[Haskell-cafe] RE: simple function: stack overflow in hugs vsnonein ghc

[Claus Reinke]

> afraid not
> 
> the given example is too strict, the requirement is to generate the matched 
> portion lazilly, and return the tail (unconsumed portion).

ah yes, without optimisations, Prelude.span builds up stack,
while the continuation-based alternative i mentioned is too
strict for some uses.
 
> In principle the function should be capable of being written to run in 
> constant space which the given example dose not.


>>>return (repeat 'a') >>= \ x -> print $ span (const True) x

how about the old spec, then?

    span p l = (takeWhile p l,dropWhile p l)

since takeWhile takes forever, here, it isn't even inefficient!-)

claus

>>>with hugs you will get a stack error, in ghc it executes in constant 
>>>space, i.e. indefinitely. In essenece the above example does exactly the 
>>>same as my ealier code.
>>
>>this thread might be relevant:
>>
>>http://www.haskell.org/pipermail/hugs-bugs/2007-June/001815.html
>>http://www.haskell.org/pipermail/hugs-bugs/2007-June/001816.html
>>http://www.haskell.org/pipermail/hugs-bugs/2007-June/001817.html