Hi Duncan On 30 Jan 2008, at 10:53, Duncan Coutts wrote:
On Tue, 2008-01-29 at 12:53 +0000, Conor McBride wrote:
What's orthogonal to what, here? I'm trying to understand why the generalization is irrelevant (or merely unwelcome) to the discussion of whether or not the special case should be added to the library.
I think the point is that it'd be a bit inconsistent if the only function in Control.Monad that was generalised to arbitrary Traversable, Monoid etc was concatMapM.
This may be a reasonable distinction between concerns, but they're hardly orthogonal. The question of whether or not such generalizations are likely or desirable, sooner or later, is clearly relevant to the issue of how lame a duck concatMapM is likely to be, hence whether or not it should be added.
The point continues that if we do indeed want to generalise all the functions in Data.List and/or Control.Monad then that should be a separate proposal.
An interesting question is the extent to which this has already happened. My issue with concatMapM has, all along, been that it is a special case of foldMap, given the appropriate instance of Data.Monoid. We don't have to contemplate adding a new super duper general thrudsplingblatter to the library, giving it a silly name the better to travesty the mere suggestion, because foldMap is already in the library: it's a question of learning to make more and better use of it. I guess I'm advocating a general policy of "more structures; fewer functions", exploiting the power of instance inference to deliver the usual structure-respecting whatever, rather than giving all the instances individual names. Less is more. Or maybe I'm just a grumpy old math-troll trying to hassle people away from getting on in the obvious manner with some obviously desirable functionality. You choose Conor