Re: Aggregate extensibility

On 30/03/2010 5:59 PM, Steve Harris wrote:
> On 2010-03-30, at 16:52, Andy Seaborne wrote:
>>
>> On 30/03/2010 3:26 PM, Ivan Mikhailov wrote:
>>> Hello Axel,
>> ...
>>
>>>> Would we also allow aggregates with arityy different to 1 (like common in functions) e.g. agg-uri(?X ?Y)
>>>
>>> That's absolutely required for statistical calculations of all sorts,
>>> from regression to likelihood detection.
>>>
>>> Best Regards,
>>>
>>> Ivan.
>>
>> At F2F3, we decided to not restrict the arity to 1 for custom aggregates (and they can have parameters to the aggregator as well as arguments to aggregate over).
>
> We did? It sounds plausible, but I don't actually remember that. Is it a parser issue?

Stats aggregates were the main argument for n-ary custom aggregates. 
The whole multiset of tuples is based around it as well.

Several of the built-ins are 1-ary ; some are n-ary.
(and some I'm not clear about).

The parser will parse N-ary custom aggregates regardless because a 
custom aggregate look, syntactically, exactly like a custom function.

I don't see why we should not be able to write any built-in as a custom 
aggregate.  We should give URIs the aggregates we do define.

	Andy

>
> - Steve
>

Received on Tuesday, 30 March 2010 17:32:42 UTC