0
$\begingroup$

I stumbled upon some question that puzzled me, maybe it's just a feature (or simply because I am doing first ``Haskell-steps'' without studying the manual too deeply, which I guess I should...

Anyway, the observation want to ask about is as follows: I did some inductive data type definition, though it's ``non-regular'', i.e., requiring polymorphic recursion. I am aware that this is problematic, so I wanted to see to what extent Haskell can handle it. The actual code is at the end of the post.

Ultimately, what I don't understand is the difference in behavior comparing - doing a :load file.hs in the ghci as opposed to - typing in the definitions manually inside the ghci.

Actually, I am not 100% sure if it has to do with the particular kind of recursion (though for simpler things I did, no such discrepancy showed up.

For concreteness sake, the code looks like the one below: The non-regular data structure is SList (strange list). The datatype in isolation works both with loading the file or typing it manually. I am not surprized by that part. More interesting and the cause of the problem is the function slength. It's kind of an instance of polymorphic recursion and thus potentially problematic. On the other hand, the non-regularity of the type parameter a vs. Node a, vs. Node (Node a)) etc is not really relevant in the inductive definition.

Indeed, the whole code loads fine in GHCi version 8.6.5, though clipping it in fails, reporting as failure that ``Occurs check: cannot construct the infinite type''. Also this seems understandable, caused by the polymorphic recursive definition of slength. What I don't understand is, why both ways of doing things in ghci behave differently (BTW also compiling with ghc works)

data Node a  = Pair a a 

data SList a =  Nil  |   Cons a (SList (Node a))

slength :: SList a -> Int
slength Nil = 0
slength (Cons n r) = 1 + (slength r)
$\endgroup$
2
  • 2
    $\begingroup$ "clipping it in fails". Are you, but any chance, copy-and-pasting into GHCi one line at a time? If so, that's your problem. GHCi is a REPL. It'll read and evaluate each line when you paste it. When you paste a multi-line definition one at a time, it'll try to evaluate those as incomplete expressions. Pasting in the entire slength function in one go, using :{ and :} as delimiters, causes no errors on my machine. $\endgroup$ – Mark Seemann Feb 23 '20 at 11:34
  • $\begingroup$ Thanks, it solves it! I was not sure if it was something ``esoteric'', having to do with perhaps that these kind of definitions are generally problematic (other functional languages like ocaml cannot deal with it), and perhaps loading code of a module activates some language extensions (the code is in some module) in the background. On the other hand, I could not rule out something trivial like basic misunderstanding So I was not sure at which end I should start looking. As you pointed out; the problem was the basic interactions. Thanks for the super-fast help. $\endgroup$ – Martin Steffen Feb 23 '20 at 12:17

Your Answer

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

Browse other questions tagged or ask your own question.