On Mar 28, 2011, at 12:54 AM, wren ng thornton wrote:
On 3/27/11 9:58 PM, John Millikin wrote:
Resending is slightly more complex -- if the other end can say
"resend that
last chunk", then it should be easy enough, but "resend the last 2
bytes of
that chunk you sent 5 minutes ago" would be much harder. What is
your use
case?
This does highlight one of the restrictions I've lamented about the
iteratee framework. Namely that the current versions I've seen place
unnecessary limitations on the communication the iteratee is allowed
to give to the enumerator/enumeratees above it. This is often
conflated with the iteratee throwing an error/exception, which is
wrong because we should distinguish between bad program states and
argument passing. Moreover the type system doesn't capture the kinds
of communication iteratees assume of their enumerators/enumeratees,
nor the kinds of communication supported by the enumerators/
enumeratees, which means that failure to hook them up in the right
(non-typechecked) way /does/ constitute an error.
The one example that tends to be supported is the iteratee
requesting that the enumerator/enumeratees seek to a given position
in a file. Which is a good example, but it's not the only one.
Requesting the resending of chunks is another good example. But
there's no limit to the reasonable kinds of communication an
iteratee could want.
In an ideal framework the producers, transformers, and consumers of
stream data would have a type parameter indicating the up-stream
communication they support or require (in addition to the type
parameters for stream type, result type, and side-effect type). That
way clients can just define an ADT for their communication protocol,
and be done with it. There may still be issues with the Expression
Problem, but at least those are pushed out of the stream processing
framework itself which really shouldn't care about the types of
communication used.
It's somewhat outdated and underdeveloped (I was writing for myself so
I never really bothered finishing it), but I wrote an exploration of
iteratee semantics[1] a while back in which I specified an iteratee as
a monad-transformer stack involving, at its core, the "PromptT" or
"ProgramT" monad transformers (as far as I know, the same could be
done with the "Coroutine" monad). I personally found that
construction far more lucid than the usual ad-hoc view, and it also
makes it very clear how the model can be trivially extended to support
additional operations such as these.
Based on what I learned while writing that (and on the similarity
between coroutines and the concepts I used), I strongly agree with
Mario Blažević, suggestion to look at his monad-coroutine library as
a way of understanding where they fit in some larger design space. I
would even go so far as to suggest that something like it could be
considered as either a replacement for iteratees or as the underlying
implementation of an iteratee library, because the concept not only
subsumes iteratees and enumerators, but also delegates control to code
that can be independent of both rather than simply reversing the
"conventional" iterator concept. I believe it also subsumes iterators
and whatever their corresponding parts are called.
As he mentions, his implementation does not come with all the
"plumbing", but I think it would be worthwhile to create that
plumbing, because either coroutines or "operational monads" may very
well be the basis needed to develop a "grand unified theory" of
composable stream processing. If nothing else, the isomorphisms in
his coroutine-enumerator[2] and coroutine-iteratee[3] packages seem to
give a much more direct and useful iteratee semantics than I've seen
given anywhere else, and at the same time they are much more readily
extended to cover additional operations.
-- James
1. https://github.com/mokus0/junkbox/tree/master/Papers/HighLevelIteratees
2. http://hackage.haskell.org/package/coroutine-enumerator
3. http://hackage.haskell.org/package/coroutine-iteratee
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe