From 11cb6504d741c3859c46d25ac24ca6341108cb01 Mon Sep 17 00:00:00 2001
From: Tissevert <tissevert+devel@marvid.fr>
Date: Fri, 24 May 2019 10:48:09 +0200
Subject: [PATCH] Go strict ByteStrings with attoparsec

---
 ChangeLog.md                           |   5 ++
 Hufflepdf.cabal                        |  11 +--
 src/Data/ByteString/Char8/Util.hs      |  16 ++++
 src/Data/ByteString/Lazy/Char8/Util.hs |  17 ----
 src/PDF.hs                             |  44 +++++-----
 src/PDF/Body.hs                        |  46 +++++------
 src/PDF/EOL.hs                         |   9 ++-
 src/PDF/Object.hs                      | 107 +++++++++++++------------
 src/PDF/Output.hs                      |  18 ++---
 src/PDF/Parser.hs                      |  98 ++++++++++++++++++++++
 10 files changed, 237 insertions(+), 134 deletions(-)
 create mode 100644 src/Data/ByteString/Char8/Util.hs
 delete mode 100644 src/Data/ByteString/Lazy/Char8/Util.hs
 create mode 100644 src/PDF/Parser.hs

diff --git a/ChangeLog.md b/ChangeLog.md
index 65f4d13..6bde38f 100644
--- a/ChangeLog.md
+++ b/ChangeLog.md
@@ -1,5 +1,10 @@
 # Revision history for Hufflepdf
 
+## 0.1.1.0  -- 2019-05-23
+
+* Rewrite the parser using attoparsec instead of parsec (about 1.5x faster)
+* Use strict ByteStrings instead of lazy ones for an additional gain in performance, simpler type interface (lazy ByteStrings are now used only for output because that's required by the ByteString Builder) and fewer import lines
+
 ## 0.1.0.0  -- 2019-05-18
 
 * First version. Released on an unsuspecting world.
diff --git a/Hufflepdf.cabal b/Hufflepdf.cabal
index 2236845..c695de0 100644
--- a/Hufflepdf.cabal
+++ b/Hufflepdf.cabal
@@ -17,17 +17,18 @@ cabal-version:       >=1.10
 
 library
   exposed-modules:     PDF
-                     , PDF.Object
-  other-modules:       PDF.Body
                      , PDF.EOL
+                     , PDF.Object
                      , PDF.Output
-                     , Data.ByteString.Lazy.Char8.Util
+  other-modules:       Data.ByteString.Char8.Util
+                     , PDF.Body
+                     , PDF.Parser
   -- other-extensions:    
-  build-depends:       base >=4.9 && <4.13
+  build-depends:       attoparsec
+                     , base >=4.9 && <4.13
                      , bytestring
                      , containers
                      , mtl
-                     , parsec
   hs-source-dirs:      src
   ghc-options:         -Wall
   default-language:    Haskell2010
diff --git a/src/Data/ByteString/Char8/Util.hs b/src/Data/ByteString/Char8/Util.hs
new file mode 100644
index 0000000..c091351
--- /dev/null
+++ b/src/Data/ByteString/Char8/Util.hs
@@ -0,0 +1,16 @@
+module Data.ByteString.Char8.Util (
+      previous
+    , subBS
+  ) where
+
+import Data.ByteString.Char8 (ByteString)
+import qualified Data.ByteString.Char8 as BS (drop, index, take)
+import Prelude hiding (length)
+
+previous :: Char -> Int -> ByteString -> Int
+previous char position byteString
+  | BS.index byteString position == char = position
+  | otherwise = previous char (position - 1) byteString
+
+subBS :: Int -> Int -> ByteString -> ByteString
+subBS offset length = BS.take length . BS.drop offset
diff --git a/src/Data/ByteString/Lazy/Char8/Util.hs b/src/Data/ByteString/Lazy/Char8/Util.hs
deleted file mode 100644
index 831be25..0000000
--- a/src/Data/ByteString/Lazy/Char8/Util.hs
+++ /dev/null
@@ -1,17 +0,0 @@
-module Data.ByteString.Lazy.Char8.Util (
-      previous
-    , subBS
-  ) where
-
-import Data.ByteString.Lazy.Char8 (ByteString)
-import qualified Data.ByteString.Lazy.Char8 as BS (drop, index, take)
-import Data.Int (Int64)
-import Prelude hiding (length)
-
-previous :: Char -> Int64 -> ByteString -> Int64
-previous char position byteString
-  | BS.index byteString position == char = position
-  | otherwise = previous char (position - 1) byteString
-
-subBS :: Int64 -> Int64 -> ByteString -> ByteString
-subBS offset length = BS.take length . BS.drop offset
diff --git a/src/PDF.hs b/src/PDF.hs
index 8c96efd..6c3dffd 100644
--- a/src/PDF.hs
+++ b/src/PDF.hs
@@ -5,12 +5,12 @@ module PDF (
     , render
   ) where
 
-import Data.ByteString.Lazy.Char8 (ByteString)
-import qualified Data.ByteString.Lazy.Char8 as BS (
+import Data.ByteString.Char8 (ByteString)
+import qualified Data.ByteString.Char8 as BS (
     drop, findIndex, head, isPrefixOf, last, length, span, unpack
   )
-import Data.ByteString.Lazy.Char8.Util (previous, subBS)
-import Data.Int (Int64)
+import Data.ByteString.Char8.Util (previous, subBS)
+import qualified Data.ByteString.Lazy as Lazy (ByteString)
 import qualified Data.Map as Map (lookup)
 import PDF.Body (populate)
 import qualified PDF.EOL as EOL (Style(..), charset, parser)
@@ -21,10 +21,7 @@ import PDF.Object (
   )
 import qualified PDF.Output as Output (render, string)
 import PDF.Output (Output(..), nextLine)
-import Text.Parsec
-import Text.Parsec.ByteString.Lazy (Parser)
-import Text.Parsec.Pos (newPos)
-import Text.ParserCombinators.Parsec.Error (Message(..), newErrorMessage)
+import PDF.Parser (Parser, runParser, string, takeAll)
 import Text.Printf (printf)
 
 data Document = Document {
@@ -38,20 +35,17 @@ instance Output Document where
       Output.string (printf "%%PDF-%s" pdfVersion)
     `nextLine` output contents
 
-render :: Document -> ByteString
+render :: Document -> Lazy.ByteString
 render document@(Document {eolStyle}) = Output.render eolStyle document
 
-version :: Parser String
-version = string magicNumber *> many (noneOf EOL.charset)
+version :: Parser () String
+version = BS.unpack <$>
+  (string magicNumber *> takeAll (not . (`elem` EOL.charset)))
 
-parseError :: String -> Either ParseError a
-parseError errorMessage =
-  Left $ newErrorMessage (Message errorMessage) (newPos "" 0 0)
+check :: Bool -> String -> Either String ()
+check test errorMessage = if test then return () else Left errorMessage
 
-check :: Bool -> String -> Either ParseError ()
-check test errorMessage = if test then return () else parseError errorMessage
-
-readStartXref :: EOL.Style -> ByteString -> Either ParseError Int64
+readStartXref :: EOL.Style -> ByteString -> Either String Int
 readStartXref eolStyle input =
   check (eofMarker `BS.isPrefixOf` (BS.drop eofMarkerPosition input))
     "Badly formed document : missing EOF marker at the end"
@@ -68,7 +62,7 @@ readStartXref eolStyle input =
       previous eolLastByte (eofMarkerPosition - eolOffset - 1) input + 1
     startXrefLength = eofMarkerPosition - eolOffset - startXrefPosition
 
-findNextLine :: ByteString -> Int64
+findNextLine :: ByteString -> Int
 findNextLine input =
   let (line, eolPrefixed) = BS.span notInEol input in
   let nextNotInEol = BS.findIndex notInEol eolPrefixed in
@@ -76,7 +70,7 @@ findNextLine input =
   where
     notInEol = not . (`elem` EOL.charset)
 
-findNextSection :: Int64 -> ByteString -> Int64
+findNextSection :: Int -> ByteString -> Int
 findNextSection offset input =
   case BS.findIndex (== BS.head eofMarker) input of
     Nothing -> 0
@@ -87,9 +81,9 @@ findNextSection offset input =
       then newOffset + findNextLine newInput
       else findNextSection (newOffset + 1) (BS.drop 1 newInput)
 
-readStructures :: Int64 -> ByteString -> Either ParseError [InputStructure]
+readStructures :: Int -> ByteString -> Either String [InputStructure]
 readStructures startXref input =
-  parse structure ('@':show startXref) (BS.drop startXref input) >>= stopOrFollow
+  runParser structure () (BS.drop startXref input) >>= stopOrFollow
   where
     stopOrFollow s@(Structure {trailer}) =
       case Map.lookup (Name "Prev") trailer of
@@ -98,11 +92,11 @@ readStructures startXref input =
           let offset = truncate newStartXref in
           let startOffset = findNextSection offset (BS.drop offset input) in
           (InputStructure startOffset s:) <$> (readStructures offset input)
-        Just v -> parseError $ "Bad value for Prev entry in trailer: " ++ show v
+        Just v -> Left $ "Bad value for Prev entry in trailer: " ++ show v
 
-parseDocument :: ByteString -> Either ParseError Document
+parseDocument :: ByteString -> Either String Document
 parseDocument input = do
-  (pdfVersion, eolStyle) <- parse ((,) <$> version <*> EOL.parser) "" input
+  (pdfVersion, eolStyle) <- runParser ((,) <$> version <*> EOL.parser) () input
   startXref <- readStartXref eolStyle input
   structuresRead <- readStructures startXref input
   let contents = populate input <$> structuresRead
diff --git a/src/PDF/Body.hs b/src/PDF/Body.hs
index 23d4a9c..53fa6f0 100644
--- a/src/PDF/Body.hs
+++ b/src/PDF/Body.hs
@@ -1,21 +1,23 @@
 {-# LANGUAGE NamedFieldPuns #-}
+{-# LANGUAGE OverloadedStrings #-}
 module PDF.Body (
     populate
   ) where
 
-import Data.ByteString.Lazy.Char8 (ByteString)
-import qualified Data.ByteString.Lazy.Char8 as BS (drop, pack)
-import Data.Int (Int64)
+import Control.Applicative ((<|>))
+import Control.Monad.State (get, gets, modify)
+import Data.ByteString.Char8 (ByteString)
+import qualified Data.ByteString.Char8 as BS (drop, unpack)
 import Data.Map ((!))
 import qualified Data.Map as Map (empty, insert, lookup)
 import qualified PDF.EOL as EOL (charset, parser)
 import PDF.Object (
       Content(..), DirectObject(..), Flow(..), IndirectObjCoordinates(..)
     , InputStructure(..), Name(..), Number(..), Object(..), Occurrence(..)
-    , Parser, Structure(..), XRefEntry(..), XRefSection, XRefSubSection(..)
+    , Structure(..), XRefEntry(..), XRefSection, XRefSubSection(..)
     , blank, dictionary, directObject, integer, line
   )
-import Text.Parsec
+import PDF.Parser (Parser, block, char, on, runParser, takeAll)
 
 data UserState = UserState {
       input :: ByteString
@@ -26,7 +28,7 @@ data UserState = UserState {
 type SParser = Parser UserState
 
 modifyFlow :: (Flow -> Flow) -> SParser ()
-modifyFlow f = modifyState $ \state -> state {flow = f $ flow state}
+modifyFlow f = modify $ \state -> state {flow = f $ flow state}
 
 addObject :: Int -> Object -> SParser ()
 addObject objectId newObject = modifyFlow $ \flow -> flow {
@@ -39,9 +41,10 @@ pushOccurrence newOccurrence = modifyFlow $ \flow -> flow {
   }
 
 comment :: Parser u String
-comment = char '%' *> many (noneOf EOL.charset) <* EOL.parser
+comment = BS.unpack <$>
+  (char '%' *> takeAll (not . (`elem` EOL.charset)) <* EOL.parser)
 
-lookupOffset :: Int -> XRefSection -> Maybe Int64
+lookupOffset :: Int -> XRefSection -> Maybe Int
 lookupOffset _ [] = Nothing
 lookupOffset objectId (xrefSubSection:others) =
   let XRefSubSection {firstObjectId, entriesNumber, entries} = xrefSubSection in
@@ -53,25 +56,20 @@ lookupOffset objectId (xrefSubSection:others) =
       _ -> Nothing
   else lookupOffset objectId others
 
-getOffset :: Int -> SParser Int64
+getOffset :: Int -> SParser Int
 getOffset objectId = do
-  table <- xreferences <$> getState
+  table <- gets xreferences
   case lookupOffset objectId table of
     Nothing -> fail $
       "obj " ++ show objectId ++ " is referenced but missing in XRef table"
     Just offset -> return offset
 
-on :: Monad m => ParsecT s u m a -> s -> ParsecT s u m a
-on parser input = do
-  originalInput <- getInput
-  setInput input >> parser <* setInput originalInput
-
 loadNumber :: Int -> SParser Float
 loadNumber objectId = do
   offset <- getOffset objectId
-  objectStart <- BS.drop offset . input <$> getState
-  indirectObjCoordinates `on` objectStart >> return ()
-  objectValue <- (!objectId) . tmpObjects . flow <$> getState
+  objectStart <- BS.drop offset <$> gets input
+  indirectObjCoordinates `on` (objectStart :: ByteString) >> return ()
+  objectValue <- (!objectId) . tmpObjects <$> gets flow
   case objectValue of
     Direct (NumberObject (Number n)) -> return n
     obj -> fail $ "Expected obj@" ++ show offset ++ " (" ++ show obj ++ ") to be a Number"
@@ -83,7 +81,7 @@ getSize :: Maybe DirectObject -> SParser Float
 getSize Nothing = fail "Missing '/Length' key on stream"
 getSize (Just (NumberObject (Number size))) = return size
 getSize (Just (Reference (IndirectObjCoordinates {objectId}))) = do
-  Flow {tmpObjects} <- flow <$> getState
+  Flow {tmpObjects} <- gets flow
   case Map.lookup objectId tmpObjects of
     Nothing -> loadNumber objectId
     Just (Direct (NumberObject (Number size))) -> return size
@@ -92,13 +90,13 @@ getSize (Just (Reference (IndirectObjCoordinates {objectId}))) = do
 getSize (Just v) = fail $ invalidValue (Direct v) ++ " for /Length"
 
 streamObject :: SParser Object
-streamObject = try $ do
+streamObject = do
   header <- dictionary <* blank
   size <- getSize (Map.lookup (Name "Length") header)
-  streamContent <- BS.pack <$> stream (truncate size)
+  streamContent <- stream (truncate size)
   return $ Stream {header, streamContent}
   where
-    stream size = line "stream" *> count size anyChar <* blank <* line "endstream"
+    stream size = line "stream" *> block size <* blank <* line "endstream"
 
 object :: SParser Object
 object = streamObject <|> Direct <$> directObject
@@ -117,7 +115,7 @@ occurrence = Comment <$> comment <|> Indirect <$> indirectObjCoordinates
 populate :: ByteString -> InputStructure -> Content
 populate input structure =
   let bodyInput = BS.drop (startOffset structure) input in
-  case runParser recurseOnOccurrences initialState "" bodyInput of
+  case runParser recurseOnOccurrences initialState bodyInput of
     Left _ -> Content {occurrences = [], objects = Map.empty, docStructure}
     Right finalState ->
       let Flow {occurrencesStack, tmpObjects} = flow finalState in
@@ -135,4 +133,4 @@ populate input structure =
 
 recurseOnOccurrences :: SParser UserState
 recurseOnOccurrences =
-  (occurrence >>= pushOccurrence >> recurseOnOccurrences) <|> getState
+  (occurrence >>= pushOccurrence >> recurseOnOccurrences) <|> get
diff --git a/src/PDF/EOL.hs b/src/PDF/EOL.hs
index 4d88751..0493ea0 100644
--- a/src/PDF/EOL.hs
+++ b/src/PDF/EOL.hs
@@ -1,19 +1,20 @@
+{-# LANGUAGE OverloadedStrings #-}
 module PDF.EOL (
       Style(..)
     , charset
     , parser
   ) where
 
-import Data.ByteString.Lazy.Char8 (ByteString)
-import Text.Parsec ((<|>), Parsec, string, try)
+import Control.Applicative ((<|>))
+import PDF.Parser (Parser, string)
 
 data Style = CR | LF | CRLF deriving Show
 
 charset :: String
 charset = "\r\n"
 
-parser :: Parsec ByteString u Style
+parser :: Parser s Style
 parser =
-    try (string "\r\n" >> return CRLF)
+    (string "\r\n" >> return CRLF)
   <|> (string "\r" >> return CR)
   <|> (string "\n" >> return LF)
diff --git a/src/PDF/Object.hs b/src/PDF/Object.hs
index f67d130..17c37e4 100644
--- a/src/PDF/Object.hs
+++ b/src/PDF/Object.hs
@@ -11,7 +11,6 @@ module PDF.Object (
     , Number(..)
     , Object(..)
     , Occurrence(..)
-    , Parser
     , Structure(..)
     , XRefEntry(..)
     , XRefSection
@@ -26,8 +25,11 @@ module PDF.Object (
     , structure
   ) where
 
-import Data.ByteString.Lazy.Char8 (ByteString)
-import Data.Int (Int64)
+import Control.Applicative ((<|>))
+import Data.ByteString.Char8 (ByteString)
+import qualified Data.ByteString.Char8 as BS (
+    concat, cons, pack, singleton, unpack
+  )
 import Data.Map (Map, (!), mapWithKey)
 import qualified Data.Map as Map (elems, fromList, toList)
 import qualified PDF.EOL as EOL (charset, parser)
@@ -36,15 +38,17 @@ import PDF.Output (
       OBuilder, Offset(..), Output(..)
     , byteString, getOffsets, join, newLine, nextLine, saveOffset
   )
-import Text.Parsec
+import PDF.Parser (
+      Parser, (<?>)
+    , char, choice, count, decNumber, hexNumber, many, octDigit, oneOf, option
+    , sepBy, string, takeAll, takeAll1
+  )
 import Text.Printf (printf)
 
-type Parser u = Parsec ByteString u
-
 line :: String -> Parser u ()
-line l = string l *> EOL.parser *> return ()
+line l = string (BS.pack l) *> EOL.parser *> return () <?> printf "line «%s»" l
 
-magicNumber :: String
+magicNumber :: ByteString
 magicNumber = "%PDF-"
 
 eofMarker :: ByteString
@@ -53,20 +57,17 @@ eofMarker = "%%EOF"
 whiteSpaceCharset :: String
 whiteSpaceCharset = "\0\t\12 "
 
-whiteSpace :: Parser u ()
-whiteSpace = oneOf whiteSpaceCharset *> return () <|> EOL.parser *> return ()
-
 blank :: Parser u ()
-blank = skipMany whiteSpace
+blank = takeAll (`elem` (EOL.charset ++ whiteSpaceCharset)) *> return ()
 
 delimiterCharset :: String
 delimiterCharset = "()<>[]{}/%"
 
-regular :: Parser u Char
-regular = noneOf $ EOL.charset ++ whiteSpaceCharset ++ delimiterCharset
+regular :: Char -> Bool
+regular = not . (`elem` (EOL.charset ++ whiteSpaceCharset ++ delimiterCharset))
 
 integer :: (Read a, Num a) => Parser u a
-integer = read <$> many1 digit <* whiteSpace
+integer = read . BS.unpack <$> decNumber <* blank <?> "decimal integer"
 
 -------------------------------------
 --          OBJECTS
@@ -76,7 +77,8 @@ integer = read <$> many1 digit <* whiteSpace
 -- Boolean
 --
 boolean :: Parser u Bool
-boolean = (string "true" *> return True) <|> (string "false" *> return False)
+boolean =
+  (string "true" *> return True) <|> (string "false" *> return False) <?> "boolean"
 
 --
 -- Number
@@ -90,12 +92,13 @@ instance Output Number where
       _ -> printf "%f" f
 
 number :: Parser u Number
-number = Number . read <$>
-  (mappend <$> sign <*> (integerPart <|> ('0':) <$> floatPart))
+number = Number . read . BS.unpack <$>
+  (mappend <$> sign <*> (integerPart <|> BS.cons '0' <$> floatPart))
+  <?> "number"
   where
     sign = string "-" <|> option "" (char '+' >> return "")
-    integerPart = mappend <$> many1 digit <*> option "" floatPart
-    floatPart = (:) <$> char '.' <*> (option "0" $ many1 digit)
+    integerPart = mappend <$> decNumber <*> option "" floatPart
+    floatPart = BS.cons <$> char '.' <*> (option "0" $ decNumber)
 
 --
 -- StringObject
@@ -108,14 +111,18 @@ instance Output StringObject where
 
 stringObject :: Parser u StringObject
 stringObject =
-    Literal <$> (char '(' *> (concat <$> many literalStringBlock) <* char ')')
-  <|> Hexadecimal <$> (char '<' *> many hexDigit <* char '>')
+    Literal . BS.unpack <$> (char '(' *> (BS.concat <$> literalString) <* char ')')
+  <|> Hexadecimal . BS.unpack <$> (char '<' *> hexNumber <* char '>')
+  <?> "string object (literal or hexadecimal)"
   where
-    literalStringBlock = many1 (noneOf "\\()") <|> matchingParenthesis <|> escapedChar
+    literalString = many literalStringBlock
+    literalStringBlock = takeAll1 normalChar <|> matchingParenthesis <|> escapedChar
+    normalChar = not . (`elem` ("\\()" :: String))
     matchingParenthesis =
-      (++) <$> ((:) <$> char '(' <*> literalStringBlock) <*> string ")"
-    escapedChar = (:) <$> char '\\' <*> ((:[]) <$> oneOf "nrtbf()\\" <|> octalCode)
-    octalCode = choice $ (\n -> count n octDigit) <$> [1..3]
+      mappend <$> (BS.cons <$> char '(' <*> literalStringBlock) <*> string ")"
+    escapedChar =
+      BS.cons <$> char '\\' <*> (BS.singleton <$> oneOf "nrtbf()\\" <|> octalCode)
+    octalCode = choice $ (\n -> BS.pack <$> count n octDigit) <$> [1..3]
 
 --
 -- Name
@@ -126,13 +133,14 @@ instance Output Name where
   output (Name n) = "/" `mappend` Output.string n
 
 name :: Parser u Name
-name = Name <$> (char '/' *> many regular)
+name = Name . BS.unpack <$> (char '/' *> takeAll regular) <?> "name"
 
 --
 -- Array
 --
 array :: Parser u [DirectObject]
-array = char '[' *> blank *> directObject `endBy` blank <* char ']'
+array =
+  char '[' *> blank *> directObject `sepBy` blank <* blank <* char ']' <?> "array"
 
 --
 -- Dictionary
@@ -149,16 +157,16 @@ instance Output Dictionary where
 
 dictionary :: Parser u Dictionary
 dictionary =
-  try (string "<<" *> blank *> keyValPairs <* string ">>")
+  string "<<" *> blank *> keyValPairs <* string ">>" <?> "dictionary"
   where
     keyVal = (,) <$> name <* blank <*> directObject
-    keyValPairs = Map.fromList <$> keyVal `endBy` blank
+    keyValPairs = Map.fromList <$> keyVal `sepBy` blank <* blank
 
 --
 -- Null
 --
 nullObject :: Parser u ()
-nullObject = string "null" *> return ()
+nullObject = string "null" *> return () <?> "null object"
 
 --
 -- Reference
@@ -169,7 +177,7 @@ data IndirectObjCoordinates = IndirectObjCoordinates {
   } deriving Show
 
 reference :: Parser u IndirectObjCoordinates
-reference = IndirectObjCoordinates <$> integer <*> integer <* char 'R'
+reference = IndirectObjCoordinates <$> integer <*> integer <* char 'R' <?> "reference to an object"
 
 --
 -- DirectObject
@@ -198,14 +206,15 @@ instance Output DirectObject where
 
 directObject :: Parser u DirectObject
 directObject =
-    Boolean <$> try boolean
-  <|> Reference <$> try reference {- defined before Number because Number is a prefix of it -}
-  <|> NumberObject <$> try number
-  <|> StringObject <$> try stringObject
-  <|> NameObject <$> try name
-  <|> Array <$> try array
-  <|> Dictionary <$> try dictionary
-  <|> const Null <$> try nullObject
+    Boolean <$> boolean
+  <|> Reference <$> reference {- defined before Number because Number is a prefix of it -}
+  <|> NumberObject <$> number
+  <|> StringObject <$> stringObject
+  <|> NameObject <$> name
+  <|> Array <$> array
+  <|> Dictionary <$> dictionary
+  <|> const Null <$> nullObject
+  <?> "direct object"
 
 --
 -- Object
@@ -226,7 +235,6 @@ instance Output Object where
     `nextLine` byteString streamContent
     `mappend` "endstream"
 
-
 --
 -- Occurrence
 --
@@ -249,10 +257,10 @@ outputOccurrence objects (Indirect (IndirectObjCoordinates {objectId, versionNum
 -- XRefEntry
 --
 data XRefEntry = InUse {
-      offset :: Int64
+      offset :: Int
     , generation :: Int
   } | Free {
-      nextFree :: Int64
+      nextFree :: Int
     , generation :: Int
   } deriving Show
 
@@ -265,14 +273,13 @@ instance Output XRefEntry where
 entry :: Parser u XRefEntry
 entry = do
   (big, small) <- (,) <$> integer <*> integer
-  (inUse big small <|> free big small) <* blank
+  (inUse big small <|> free big small <?> "XRef entry") <* blank
   where
-    inUse :: Int64 -> Int -> Parser u XRefEntry
+    inUse :: Int -> Int -> Parser u XRefEntry
     inUse offset generation = char 'n' *> return (InUse {offset, generation})
-    free :: Int64 -> Int -> Parser u XRefEntry
+    free :: Int -> Int -> Parser u XRefEntry
     free nextFree generation = char 'f' *> return (Free {nextFree, generation})
 
-
 --
 -- XRefSubSection
 --
@@ -289,7 +296,7 @@ instance Output XRefSubSection where
 
 xrefSubSection :: Parser u XRefSubSection
 xrefSubSection = do
-  (firstObjectId, entriesNumber) <- (,) <$> integer <*> integer
+  (firstObjectId, entriesNumber) <- (,) <$> integer <*> integer <?> "XRef subsection"
   entries <- Map.fromList . zip [firstObjectId..] <$> count entriesNumber entry
   return $ XRefSubSection {firstObjectId, entriesNumber, entries}
 
@@ -299,7 +306,7 @@ type XRefSection = [XRefSubSection]
 -- Structure
 --
 data InputStructure = InputStructure {
-      startOffset :: Int64
+      startOffset :: Int
     , inputStructure :: Structure
   }
 
@@ -314,7 +321,7 @@ structure =
   <$> (line "xref" *> xrefSubSection `sepBy` EOL.parser)
   <*> (line "trailer" *> dictionary <* EOL.parser)
 
-updateXrefs :: XRefSection -> Map Offset Int64 -> (XRefSection, Int64)
+updateXrefs :: XRefSection -> Map Offset Int -> (XRefSection, Int)
 updateXrefs xrefSection offsets = (
       updateSubSection <$> xrefSection
     , offsets ! StartXRef
diff --git a/src/PDF/Output.hs b/src/PDF/Output.hs
index bcbb01a..27a77f1 100644
--- a/src/PDF/Output.hs
+++ b/src/PDF/Output.hs
@@ -11,7 +11,6 @@ module PDF.Output (
     , char
     , getOffsets
     , join
-    , lift
     , newLine
     , nextLine
     , saveOffset
@@ -19,10 +18,11 @@ module PDF.Output (
     , render
   ) where
 
-import Data.ByteString.Builder (Builder, char8, lazyByteString, string8, toLazyByteString)
-import Data.ByteString.Lazy.Char8 (ByteString)
-import qualified Data.ByteString.Lazy.Char8 as BS (length)
-import Data.Int (Int64)
+import Data.ByteString.Builder (Builder, char8, string8, toLazyByteString)
+import qualified Data.ByteString.Builder as B (byteString)
+import Data.ByteString (ByteString)
+import qualified Data.ByteString as BS (length)
+import qualified Data.ByteString.Lazy as Lazy (ByteString)
 import Data.Map (Map)
 import qualified Data.Map as Map (singleton)
 import Data.String (IsString(..))
@@ -31,7 +31,7 @@ import qualified PDF.EOL as EOL (Style(..))
 
 data Offset = StartXRef | ObjectId Int deriving (Eq, Ord)
 
-newtype OContext a = OContext (RWS EOL.Style (Map Offset Int64) Int64 a)
+newtype OContext a = OContext (RWS EOL.Style (Map Offset Int) Int a)
 type OBuilder = OContext Builder
 
 instance Functor OContext where
@@ -51,7 +51,7 @@ saveOffset offset = OContext $
 lift :: (a -> Builder) -> a -> OBuilder
 lift f a = return (f a)
 
-getOffsets :: OBuilder -> OContext (OBuilder, Map Offset Int64)
+getOffsets :: OBuilder -> OContext (OBuilder, Map Offset Int)
 getOffsets (OContext builder) =
   OContext (listen builder >>= \(a, w) -> return (return a, w))
 
@@ -108,9 +108,9 @@ string :: String -> OBuilder
 string s = lift string8 s <* OContext (modify (+ toEnum (length s)))
 
 byteString :: ByteString -> OBuilder
-byteString bs = lift lazyByteString bs <* OContext (modify (+ BS.length bs))
+byteString bs = lift B.byteString bs <* OContext (modify (+ BS.length bs))
 
-render :: Output a => EOL.Style -> a -> ByteString
+render :: Output a => EOL.Style -> a -> Lazy.ByteString
 render eolStyle a =
   let OContext builder = output a in
   let (outputByteString, _, _) = runRWS builder eolStyle 0 in
diff --git a/src/PDF/Parser.hs b/src/PDF/Parser.hs
new file mode 100644
index 0000000..3f170e7
--- /dev/null
+++ b/src/PDF/Parser.hs
@@ -0,0 +1,98 @@
+module PDF.Parser (
+      Parser
+    , (<?>)
+    , block
+    , char
+    , choice
+    , count
+    , decNumber
+    , hexNumber
+    , many
+    , octDigit
+    , on
+    , oneOf
+    , option
+    , runParser
+    , sepBy
+    , string
+    , takeAll
+    , takeAll1
+  ) where
+
+import Control.Applicative ((<|>), empty)
+import Control.Monad.State (StateT(..), evalStateT)
+import Control.Monad.Trans (lift)
+import qualified Data.Attoparsec.ByteString.Char8 as Atto (
+    Parser, char, parseOnly, satisfy, string, take, takeWhile, takeWhile1
+  )
+import Data.ByteString (ByteString)
+import Data.Char (toLower)
+import Data.Set (Set)
+import qualified Data.Set as Set (fromList, member, unions)
+
+type Parser s = StateT s Atto.Parser
+
+(<?>) :: Parser s a -> String -> Parser s a
+(<?>) parser debugMessage = parser <|> fail debugMessage
+
+block :: Int -> Parser s ByteString
+block = lift . Atto.take
+
+char :: Char -> Parser s Char
+char = lift . Atto.char
+
+choice :: [Parser s a] -> Parser s a
+choice = foldr (<|>) empty
+
+count :: Int -> Parser s a -> Parser s [a]
+count 0 _ = return []
+count n p = (:) <$> p <*> count (n-1) p
+
+decNumber :: Parser s ByteString
+decNumber = lift $ Atto.takeWhile1 (`Set.member` digits)
+
+digits :: Set Char
+digits = Set.fromList ['0'..'9']
+
+hexDigits :: Set Char
+hexDigits = Set.unions [digits, Set.fromList af, Set.fromList $ toLower <$> af]
+  where
+    af = ['A'..'F']
+
+hexNumber :: Parser s ByteString
+hexNumber = lift $ Atto.takeWhile1 (`Set.member` hexDigits)
+
+many :: Parser s a -> Parser s [a]
+many parser = (:) <$> parser <*> many parser <|> return []
+
+octDigit :: Parser s Char
+octDigit = oneOf ['0'..'7']
+
+on :: Parser s a -> ByteString -> Parser s (Either String a)
+on (StateT parserF) input = StateT $ \state ->
+  case Atto.parseOnly (parserF state) input of
+    Left errorMsg -> return (Left errorMsg, state)
+    Right (result, newState) -> return (Right result, newState)
+
+oneOf :: String -> Parser s Char
+oneOf charSet = lift $ Atto.satisfy (`elem` charSet)
+
+option :: a -> Parser s a -> Parser s a
+option defaultValue p = p <|> pure defaultValue
+
+runParser :: Parser s a -> s -> ByteString -> Either String a
+runParser parser initState =
+  Atto.parseOnly (evalStateT parser initState)
+
+sepBy :: Parser s a -> Parser s b -> Parser s [a]
+sepBy parser separator =
+  option [] $ (:) <$> parser <*> many (separator *> parser)
+
+string :: ByteString -> Parser s ByteString
+string = lift . Atto.string
+
+takeAll :: (Char -> Bool) -> Parser s ByteString
+takeAll = lift . Atto.takeWhile
+
+takeAll1 :: (Char -> Bool) -> Parser s ByteString
+takeAll1 = lift . Atto.takeWhile1