[Grace-core] The arity problem

[James Noble]

> On 24/10/2015, at 11:46am, Andrew P Black <black at cs.pdx.edu> wrote:
> 
> I just came across this paper (or mine!) which nicely describes why we need overloading by arity.

indeed! 	

> The relevant sections are "The Partial Order of Types" on page 11-18 and "Form Partial order to Lattice", which follows.

So that makes the case that *types* need overloading on arity, to make type checking more elegant.
it doesn’t necessarily follow that all types must (potentially) be inhabited. 
In Grace we have the option of making different choices for classes and for types
(even though usually I’d argue against that in general…).
But that would give you a theoretically clean type lattice, 
without the practical / understandability / philosophical  
issues of actual overriding. 

I feel slightly dirty for suggesting it but I am on a 12 hour flight… 


In the paper you write:

> The solution, like many good ideas, is quite simple and in hindsight quite obvious: we needed to treat operations with different arities as if they had different names.

so we *already have* ways for operations with different arities to have different names:
one can always write multi-part methods and then this “problem” doesn’t exist.
There is a question about what we want to do about it - perhaps more generally about
method calls - varargs, arity, generalised messages, 
(everyone should go to Michael’s talk at the DLS, if that doesn’t leave NOOL empty :-)
I wonder if there are again simplicifications (perhaps on the object side
rather than in the types) that can rule out the more complex cases? 
Once again: orthogonality vs simplicity vs pragmatism.

anyway: something to talk about this week! 

and: I’m interested in Emeralds rules for manifest declarations & immutable objects.

James