Hi Dan, thank you for the solution. It looks pretty interesting and usable, however I'll have to spend some time understanding arrows: I never had an opportunity to use them before. Anyway, it looks very close to what I actually need, and in any case much less ugly than breaking the GenParser encapsulation 2009/8/6 Dan Weston

Of course, since ParsecT s u m is a functor, feel free to use fmap instead of parsecMap. Then you don't need to import from Text.Parsec.Prim. And in hindsight, I might prefer the name (<:>) or cons to (<>) for the first function, but now I'm just obsessing. :)

Dan

Dan Weston wrote:

...

I think parsecMap does the job here:

----------------------- import Text.ParserCombinators.Parsec hiding ((<|>)) import Text.Parsec.Prim(parsecMap) import Control.Applicative((<|>)) import Control.Arrow((|||),(&&&))

-- Tagged (:) (<>) :: Either Char Char -> Either String String -> Either String String Left a <> Left b = Left (a:b) Left a <> Right b = Left (a:b) Right a <> Left b = Left (a:b) Right a <> Right b = Right (a:b)

-- Tagged concat stringParser :: [Either Char Char] -> Either String String stringParser = foldr (<>) (Right "")

-- Parse Integer if properly tagged, keeping unparsed string maybeToInteger :: Either String String -> (Maybe Integer, String) maybeToInteger = (const Nothing ||| Just . read) &&& (id ||| id)

-- Tagged-choice parser intOrStringParser = parsecMap (maybeToInteger . stringParser) $ many1 (parsecMap Right digit <|> parsecMap Left (noneOf ";)"))

-- Parse between parentheses intOrStringListParser = between (char '(') (char ')') (sepBy1 intOrStringParser (char ';')) -----------------------

Then you get a tagged version of each string, along with the string itself:

*P> parseTest intOrStringListParser $ "(1;2w4;8;85)" [(Just 1,"1"),(Nothing,"2w4"),(Just 8,"8"),(Just 85,"85")]

There may be some parsecMap-fold fusion optimization possible, though I haven't looked into that.

Dan

Paul Sujkov wrote:

...

Hi everybody,

suppose I have two different parsers: one just reads the string, and another one parses some values from it. E.g.:

parseIntList :: Parser [Integer] parseIntList = do char '(' res <- liftM (map read) (sepBy1 (many1 digit) (char ';')) char ')' return res

parseIntString :: Parser String parseIntString = manyTill anyChar eof

so for some input like this - "(1;2;3;4)" - I will have two different result:

*Parlog> parseTest parseIntList "(1;2;3;4)" [1,2,3,4] *Parlog> parseTest parseIntString "(1;2;3;4)" "(1;2;3;4)"

but the thing that I actually want is something like Parser ([Integer], String) - results from both parsers at a time, no matter whether one of them fails or not:

*Parlog> parseTest parseIntListAndString "(1;2;3;4)" ([1,2,3,4], "(1;2;3;4)")

it is impossible at first sight, because first parser to use will consume all the input, and there will be nothing to parse for the second one

Parsec contains "choice" function, but it is implemented via <|> and that is mplus - so it tries second alternative only if the first one fails. Is it possible to use two parsers for the same string (with try-like backtracking, no input actually consumed till the second parser finishes)? I can assume only dirty hacks with the GenParser internals - manual position storing and backtracking - but that is obviously not good

however, my first attempt to solve the problem was kind a like that: to parse string to String, and then to use it as an input for the next level parse call:

parseIntListAndString :: Parser ([Integer], String) parseIntListAndString = do str <- parseIntString return (res str, str) where res str = case (parse parseIntList "" str) of Left err -> [] Right val -> val

but the problems with such a method began when I switched from Parser to GenParser with user state: function parseIntList have to update the state, but it can't have the same state as the parseIntListAndString any more: it has it's own. I can explicitly pass the state from parseIntListAndString to parseIntList, but I see no suitable way for the parseIntList to update it. I can return the updated state value from the parseIntList function, and call setState on a result - but it seems rather ugly to mee. However, if nothing else will do, that is an alternative

it is of course possible to use two different parsers sequentially, but it is also very ineffective: I need to use such multiple parsing on a relatively small substring of the actual input, so little backtracking would be a much nicier approach. Any suggestions?

-- Regards, Paul Sujkov

-- Regards, Paul Sujkov

The uu-parsinglib: http://hackage.haskell.org/packages/archive/uu-parsinglib/2.2.0/doc/html/Tex... contains a combinator to achieve just this: -- parsing two alternatives and returning both rsults pAscii = pSym ('\000', '\254') pIntList = pParens ((pSym ';') `pListSep` (read <$> pList (pSym ('0', '9')))) parseIntString = pList (pAscii) parseBoth = pPair pIntList parseIntString pPair p q = amb (Left <$> p <|> Right <$> q) The amb combinator tells you that it's parser parameter is ambiguous, and returns you all the possible results. Amazingly it still maintains its online behaviour. The only problem is that if either one of the parsers fails then you will get only a single result. I have added the code above to the Examples.hs contained in the uu- parsinglib (so it will show up in due time when I release a new version) which I am attaching. Just load this file, and call the function main to see what are the results of the different parsers and correction strategies. The only problem is that if either one of the parsers fails you will only get one of the results. If both fail you will get the result which fails latest and if both fail at the same place, the one which fails with the least repair costs. If you really want both results, even if the input is erroneaous, things become more complicated, especially if you want to embed this parser in a larger one, since then we have to check whether both parse the same prefix. If needed I could put some work into this, by making a slightly different version of the amb combinator. Doaitse On 6 aug 2009, at 21:03, Dan Weston wrote:

Paul,

Arrows (and category theory in general) are interesting, but you certainly don't need to understand them for this. The only arrow in this code is the lowly function arrow (->). (&&&) and (|||) are duals of each other and mean, respectively, "both" and "either" (though for some bizarre reason, "both" is usually called "fanout"!)

This style of pointfree (or "pointless") code is clearer to me because I don't have a bunch of variable names to invent and have lying around.

Anyway, if you prefer, don't import Control.Arrow at all, and just use:

-- |Both: Apply two functions to same argument and tuple the results infixr 3 &&& (&&&) :: (a -> b) -> (a -> c) -> a -> (b,c) (f &&& g) x = (f x, g x)

-- |Either: If argument is Left, apply Left function, else apply Right function infixr 2 ||| (|||) :: (a -> c) -> (b -> c) -> Either a b -> c (|||) = either

either is implicitly imported from the Prelude and is defined as:

-- | Case analysis for the 'Either' type. -- If the value is @'Left' a@, apply the first function to @a@; -- if it is @'Right' b@, apply the second function to @b@. either :: (a -> c) -> (b -> c) -> Either a b -> c either f _ (Left x) = f x either _ g (Right y) = g y

Dan

Paul Sujkov wrote:

...

...

...

Of course, since ParsecT s u m is a functor, feel free to use fmap instead of parsecMap. Then you don't need to import from Text.Parsec.Prim. And in hindsight, I might prefer the name (<:>) or cons to (<>) for

Hi Dan, thank you for the solution. It looks pretty interesting and usable, however I'll have to spend some time understanding arrows: I never had an opportunity to use them before. Anyway, it looks very close to what I actually need, and in any case much less ugly than breaking the GenParser encapsulation 2009/8/6 Dan Weston )) import Text.Parsec.Prim(parsecMap) import Control.Applicative((<|>)) import Control.Arrow((|||),(&&&)) -- Tagged (:) (<>) :: Either Char Char -> Either String String -> Either String String Left a <> Left b = Left (a:b) Left a <> Right b = Left (a:b) Right a <> Left b = Left (a:b) Right a <> Right b = Right (a:b) -- Tagged concat stringParser :: [Either Char Char] -> Either String String stringParser = foldr (<>) (Right "") -- Parse Integer if properly tagged, keeping unparsed string maybeToInteger :: Either String String -> (Maybe Integer, String) maybeToInteger = (const Nothing ||| Just . read) &&& (id ||| id) -- Tagged-choice parser intOrStringParser = parsecMap (maybeToInteger . stringParser) $ many1 (parsecMap Right digit <|> parsecMap Left (noneOf ";)")) -- Parse between parentheses intOrStringListParser = between (char '(') (char ')') (sepBy1 intOrStringParser (char ';')) ----------------------- Then you get a tagged version of each string, along with the string itself: *P> parseTest intOrStringListParser $ "(1;2w4;8;85)" [(Just 1,"1"),(Nothing,"2w4"),(Just 8,"8"),(Just 85,"85")] There may be some parsecMap-fold fusion optimization possible, though I haven't looked into that. Dan Paul Sujkov wrote: Hi everybody, suppose I have two different parsers: one just reads the string, and another one parses some values from it. E.g.: parseIntList :: Parser [Integer] parseIntList = do char '(' res <- liftM (map read) (sepBy1 (many1 digit) (char ';')) char ')' return res parseIntString :: Parser String parseIntString = manyTill anyChar eof so for some input like this - "(1;2;3;4)" - I will have two different result: *Parlog> parseTest parseIntList "(1;2;3;4)" [1,2,3,4] *Parlog> parseTest parseIntString "(1;2;3;4)" "(1;2;3;4)" but the thing that I actually want is something like Parser ([Integer], String) - results from both parsers at a time, no matter whether one of them fails or not: *Parlog> parseTest parseIntListAndString "(1;2;3;4)" ([1,2,3,4], "(1;2;3;4)") it is impossible at first sight, because first parser to use will consume all the input, and there will be nothing to parse for the second one Parsec contains "choice" function, but it is implemented via <|> and that is mplus - so it tries second alternative only if the first one fails. Is it possible to use two parsers for the same string (with try-like backtracking, no input actually consumed till the second parser finishes)? I can assume only dirty hacks with the GenParser internals - manual position storing and backtracking - but that is obviously not good however, my first attempt to solve the problem was kind a like that: to parse string to String, and then to use it as an input for the next level parse call: parseIntListAndString :: Parser ([Integer], String) parseIntListAndString = do str <- parseIntString return (res str, str) where res str = case (parse parseIntList "" str) of Left err -> [] Right val -> val but the problems with such a method began when I switched from Parser to GenParser with user state: function parseIntList have to update the state, but it can't have the same state as the parseIntListAndString any more: it has it's own. I can explicitly pass the state from parseIntListAndString to parseIntList, but I see no suitable way for the parseIntList to update it. I can return the updated state value from the parseIntList function, and call setState on a result - but it seems rather ugly to mee. However, if nothing else will do, that is an alternative it is of course possible to use two different parsers sequentially, but it is also very ineffective: I need to use such multiple parsing on a relatively small substring of the actual input, so little backtracking would be a much nicier approach. Any suggestions? -- Regards, Paul Sujkov -- Regards, Paul Sujkov

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Since the uu-parsinglib also provides a monadic interface it should not be too difficult to provide a Parsec interface on top of the uu- parsinglib combinators. so you can re-use large parts of your code. I expect that your parsers eventually will become simpler, since you do not have to add explicit control to the parsing process with try-like constructs. This being said I still think that the applicative interface is to be preferred over the monadic interface, since it does not prohibit all kind of static analases of your parser (as is done in the older parsing library which is part of the uulib package); using the monadic interface for building new parsers based on results recognised thus far is fine, but using it just to construct a parsing result is overkill. If you have any questions please let me know. Doaitse On 10 aug 2009, at 00:30, Paul Sujkov wrote:

Hi Doaitse,

that is very interesting, and I'll take a precise look at the uu- parsinglib. Regarding my original question there exist (I believe) one serious problem: existing code is written exclusively using Parsec and it's already quite complex. At first glimpse I don't see an obvious way to use both libraries in one parsing module simulatiously. However, these are a very good news indeed, thank you

2009/8/9 S. Doaitse Swierstra The uu-parsinglib:

http://hackage.haskell.org/packages/archive/uu-parsinglib/2.2.0/doc/html/Tex...

contains a combinator to achieve just this:

-- parsing two alternatives and returning both rsults pAscii = pSym ('\000', '\254') pIntList = pParens ((pSym ';') `pListSep` (read <$> pList (pSym ('0', '9')))) parseIntString = pList (pAscii)

parseBoth = pPair pIntList parseIntString

pPair p q = amb (Left <$> p <|> Right <$> q)

The amb combinator tells you that it's parser parameter is ambiguous, and returns you all the possible results. Amazingly it still maintains its online behaviour. The only problem is that if either one of the parsers fails then you will get only a single result.

I have added the code above to the Examples.hs contained in the uu- parsinglib (so it will show up in due time when I release a new version) which I am attaching. Just load this file, and call the function main to see what are the results of the different parsers and correction strategies. The only problem is that if either one of the parsers fails you will only get one of the results. If both fail you will get the result which fails latest and if both fail at the same place, the one which fails with the least repair costs.

If you really want both results, even if the input is erroneaous, things become more complicated, especially if you want to embed this parser in a larger one, since then we have to check whether both parse the same prefix. If needed I could put some work into this, by making a slightly different version of the amb combinator.

Doaitse

On 6 aug 2009, at 21:03, Dan Weston wrote:

Paul,

Arrows (and category theory in general) are interesting, but you certainly don't need to understand them for this. The only arrow in this code is the lowly function arrow (->). (&&&) and (|||) are duals of each other and mean, respectively, "both" and "either" (though for some bizarre reason, "both" is usually called "fanout"!)

This style of pointfree (or "pointless") code is clearer to me because I don't have a bunch of variable names to invent and have lying around.

Anyway, if you prefer, don't import Control.Arrow at all, and just use:

-- |Both: Apply two functions to same argument and tuple the results infixr 3 &&& (&&&) :: (a -> b) -> (a -> c) -> a -> (b,c) (f &&& g) x = (f x, g x)

-- |Either: If argument is Left, apply Left function, else apply Right function infixr 2 ||| (|||) :: (a -> c) -> (b -> c) -> Either a b -> c (|||) = either

either is implicitly imported from the Prelude and is defined as:

-- | Case analysis for the 'Either' type. -- If the value is @'Left' a@, apply the first function to @a@; -- if it is @'Right' b@, apply the second function to @b@. either :: (a -> c) -> (b -> c) -> Either a b -> c either f _ (Left x) = f x either _ g (Right y) = g y

Dan

...

...

Of course, since ParsecT s u m is a functor, feel free to use fmap instead of parsecMap. Then you don't need to import from Text.Parsec.Prim. And in hindsight, I might prefer the name (<:>) or cons to (<>) for

Paul Sujkov wrote: Hi Dan, thank you for the solution. It looks pretty interesting and usable, however I'll have to spend some time understanding arrows: I never had an opportunity to use them before. Anyway, it looks very close to what I actually need, and in any case much less ugly than breaking the GenParser encapsulation 2009/8/6 Dan Weston )) import Text.Parsec.Prim(parsecMap) import Control.Applicative((<|>)) import Control.Arrow((|||),(&&&)) -- Tagged (:) (<>) :: Either Char Char -> Either String String -> Either String String Left a <> Left b = Left (a:b) Left a <> Right b = Left (a:b) Right a <> Left b = Left (a:b) Right a <> Right b = Right (a:b) -- Tagged concat stringParser :: [Either Char Char] -> Either String String stringParser = foldr (<>) (Right "") -- Parse Integer if properly tagged, keeping unparsed string maybeToInteger :: Either String String -> (Maybe Integer, String) maybeToInteger = (const Nothing ||| Just . read) &&& (id ||| id) -- Tagged-choice parser intOrStringParser = parsecMap (maybeToInteger . stringParser) $ many1 (parsecMap Right digit <|> parsecMap Left (noneOf ";)")) -- Parse between parentheses intOrStringListParser = between (char '(') (char ')') (sepBy1 intOrStringParser (char ';')) ----------------------- Then you get a tagged version of each string, along with the string itself: *P> parseTest intOrStringListParser $ "(1;2w4;8;85)" [(Just 1,"1"),(Nothing,"2w4"),(Just 8,"8"),(Just 85,"85")] There may be some parsecMap-fold fusion optimization possible, though I haven't looked into that. Dan Paul Sujkov wrote: Hi everybody, suppose I have two different parsers: one just reads the string, and another one parses some values from it. E.g.: parseIntList :: Parser [Integer] parseIntList = do char '(' res <- liftM (map read) (sepBy1 (many1 digit) (char ';')) char ')' return res parseIntString :: Parser String parseIntString = manyTill anyChar eof so for some input like this - "(1;2;3;4)" - I will have two different result: *Parlog> parseTest parseIntList "(1;2;3;4)" [1,2,3,4] *Parlog> parseTest parseIntString "(1;2;3;4)" "(1;2;3;4)" but the thing that I actually want is something like Parser ([Integer], String) - results from both parsers at a time, no matter whether one of them fails or not: *Parlog> parseTest parseIntListAndString "(1;2;3;4)" ([1,2,3,4], "(1;2;3;4)") it is impossible at first sight, because first parser to use will consume all the input, and there will be nothing to parse for the second one Parsec contains "choice" function, but it is implemented via <|> and that is mplus - so it tries second alternative only if the first one fails. Is it possible to use two parsers for the same string (with try-like backtracking, no input actually consumed till the second parser finishes)? I can assume only dirty hacks with the GenParser internals - manual position storing and backtracking - but that is obviously not good however, my first attempt to solve the problem was kind a like that: to parse string to String, and then to use it as an input for the next level parse call: parseIntListAndString :: Parser ([Integer], String) parseIntListAndString = do str <- parseIntString return (res str, str) where res str = case (parse parseIntList "" str) of Left err -> [] Right val -> val but the problems with such a method began when I switched from Parser to GenParser with user state: function parseIntList have to update the state, but it can't have the same state as the parseIntListAndString any more: it has it's own. I can explicitly pass the state from parseIntListAndString to parseIntList, but I see no suitable way for the parseIntList to update it. I can return the updated state value from the parseIntList function, and call setState on a result - but it seems rather ugly to mee. However, if nothing else will do, that is an alternative it is of course possible to use two different parsers sequentially, but it is also very ineffective: I need to use such multiple parsing on a relatively small substring of the actual input, so little backtracking would be a much nicier approach. Any suggestions? -- Regards, Paul Sujkov -- Regards, Paul Sujkov

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

-- Regards, Paul Sujkov