On: Hacking Haskell

[parfe]

>So you could say that IO isn't well explained by computer science yet

I don't think I could ever be sold on using a language where IO is an unsolved problem. Computer Science has defined it quite well for me in Python

[Peaker → parfe]

Python doesn't really have the equivalent power of Iteratees.

Iteratees let you compose chains of highly-efficient producers, transformers and consumers.

As a relatively trivial example, how would you write something in Python that performs this:

    for word in open("really_big_file", "rb").read().split(","):
        socket.send(word[::-1])

but does so with acceptable performance and memory use?

[parfe → Peaker]

Using a generator I think this does what you want. What does the equivalent haskell code look like?

    def read_chunked(f, delim=',', size=1024):
        buf = []
        while True:
            data = f.read(size)
            if not data:
                if buf:
                    yield ''.join(buf)
                break
            if delim in data:
                spl = data.split(',')
                buf.append(spl[0])
                yield ''.join(buf)
                for s in spl[1:-1]:
                    yield s
                rest = spl[-1]
                buf = [rest] if rest else []
            else:
                buf.append(data)

    for word in read_chunked(open("really_big_file", "rb")):
        #socket.send(word[::-1])
        print word[::-1]

I'm pretty sure I got the gist of what you were after. What do you think?

[Peaker → parfe]

Well, this is basically manually implementing Iteratee inline your solution. The idea is that Iteratees allow writing this as:

  main = run_ $
       enumFile someFile
    $$ splitWhen isComma
    $$ map reverse
    $$ iterHandle stdout
  where
    isComma = (== ',') . chr . fromIntegral

Basically Iteratees are a generalization of the technique you've implemented, but:

* Generalized for any error type and support various modes of error handling/propagation (they can go up to the producer, or down towards the deeper consumer as error inputs).

* Are generalized to many types of "buffers"

* Have a high-performance counterparts of standard "list functions"

* Are a standard base for many libraries to extend (http, xml, etc), and are faster, more powerful and more useful than standard strict IO libraries (Haskell's base libs, and Python's stdlib).

So both Python and Haskell have standard "strict IO" libraries. Iteratees are really a new idea -- that Python does not do well yet. So I don't think Haskell's progression of the state of the art in IO handling is really evidence that "IO is unsolved in Haskell".

[parfe → Peaker]

That is a nice feature. thanks.