[Haskell-cafe] safe lazy IO or Iteratee?

[John Millikin]

Both have advantages and disadvantages. The primary advantage of lazy
IO over iteratees is that it's much, *much* easier to understand --
existing experience with monads can be used immediately. The downsides
of lazy IO, of course, are well documented[1][2][3].

Some are fixed by the safe/strict IO packages. However, safe lazy IO
is still "unpredictable" in that it's difficult to know how many
resources will be used, the order of some operations (eg, releasing
handles), or whether some particular expression will throw an
exception.

Iteratees are useful because the author can make and verify guarantees
about performance (eg, "this code will never read more than 4 KiB at
once"). They also allow pure code to determine the behavior of IO,
safely, which is useful when writing libraries which must deal with
large amounts of data (eg, streaming a file over HTTP). The Hyena web
server is written using iteratees, and from what I've heard the
authors have been quite happy with their properties.

I've also found iteratees to perform somewhat better than lazy IO,
though I don't know why.

Downside: iteratees are very hard to understand. I wrote a
decently-sized article about them[4] trying to figure out how to make
them useful, and some comments in one of Oleg's implementations[5]
suggest that the "iteratee" package is subtly wrong. Oleg has written
at least three versions (non-monadic, monadic, monadic CPS) and I've
no idea why or whether their differences are important. Even dons says
he didn't understand them until after writing his own iteratee-based
IO layer[6].

tl;dr: Lots of smart people, with a history of being right about this
sort of thing, say iteratees are better. Evidence suggests
iteratee-based IO is faster and more predictable than lazy IO.
Iteratees are really hard to understand.

[1] http://okmij.org/ftp/Haskell/Iteratee/Lazy-vs-correct.txt
[2] http://www.kuliniewicz.org/blog/archives/2010/01/27/happstack-and-streaming-part-4-the-flaw/
[3] http://forums.xkcd.com/viewtopic.php?f=11&t=46780
[4] http://ianen.org/articles/understanding-iteratees/
[5] http://okmij.org/ftp/Haskell/Iteratee/IterateeM.hs , search for
"Drawbacks of this encoding"
[6] http://stackoverflow.com/questions/1319705/introduction-or-simple-examples-for-iteratee

On Thu, Feb 4, 2010 at 08:29, David Leimbach <leimy2k at gmail.com> wrote:
> Hi everyone,
> This is not an attempt to start a flame war.  I'm just trying to get a good
> feel for the advantages and disadvantages of the newer safe lazy io lib
> available on Hackage vs using Iteratee.
> It does appear to me that using something like Itereatee gives a bit of room
> to really tweak the low level enumerator/iteratee relationship, and, if you
> asked my old boss, I'm just a big fan of folds too, I use them a lot in my
> Erlang :-), so Iteratee really does appeal to me on many levels.
> Yet at the same time, I'm quite enamored with the beauty of "interact" and
> functions of that sort.  I realize mixing the effects of the lazy IO and
> pure code may not be the clearest way to write code for everyone, but there
> is something about being able to get linewise data as
> interact (unlines . fmap someLineWiseFunction . lines)
> that is just kind of cool.
> Dave
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>
>