1

In Haskell, there are a few different options to "parsing text". I know of Alex & Happy, Parsec and Attoparsec. Probably some others.

I'd like to put together a library where the user can input pieces of a URL (scheme e.g. HTTP, hostname, username, port, path, query, etc.) I'd like to validate the pieces according to the ABNF specified in RFC 3986.

In other words, I'd like to put together a set of functions such as:

validateScheme    :: String -> Bool
validateUsername  :: String -> Bool
validatePassword  :: String -> Bool
validateAuthority :: String -> Bool
validatePath      :: String -> Bool
validateQuery     :: String -> Bool

What is the most appropriate tool to use to write these functions?

Alex's regexps is very concise, but it's a tokenizer and doesn't straightforwardly allow you to parse using specific rules, so it's not quite what I'm looking for, but perhaps it can be wrangled into doing this easily.

I've written Parsec code that does some of the above, but it looks very different from the original ABNF and unnecessarily long.

So, there must be an easier and/or more appropriate way. Recommendations?

  • If you want to take a BNF grammar and use that as your code to parse, you want a parser generator that takes the grammar and spits out code. I don't know off hand which parser generators might work directly off a BNF like this, but that's because I don't prefer this type of approach. The resulting code is always greater and maintenance is funky since it's nothing you can directly dig into easily. The answer below is a nice clear concise approach which takes a minimum of code and would be easy to maintain. Though I agree it's not the BNF->validation you're looking for. – Jimmy Hoffa Nov 7 '13 at 15:38
3

I'm curious why Parsec was so long, I'd write it something like

data URL = URL { scheme, hostname, username, port, path, query :: String}

parser :: String -> Either ParseError URL
parser = flip parse "Foo" $ Parser <$> parseScheme <*> parseHostname <*> parseUsername ...

validators = [validateScheme . scheme, validateHostname . hostname, .....]
validate :: URL -> Bool
validate url = all ($url) validators

This is nice and concise looking to me. As for how to implement the parser, it's a straightforward to parse each section and then apply URL to each piece.

  • To me it's not whether it's straightforward. To me, it looks much more removed from the original specification. – Ana Nov 6 '13 at 19:54
  • 1
    @Ana Define removed? I think it'd be a misfeature to mix your parsing and validation. Parse a wellformed url, and check if it fits your specifications. – jozefg Nov 6 '13 at 19:55
  • You're user shouldn't have to care/see exactly what goes on to parse a url, they just care about making sure it fits XYZ specifications. Dealing with it as a datastructure makes this much more feasible – jozefg Nov 6 '13 at 19:56
  • I would make the validators Either SomeError URL -> Either SomeError URL and compose them into the applicative parser so you have: Parser <$> (validateScheme . parseScheme) <*> (validateHostname . parseHostname) <*> (validateUsername . parseUsername) getting validation at parse/construction time, the intermediary step while good design for a larger piece seems unnecessary given what the OP is asking for. Though if this is a part of a greater whole your approach is better because the separation of concerns making for more reusable components in your approach. – Jimmy Hoffa Nov 7 '13 at 16:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.