tokenizer-monad

Motivation: Before working with tokenizer-monad, I often implemented tokenizers by recursively destroying Char lists. The resulting code was purely functional, but hardly readable - even more so, if one destroys Text instead of Char lists. In my mind, I usually imagine tokenization algorithms like flow charts, hence I wanted to code them in a similar manner.

Main idea: You walk through the input string like a turtle, and everytime you find a token boundary, you call emit. If some specific kinds of tokens should be suppressed, you can 'discard' them instead (or filter afterwards).

This package supports Strings, strict and lazy Text, as well as strict and lazy ASCII ByteStrings. The generic Tokenizer type from module Control.Monad.Tokenizer takes the text type as its first argument. In the other modules, Tokenizer is already specialized to a specific text type. It is recommendable to avoid importing more than one module from this package. Instead, you could just switch to a more general one.

Provided functions

The functions provided by this package can be divided into three categories:

• tests peek characters from the input text or check a condition; they have no effect on the turtle position nor on the emissions. Examples: peek, isEOT, lookAhead
• walkers modify the turtle position, but have no effect on the emissions. Examples: walk, walkBack, pop, restore
• commits cut off the input text at the current position, and emit or discard the visited part. Examples: emit, discard

Examples

This tokenizer is equivalent to words from Prelude:

 words' :: String -> [String]
 words' = runTokenizerCS $ untilEOT $ do
   c <- pop
   if c `elem` " \t\n\r"
     then discard
     else do
       walkWhile (not . isSpace)
       emit

...> words' "Dieses Haus ist blau."
["Dieses","Haus","ist","blau."]

This tokenizer is similar to lines from Prelude, but discards empty lines:

 lines' :: String -> [String]
 lines' = runTokenizerCS $ untilEOT $ do
   c <- pop
   if c `elem` "\n\r"
     then discard
     else do
       walkWhile (\c -> not (c `elem` "\r\n"))
       emit

...> lines' "Dieses Haus ist\n\nblau.\n"
["Dieses Haus ist","blau."]

A more advanced tokenizer, that can handle punctuation and HTTP URIs in text:

t1Tokenize' :: Tokenizer Text ()
t1Tokenize' = do
  http <- lookAhead "http://"
  https <- lookAhead "https://"
  if (http || https)
     then (walkWhile (not . isSpace) >> discard)
     else do
       c <- peek
       walk
       if isStopSym c
         then emit
         else if c `elem` (" \t\r\n" :: [Char])
              then discard
              else do
                walkWhile (\c -> (c=='_') || not (isSpace c || isPunctuation c))
                emit