Subject: matching vs. F-bounded
From: Vassily Litvinov (vass@cs.washington.edu)
Date: Wed Apr 18 2001 - 14:45:04 PDT
Here's a point of comparison between the two, hinted at in the LOOM paper.
I am not sure whether this addresses Craig's question or not.
Imagine you have type Ordered (or PartiallyOrdered as in the F-bounded
paper) denoting things that can be compared using le(). For example,
numbers are Ordered and so are strings.
Then you want to write a polymorphic sorting method, say sort(), that
takes an array and sorts it using le(). The method is polymorphic in the
type of array elements.
You want to pass an array of numbers or another array of strings to
sort(). Up to this point both matching and F-bounded quantification let
you do it.
But when you pass an array of mixed integers and floats (which both are
subtypes of number or, with matching, both match number), then matching
gives up. F-bounded still works fine.
Here's Cecil code; writing corresponding #-based code is left to the reader.
type Ordered[T];
signature le(`T,`T):bool where T<=Ordered[T];
method sort(a:array[`T<=Ordered[T]]):void {
... if( le(a!i, a!j) , (-- swap the two --) ); ...
}
type number subtypes Ordered[number];
type int subtypes number;
type float subtypes number;
implementation le(x@:number, y@:number):bool { ... }
-- can do multiple dispatch here, but that's not the point
type string subtypes Ordered[string];
implementation le(x@:string, y@:string):bool { ... }
-- examples of message sends; I onlly list argument types
sort(array[number]) -- in LOOM can't pass array of mixed ints and floats
sort(array[int|float]) -- not supported in LOOM
sort(array[string])
sort(array[number|string]) -- type error
Vass
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Wed Apr 18 2001 - 14:46:05 PDT