I was in a bit of a rush to send this email earlier before the table got
packed up from in front of me, so I forgot to mention these little
tidbits:
* container-classes uses MPTCs+fundeps to define the classes rather than
type families. My initial preference would have been to use an
associated type, except that due to the lack of superclass constraints
then the various *Functor type classes would have to have extra
constraints of the form "Value (c a) ~ a" in every single type
signature (and for every single value being used, so 3 extra
constraints of this form for zip3, etc.). I felt that this verbosity
was a bit too much and thus declined. Furthermore, I would have
needed to use an MPTC for the *Functor type-classes anyway as I would
have to state the value type somewhere in the class to ensure it was
properly constrained. However, once superclass constraints become
available (which will probably be 6.16 at the earliest from what I've
been told) then I'll switch it to using type families (assuming the
library is still being used of course).
* Container uses Monad as a super-class to avoid re-defining mempty and
mappend, though it does have (++) = mappend, so that's there for all
those people wanting to infiltrate Caleskell-style ideas into Haskell
proper...
* Someone's asked me why I don't define a Traversable-like class. The
reason is is that I haven't gotten around to it yet ;-) (as well as a
great many other things).
Ivan Lazar Miljenovic
I'm pleased to announce the fruits of my labours during AusHac [1]: version 0.0.0.0 of the container-classes library [2] I was talking about in my last blog post [3].
[1]: http://www.haskell.org/haskellwiki/AusHac2010 [2]: http://hackage.haskell.org/package/container-classes [3]: http://ivanmiljenovic.wordpress.com/2010/07/14/data-oriented-hierarchies/
The purposes of this library is two-fold:
* As a library writer, let your users choose which data structure they want to use rather than whichever one you decided upon when writing the library (e.g. you use lists predominantly, but your users may want to use Sets; by outputting a generic Container value yoru users don't need to keep using Set.fromList and Set.toList).
* When writing your own code and you realise that you're using the wrong data structure for your purposes, if you used this library (correctly) then the amount of work you need to do to swap data structures is minimised.
This current version is only a first draft. As such, I would appreciate feedback on the API. Currently, it provides its own definitions of all list-oriented functions in the Prelude plus a few more (e.g. partition), and list instances of all the classes.
In terms of definitions, most class methods have default definitions in terms of fold. Furthermore, as much as possible I've defined them using a build/fold setup outlined in the paper "A Short Cut to Deforestation" by Andrew Gill, John Launchbury and Simon Peyton Jones (however I haven't yet defined the rules to get this used yet). However, the current list instances use the pre-defined versions in the Prelude and Data.List; I'm not sure if I should keep it this way or - where equivalent - use the pre-defined ones so that hopefully conversions from lists to other data types would also be removed without needing intermediary data structures.
I also plan on adding more instances and a Lookup class for Maps, etc. However, we have to start packing up the room now and so my hacking is at an end for this weekend :(
-- Ivan Lazar Miljenovic Ivan.Miljenovic@gmail.com IvanMiljenovic.wordpress.com