where readsPrec d = map (\(n, s) -> (fromInteger n, s)) . filter ((>= 0) . (\(x,_)->x)) . readsPrec d -- | @since 2.01 instance Read Float where readPrec = readNumber convertFrac readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance Read Double where readPrec = readNumber convertFrac readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Integral a, Read a) => Read (Ratio a) where readPrec = parens ( prec ratioPrec ( do x <- step readPrec expectP (L.Symbol "%") y <- step readPrec return (x % y) ) ) readListPrec = readListPrecDefault readList = readListDefault ------------------------------------------------------------------------ -- Tuple instances of Read, up to size 15 ------------------------------------------------------------------------ -- | @since 2.01 instance Read () where readPrec = parens ( paren ( return () ) ) readListPrec = readListPrecDefault readList = readListDefault -- | @since 4.15 deriving instance Read a => Read (Solo a) -- | @since 2.01 instance (Read a, Read b) => Read (a,b) where readPrec = wrap_tup read_tup2 readListPrec = readListPrecDefault readList = readListDefault wrap_tup :: ReadPrec a -> ReadPrec a wrap_tup p = parens (paren p) read_comma :: ReadPrec () read_comma = expectP (L.Punc ",") read_tup2 :: (Read a, Read b) => ReadPrec (a,b) -- Reads "a , b" no parens! read_tup2 = do x <- readPrec read_comma y <- readPrec return (x,y) read_tup4 :: (Read a, Read b, Read c, Read d) => ReadPrec (a,b,c,d) read_tup4 = do (a,b) <- read_tup2 read_comma (c,d) <- read_tup2 return (a,b,c,d) read_tup8 :: (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h) => ReadPrec (a,b,c,d,e,f,g,h) read_tup8 = do (a,b,c,d) <- read_tup4 read_comma (e,f,g,h) <- read_tup4 return (a,b,c,d,e,f,g,h) -- | @since 2.01 instance (Read a, Read b, Read c) => Read (a, b, c) where readPrec = wrap_tup (do { (a,b) <- read_tup2; read_comma ; c <- readPrec ; return (a,b,c) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d) => Read (a, b, c, d) where readPrec = wrap_tup read_tup4 readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e) => Read (a, b, c, d, e) where readPrec = wrap_tup (do { (a,b,c,d) <- read_tup4; read_comma ; e <- readPrec ; return (a,b,c,d,e) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f) => Read (a, b, c, d, e, f) where readPrec = wrap_tup (do { (a,b,c,d) <- read_tup4; read_comma ; (e,f) <- read_tup2 ; return (a,b,c,d,e,f) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g) => Read (a, b, c, d, e, f, g) where readPrec = wrap_tup (do { (a,b,c,d) <- read_tup4; read_comma ; (e,f) <- read_tup2; read_comma ; g <- readPrec ; return (a,b,c,d,e,f,g) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h) => Read (a, b, c, d, e, f, g, h) where readPrec = wrap_tup read_tup8 readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i) => Read (a, b, c, d, e, f, g, h, i) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; i <- readPrec ; return (a,b,c,d,e,f,g,h,i) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j) => Read (a, b, c, d, e, f, g, h, i, j) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j) <- read_tup2 ; return (a,b,c,d,e,f,g,h,i,j) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j, Read k) => Read (a, b, c, d, e, f, g, h, i, j, k) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j) <- read_tup2; read_comma ; k <- readPrec ; return (a,b,c,d,e,f,g,h,i,j,k) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j, Read k, Read l) => Read (a, b, c, d, e, f, g, h, i, j, k, l) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j,k,l) <- read_tup4 ; return (a,b,c,d,e,f,g,h,i,j,k,l) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j, Read k, Read l, Read m) => Read (a, b, c, d, e, f, g, h, i, j, k, l, m) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j,k,l) <- read_tup4; read_comma ; m <- readPrec ; return (a,b,c,d,e,f,g,h,i,j,k,l,m) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j, Read k, Read l, Read m, Read n) => Read (a, b, c, d, e, f, g, h, i, j, k, l, m, n) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j,k,l) <- read_tup4; read_comma ; (m,n) <- read_tup2 ; return (a,b,c,d,e,f,g,h,i,j,k,l,m,n) }) readListPrec = readListPrecDefault readList = readListDefault -- | @since 2.01 instance (Read a, Read b, Read c, Read d, Read e, Read f, Read g, Read h, Read i, Read j, Read k, Read l, Read m, Read n, Read o) => Read (a, b, c, d, e, f, g, h, i, j, k, l, m, n, o) where readPrec = wrap_tup (do { (a,b,c,d,e,f,g,h) <- read_tup8; read_comma ; (i,j,k,l) <- read_tup4; read_comma ; (m,n) <- read_tup2; read_comma ; o <- readPrec ; return (a,b,c,d,e,f,g,h,i,j,k,l,m,n,o) }) readListPrec = readListPrecDefault readList = readListDefault