Re: [stack] array theory question

On Dec 22, 2008, at 8:56 PM, John Cowan wrote:

> John Nowak scripsit:
>
>> What I'm hoping might make sense here is making the atom 'y'
>> equivalent to '{y}'. This would allow the second form of '&' shown
>> above to be used in the general case. Do you think that may be
>> workable? Is it comparable to what Nial or APL do?
>
> Ah^2.  What you want is Algol 68's "rowing coercion" ("row" being A68
> jargon for vector, or 1-D array).  This coercion transforms a non-row
> value into the corresponding row value with one element.

Very interesting! I suppose the question I'd need to answer is if it's  
sufficient to only go from scalar -> vector. I think the answer is  
"no" if functions like 'square' are meant to operate on vectors rather  
than scalars as then 'square 5' would return a one element vector and  
I'd be back to the same problem regarding construction. It would be  
sufficient however for functions like 'sum' that always return a  
scalar but can consume either a scalar or vector as input. For things  
like 'square', there's always map.

The choices I make here are going to end up depending on the  
complexity of the necessary type system more than anything else. It's  
quite likely that I won't be able to offer a map-like operation that  
works on the vector passed to a function. This is essentially the same  
state of things in a language like Factor where there's no way of  
applying a function to every element of the stack (without using with- 
datastack and then replacing the main stack or something awful).  
Instead, you have stack and arrays as separate concepts. If I go this  
route, the rowing coercion will be sufficient. The only place I think  
I really need to use it is for partial application anyway. It seems a  
shame though to need to add this notion to the type system just for  
this one case, although perhaps there's a nice way of working it into  
the notation so it seems obvious.

Thanks again.

- John