C9 Lectures: Erik Meijer - Functional Programming Fundamentals

C9 Lectures: Dr. Erik Meijer - Functional Programming Fundamentals Chapter 8 of 13

Download this episode

Download Video

Description

In Chapter 8, Functional Parsers, it's all about parsing and parsers. A parser is a program that analyses a piece of text to determine its syntactic structure. In a functional language such as Haskell, parsers can naturally
be viewed as functions.

  type Parser = String -> Tree

A parser is a function that takes a string and returns some form of tree.

You should watch these in sequence (or skip around depending on your curent level of knowledge in this domain):

Get the presentation slides here

Chapter 1
Chapter 2
Chapter 3
Chapter 4
Chapter 5

Chapter 6
Chapter 7
Chapter 8
Chapter 9
Chapter 10
Chapter 11
Chapter 12
Chapter 13

 

Embed

Format

Available formats for this video:

Actual format may change based on video formats available and browser capability.

    More episodes in this series

    Related episodes

    The Discussion

    • lucjansz

      Very nice shirt Smiley

    • AliPang

      Hope you get the time to do an introduction on QuickCheck. This is one of the things in Haskell that seems the most interesting to me, and I think it is a pretty unique to functional programming languages, if I understand it correctly.

    • paks8150

      Excelent lecture as always. Keep up the good work. Smiley

       

      In the book Hutton didn't explain why he chose List instead of the Maybe type. I'm glad you touch that subject in the lecture.

    • exoteric

      Erik, could you make the generalization offered by lists over optional types more statically correct with dependent typing? Saying that Maybe is just a list of zero or one element?

       

      In the ideal world we want to have lists as a generalization of Maybe but we don't want to loose the type safety that Maybe offers.

       

      This problem is complete Deja Vu for me: I encountered the same tension between the option type and the list type in an attempt to implement a lazy functional L-system library in haXe back in 2007.

       

      Thanks for the "epic fail" parser, haha.

       

      Deja Vu again! Brian Beckman repeated this so many times it was burned into my memory: the state monad is a function from state to state contents pair. This looks exactly as what we see here: you start with a string and get back a string and a tree.

       

      Erik, why do you not like the layout rule in this case and use explicit delineation. I don't think you mention this.

       

      This whole parsing business is reminiscent of unfolding. Expansion of some input into a more complex form. Composition, application and unfolding, hmm...

    • naasking

      While parser combinators can be expressed functionally on .NET, even in C#, after having written a few such combinator libraries I've found that Pratt parsers strike a better balance between simplicity and performance on the CLR. Pratt parsers also naturally handle operator precedence, while parser combinator libraries require more complicated structures to express these relationships. The extensible parser in the above link will be available in an upcoming release of my open source Sasa class library, so check it out if you're interested in this domain.

    • exoteric

      I've seen your blog before naasking and will check this out, thanks!

    • dpratt71

      Excellent as always, Erik.

       

      Perhaps like exoteric, I felt a little dissatisfied with the return type of Parser being "list of..." rather than "Maybe...". I've come to admire the expressive power of Haskell types and this seems like a bit of a hack. Considering, as you pointed out, that both the list and Maybe types comprise Monads, I am wondering where the advantage exists for using the list type. Monads allow for map, filter, lift operations, no?

    • ShinNoNoir

      The list type is generally used for parser combinators to return multiple possible parses.

       

      I don't own Hutton's book, but I can think of a few reasons for not using Maybe. It would require introducing a new datatype to the reader, while the reader is already familiar with the list type. Also, it takes more characters to write when pattern matching on a Maybe value Tongue Out.

       

      Now, in a real project you'd use the appropiate data type. If you know that at most only a single answer will be produced, use Maybe.

    • exoteric

      In general, this tension between structural abstraction and type-safety is conceptually dissatisfying.

       

      There must be a way to use the same list abstraction, only constrained to apply to [n,m] elements. Such value-restrictions upheld by a type system appear to be what dependent typing does. This provides a new expressive dimension which is missing in order to type programs correctly without also loosing the abstraction provided by lists over options. The ideal must be a way to fluently move from intensional to extensional semantics by explicitly coercing a type into the extensional domain of structural compatibility.

       

      It also reminds me of template metaprogramming. Someone built a compile-time regexp module in D and there's also a compile-time raytracer in D. In the last case you may discuss the applicability of that but at least it shows the expressive power.

       

      http://www.digitalmars.com/d/2.0/templates-revisited.html

       

      Now time to play with parsers in C# using LINQ...

    • Magnus Lidbom

      I'm mystified by Erik's comment that the inserting calls to parse "makes the type conversion explicit".

       

      I can certainly see that:

      parse p "something"

      is preferable to

      p "something"

       

      However, to me it seems that it makes semantics exlicit rather than type conversion.

       

      Erik hardly seems more likely to be confused here than I am, so I think I'm missing some fundamental alternative way of viewing this.

       

      As I'm writing this I think I'm starting to get it. Using parse makes it explicit that p is a parser. Since applying a parser to a value always returns a value with a type that has the same relationship to the input value's type, using parse makes both the semantics and the type conversions explicit.

       

      Am I understanding this correctly?

      If not, I'd appreciate it if someone could expand upon Erik's statements.

    • head.in.the.​box

      The balance between too little and too much typing is subtle.  You can goo all the way and have your types be a very precise specification of your code http://www.cs.chalmers.se/~bengt/talks/tt-snu.pdf or have no static types at all. I think Hindley-Milner is a good compromise, although I think something like Cayenne http://www.cs.chalmers.se/~augustss/cayenne/ should work in practice as well.

    • leomm

      Hello, I've tried the following code with Hugs:

       

      type Parser a = String -> [(a,String)]

       

      item :: Parser Char
      item = \inp -> case inp of
                     []     -> []
                     (x:xs) -> [(x,xs)]
                    
      return' :: a -> Parser a
      return' v = \inp -> [(v,inp)]

       

      p' :: Parser (Char,Char)
      p' = do x <- item
                  z <- item
                  y <- item
                  return' (x,y)

       

      It gives the following error:

       

      Prelude>

      p' "12345"

      ERROR file:{Hugs}\packages\hugsbase\Hugs.hs:94 - Type error in explicitly typed binding

      *** Term : p'

      *** Type : [Char] -> [(([(Char,[Char])],[(Char,[Char])]),[Char])]

      *** Does not match : Parser (Char,Char)

       

      Does anyone had this problem? Thanks in advance for any help.

    • zurk

      I must be going blind, but I can't seem to find the slides anywhere.  Also, did anyone manage to collect all the links Erik pointed-to/talked-about at the beginning of the presentation?

    • leomm

      I also couldn't find the slides.

       

      One of the links is this: http://www.comlab.ox.ac.uk/people/Richard.Bird/index.html

       

      Another one is this: http://portal.acm.org/portal.cfm?coll=GUIDE&dl=GUIDE&CFID=63594272&CFTOKEN=61762843

       

      Regards

    • Charles
    • leomm

      Thank you Charles. The source code, available at the same link, is great. Now I understand why my code wasn´t running.

       

      Regards

    • paks8150

      homework:

       

      2) Extend the parser for arithmetic expressions to support substraction and
         division, based upon the following extensions to the grammar:
              
              expr ::= term (+ expr | - expr | e)
              term ::= factor (* term | / term | e)

      expr' :: Parser Int expr' = do t <- term' do symbol "+" e <- expr' return (t + e) +++ do symbol "-" e <- expr' return (t - e) +++ return t term' :: Parser Int term' = do f <- factor' do symbol "*" t <- term' return (f * t) +++ do symbol
       "/" e <- expr' return (f `div` e) +++ return f factor' :: Parser Int factor' = do symbol "(" e <- expr' symbol ")" return e +++ natural eval' :: String -> Int eval' xs = case parse expr' xs of [(n,[])] -> n [(_,out)] -> error ("unused input " ++ out) [] ->
       error "invalid input" 

    • rehaskell9

      I'm getting the same error from my code. How did you fix your code?

    • siryc

      I cannot find a Gilat Braga's blog (I am not sure if I write it correct)... Can anyone help me?

    • Charles

      Hi,

       

      His name is Gilad Bracha. His blog is http://gbracha.blogspot.com/ 

       

      You can also watch him on C9: http://channel9.msdn.com/tags/Gilad+Bracha

       

      C

    • Sachin

      The second Paper is "The Genuine sieve of Eratosthenes" by Melissa O'Neill
      http://www.cs.hmc.edu/~oneill/papers/Sieve-JFP.pdf

    • David

      I do not agree with most as far as this one presentation goes... I really appreciated all the other ones but here...
      I lost you or you lost me or you lost it... whichever.

    Comments closed

    Comments have been closed since this content was published more than 30 days ago, but if you'd like to continue the conversation, please create a new thread in our Forums, or Contact Us and let us know.