[Haskell-cafe] Re: Numeric type classes
Jacques Carette
carette at mcmaster.ca
Thu Sep 14 15:54:21 EDT 2006
David Menendez wrote:
>> * Having (+) work on lists, tuples and all the other monoids would
>> make error messages more complicated.
>>
>
> It gets worse than that. Imagine trying to explain to someone why "1 +
> sin" is actually "\a -> const 1 a + sin a".
>
It isn't that hard - it is done routinely in mathematics courses. In
fact, that is what 1+sin means in Maple today (and has for 25 years).
It is also what it means in MuPAD. AFAIK, that is also what 1+Sin means
in Mathematica. That is what polymorphism is all about! [This is really
equational-theory polymorphism rather than parametric polymorphism, but
that's a minor detail, since Monad polymorphism is _also_
equational-theory polymorphism].
This kind of polymorphism [where you add the 'right number' of arrows on
the left] is quite useful. Things like differential operators become
quite tiresome to write down if you have to pedantically spell
everything out, even though there is only one 'sensible' way to
interpret a given expression [1].
In the very same way that fromInteger can project a literal integer into
other typeclasses, one can project values into spaces of functions by
just "adding arrows on the left" (ie exactly what const does). It is
possible to make this quite formal, but you need Natural(s) (as an
additive monoid) on the type level, and then be able to be polymorphic
over _those_ to do make it all work. It should even be decidable [but
that part I have not checked]. Something I should write up one of these
days, but in the meantime go read [1]!
Jacquces
[1] Bjorn Lisper and Claes Thomberg have already investigated something
very close to this, see
http://www.mrtc.mdh.se/index.php?choice=publications&id=0245
More information about the Haskell-Cafe
mailing list