[GHC] #971: Add intercalate and split to Data.List

[Josef Svenningsson]

OK. This suggests that there is a lot more to think about and discuss
when it comes to split. I'm going to remove split from my patch. It
should have its own proposal and I'll leave it for someone else to get
that ball rolling.

Thanks for the information.

Josef

On 30 Oct 2006 18:25:43 +0000, Jón Fairbairn <jon.fairbairn at cl.cam.ac.uk> wrote:
> "Josef Svenningsson" <josef.svenningsson at gmail.com> writes:
>
> > On 10/30/06, Christian Maeder <maeder at tzi.de> wrote:
> > > I also agree to add a split function. The problem was to agree which
> > > variant should be called split (wrt. to treatment of multiple and final
> > > separators as well as to "Eq" context or predicate argument).
> > >
> > > http://article.gmane.org/gmane.comp.lang.haskell.libraries/4862
> > > http://article.gmane.org/gmane.comp.lang.haskell.libraries/4869
> > >
> > Right. I had totally forgotten about that discussion. Was there any
> > consensus about the names?
>
> I'm not sure there was yet consensus about the functions --
> it seems I tried to have the bikeshed demolished, rebuilt
> as a pagoda and painted red:
>
> http://article.gmane.org/gmane.comp.lang.haskell.cafe/13807
>
> while there's some merit in ending discussion and deciding
> on /something/, I reckon it's worth trying to think about
> laws and such for a while yet.
>
> It strikes me that the discussion suggests that there are
> several distinct design cases: does a separator at the
> end/do two consecutive separators result in empty elements?
> Do the separators form part of the output?
>
> To sloganise for a moment, libraries should provide
> components, not finished products.  So what we should aim
> for is a collection of functions that facilitate the
> construction of all the design cases.
>
> One of the components /might/ be:
>
> > spanMb p [] = Nothing
> > spanMb p l = Just $ span p l
>
> As to nomenclature, since we already have span, I'd suggest
> spans p l should return a list of all the contiguous
> segments of l with members satisfying p (and discard the
> rest).
>
> Here's a definition:
>
> > spans p = unfoldr (fmap (id >< dropWhile (not . p)) . spanMb p)
> > (f >< g) (a,b) = (f a, g b)
>
> We can write "groupBy (\a b -> p a && p b)" for another of
> the design cases, though I think some discussion of groupBy
> belongs in here: why does it require an equivalence
> relation? Shouldn't we at least have a version that works
> for any binary predicate?
>
> Here's an off the top of my head ugly version:
>
>
> > runsBy = unfoldr . splitRun
>
> >          where splitRun p [] = Nothing
> >                splitRun p [x] = Just ([x],[])
> >                splitRun p (a:rest@(b:_))
> >                     | p a b = fmap ((a:) >< id) (splitRun p rest)
> >                     | otherwise = Just ([a], rest)
>
> (maybe splitRun is generally useful?)
>
>
> --
> J�n Fairbairn                                 Jon.Fairbairn at cl.cam.ac.uk
>
> _______________________________________________
> Libraries mailing list
> Libraries at haskell.org
> http://www.haskell.org/mailman/listinfo/libraries
>