Monad composition can be used for lookbehind parsers, we can make the future parses depend on the previously parsed values.

If we have to parse "a9" or "9a" but not "99" or "aa" we can use the following parser:

backtracking :: MonadCatch m => PR.Parser m Char String
backtracking =
    sequence [PR.satisfy isDigit, PR.satisfy isAlpha]
    <|>
    sequence [PR.satisfy isAlpha, PR.satisfy isDigit]

We know that if the first parse resulted in a digit at the first place then the second parse is going to fail. However, we waste that information and parse the first character again in the second parse only to know that it is not an alphabetic char. By using lookbehind in a Monad composition we can avoid redundant work:

data DigitOrAlpha = Digit Char | Alpha Char

lookbehind :: MonadCatch m => PR.Parser m Char String
lookbehind = do
    x1 <-    Digit <$> PR.satisfy isDigit
         <|> Alpha <$> PR.satisfy isAlpha

    -- Note: the parse depends on what we parsed already
    x2 <- case x1 of
        Digit _ -> PR.satisfy isAlpha
        Alpha _ -> PR.satisfy isDigit

    return $ case x1 of
        Digit x -> [x,x2]
        Alpha x -> [x,x2]

See also concatMap. This monad instance does not fuse, use concatMap when you need fusion.

Maps a function over the output of the parser.

Applicative form of serialWith. Note that this operation does not fuse, use serialWith when fusion is important.

Alternative form of alt. Backtrack and run the second parser if the first one fails.

The "some" and "many" operations of alternative accumulate results in a pure list which is not scalable and streaming. Instead use some and many for fusible operations with composable accumulation of results.

See also alt. This Alternative instance does not fuse, use alt when you need fusion.

mzero is same as empty, it aborts the parser. mplus is same as <|>, it selects the first succeeding parser.

Pre-release