varying number of arguments restriction

[Brian Boutel]

Ashley Yakeley wrote:
> 
> At 2001-10-30 11:01, Hal Daume wrote:
> 
> >obviously i can rewrite:
> >
> >foo [] = ""
> >foo s  = (snd . head) s
> >
> >but this is uglier.
> 
> I'm not sure. I actually prefer it written out so that the number of
> arguments in the cases matches (as GHC enforces).
> 

It's defined in the Report, not a GHC idiosyncracy. As to why, I don't
really remember, but I suspect it had to do with a desire by some
members of the Haskell Committee to require that the patterns in all
clauses of a function binding were disjoint, so that reasoning about
programs could deal with each clause independently. This was not
adopted, and the alternative top-to-bottom, left-to-right, semantics
were, but there was still a feeling that good style demanded
disjointness.

In that style, the second clause could be written

foo ((x,y):xs) = y

I don't know whether this is still true, but it used to be argued that
this was likely to be more efficient because compilers could produce
really good pattern-matching code.

--brian