I guess we just need to agree that we disagree. Of course, you are right that I am just a newbie when it comes to Haskell. How did you know that I am actually a Cobol programmer? Did you google? http://www.google.com/search?biw=1003&hl=en&q=Ralf+Cobol Seriously (?): 1. I never intended to impose any particular OO style on Haskeller's. Its more about the *intellectual* question whether we can take C++/C#/Java code and do a structure-preserving conversion to Haskell without starting to argue on paradigm conversion. Or as Alistair Bayley argues (Alistair, thanks for helping out!) http://www.haskell.org/tmrwiki/FpVsOo "... it might be that you need to interface with external OO code, or you are porting an existing program and need to remain faithful to the original design, if only because you don't understand it well enough to convert it to a functional design. ..." Bulat, why not read all of this? Also see the end of that Wiki page, where a reference to a paper is given, which seems to solve all these problems, but my brain is to small to really be sure. ;-) 2. No matter how much you try, the code that you showed was not very close to a faithful Shapes implementation. If you dare to read the spec: http://onestepback.org/articles/poly/ http://www.angelfire.com/tx4/cus/shapes/ You made two proposals that don't solve the problem no matter of *style*: 1st proposal of Bulat:
just create list of draw functions itself:
[drawCircle (10,10) 5, drawSquare (20,20) 10]
you are not expected that this problem can be solved with one line of code? :)
My reply: you are not building a list of objects. You are not organizing a loop whose body iterates over a list and executes functionality. 2nd proposal of Bulat:
for more complex tasks - declare interface as a structure:
data ShapeInterface = Shape { draw :: IO (), moveTo :: Point -> IO (), calcArea :: Float }
and return this structures from "constructor" functions:
circle x y r = Shape { draw = drawCircle center r, moveTo newCenter = ...., calcArea = pi*r*r } where center = Point x y
square x y size = Shape { draw = ...., moveTo newCenter = ...., calcArea = size*szie }
figures = [circle 1 2 3, square 4 5 6, circle 7 8 9]
The tragedy here is that this stops to work once circles and squares do have different interfaces. You solution does not scale. Subtyping allows for interface extension, and the actual benchmark exploits this, but your code avoids it. Anyway the solution is trivial: ex. quantification, you didn't propose that. Cheers, Ralf (back to Cobol)
-----Original Message----- From: Bulat Ziganshin [mailto:bulatz@HotPOP.com] Sent: Thursday, June 23, 2005 11:37 PM To: Ralf Lammel Cc: Pal-Kristian Engstad; haskell-cafe@haskell.org Subject: Re: [Haskell-cafe] RE: Re[4]: [Haskell] Dynamic binding
Hello Ralf,
Thursday, June 23, 2005, 11:40:13 PM, you wrote:
RL> a) building (i) a list of data is fundamentally different RL> from building (ii) a list of anticipated results of each datum. RL> I would be surprised to hear that this counts as a valid technique. RL> BTW, you can do the *same* in a lazy OO language. (No lazy OO language RL> at hand -- well lazyness can be simulated.)
sorry, "valid" technique is technique that work :) i use in my own program all the three ways i mentioned to solve problems of different complexity
afair, you are relatively new to Haskell (relatively to my 1-year experience :) and, i think, you are not share FP thinking style. when programming in Haskell, i think in terms "what thing i will need in this point of program?". if i need, for example, a possibility to check string against regular expression, then i will pass a function which does this check, not original RE. if i need a possibility to draw a shape, i will pass action which draws this shape. if i need several functions, i just combine them in a tuple
there is one interesting example in my program. i have a list of wildcards and list of filenames and for each filename i need to know number of first wildcard to which this filename matched. in early stages of my program development i just passed list of wildcards to file-finding routine (as [Wildcard]). then, i changed list of wildcards to list of functions which check match against each wildcard ([Filename->Bool]). and after that, i changed this to one function which just finds first True answer (Filename->Int). it was also more effective to compute this function only one time (it was something compiled on moment of computing and worked as fast as hand-written analyzer for given set of wildcards)
as you see, i slowly migrated from traditional way of solving this problem to perfectly functional way and it was required several months
RL> a) building (i) a list of data is fundamentally different RL> from building (ii) a list of anticipated results of each datum.
i think that you don't have "Haskell brain" ;) and therefore don't "trust" functions, which are passed as parameters, collected in lists, saved in data structures and so on. you are prefer to "put hands on" some data, preferably an object, which can be manipulated with any method declared in his interface. i'm right? ;)
FP encourage another way - passing functions which will then be applied to some additional arguments, as in my program, where file-finding function absolutely don't need list of wildcards. it just need to map filename to wildcard number, so a Filename->Int parameter is just enough
in the draw example, each elment in a list was an action (having type IO() ), so i don't create list of anticipated results, i created list of actions which can be performed, for example, by sequence_
RL> Anyway, even if people end up coding as you propose, RL> it won't work in general. Think of mutation methods that RL> change the state but preserve the type. Then your list will RL> still be heterogonous. NO?
my second example was just of this type. it uses IORefs to hold current state, but this IORefs don't need to appear in interface
see method moveTo, which changes state variable center, and method draw, which uses this variable
this state is just don't need to appear in interface definition :) circle x y r
RL> You are not talking about state but constructor arguments.
"interface definition" is a structure ShapeInterface, which have only fields for exposed object methods. so it supports any figures in universe :)
RL> In OO, mutable state tends to leak to the interface, RL> perhaps not as public fields, perhaps not even as public
only in C++ and other languages which need to calculate object size :) declaration of _non-public_ fields in _interface_ is something strange ;)
this state is just don't need to appear in interface definition :)
if you need to maintain mutable state, this is also not a problem:
data ShapeInterface = Shape { draw :: IO (), moveTo :: (Int,Int) -> IO (), calcArea :: Float } circle x y r = do center <- ref (x,y) return Shape { draw = val center >>= drawCircle r , moveTo = (center=:) , calcArea = pi*r*r } main = do figures <- sequence [circle 1 2 3, square 4 5 6, circle 7 8 9] mapM_ draw figures mapM_ (moveTo (0,0)) figures mapM_ draw figures
ref=newIORef val=readIORef (=:)=writeIORef
-- Best regards, Bulat mailto:bulatz@HotPOP.com