good point: the core libraries really should strike a balance between generality, consistency, and having informative / useful type inference for client users.

by the criterion, the foldable and traversable in prelude fits that objective to a T. (but i'm just repeating what other people have already said I suppose)


On Tue, May 21, 2013 at 5:27 PM, Edward Kmett <ekmett@gmail.com> wrote:
One nice thing about the limited palette of tools we're generally willing to allow ourselves in the base libraries is that it steers us towards a pretty reasonable middle ground.

Since MPTCs and fundeps are off the table for the forseeable future in terms of Prelude library design, we're more or less forced into the subset of Haskell that works and infers nicely. Is what Foldable/Traversable allow us to build strictly the most general thing that could be built? No -- but it does tend to run a pretty good balance between rigorous laws, flexibility, ease of understanding, and inference.

When I see these pie in the sky options about generalizing everything away almost universally those rely on extensions that frankly don't infer well.

Foldable and Traversable do not rise to that level of complexity and can be implemented entirely within a framework we've had fully fleshed out for 15 years.

When is it enough to just integrate an improvement and move on? Should we try again at 30 years or do you think someone might have accidentally written another book in the meantime?

-Edward


On Tue, May 21, 2013 at 4:44 PM, Henning Thielemann <lemming@henning-thielemann.de> wrote:

On Tue, 21 May 2013, Casey McCann wrote:

Being a beginner is by definition an ephemeral state; the entire
purpose of being a beginner is to eventually stop being one. Don't
design a language (or anything else) around the needs of beginners
unless you intend that only beginners will use it, in which case one
wonders why they're even bothering.

The other way round does not work, too. If you are trying to put the most general (thus most "useful" function) to Prelude, you will find that there will always be more general generalizations. Today you say

  mapM_ :: (a -> m b) -> [a] -> m ()

should be replaced by more general

  mapM_ :: (a -> m b) -> f a -> m ()

But this is not the most general form. We have discussed several times about generalizing the (), without a conclusion. But maybe some day there might be consensus to generalize the () away ... what will be the next generalization? We may generalize the (->) away. What else?

For me it is perfect to start with something simple in Prelude and build more general functions around it in other modules. This is a scheme that we can manage over time with backwards compatibility.


_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://www.haskell.org/mailman/listinfo/libraries


_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://www.haskell.org/mailman/listinfo/libraries