Re: rq25 aggregate functions

On 22 Dec 2009, at 08:52, Andy Seaborne wrote:
...
>>> [[
>>> Definition: GroupConcat
>>>
>>> The multiset of values passed as an argument is converted to a
>>> sequence S.
>>>
>>> GroupConcat(S) = fn:string-join(S, " ")
>>>
>>> @@ do we want space as the seperator? It's annoying in names for
>>> example. Other reasonable choices include ",", "\n", or "\t".
>>> ]]
>>>
>>> All reasonable, none always right. I suggest that GroupConcat take  
>>> an
>>> optional separator defaulting to a single space.
>>>
>>> If the elements are "a" "b" "c"
>>>
>>> CONCAT(?x) -> "a b c"
>>> CONCAT(?x, "/") -> "a/b/c"
>>>
>>> If the elements are integers 1 2 3
>>>
>>> CONCAT(str(?x)+1) -> "2 3 4"
>>> CONCAT(str(?x)+1, "+") -> "2+3+4"
>>
>> I did consider that, but it conflicts with the semantics of  
>> aggregates
>> (both in SQL, and what's currently in the SPARQL draft).
>>
>> One option would be a keyword, like GROUP_CONCAT(str(?x) SEPARATOR  
>> "+"),
>> this is what MySQL does, but it means some special case syntax.
>
> I don't see it does necessarily.  The "str(?x)+1" is an expression,  
> which is passed, as an expression, to the aggregation operation.   
> The aggregator applies it to each partition so it's not being  
> evaluated where used, which I agree would be confusing.
>
> We already have count(*).

I'm not sure if that's an argument in favour or against.

If there's general feeling in favour of GROUP_CONCAT([DISTINCT] e, ...  
[SEPARATOR s]) then I'm more than happy to add it. I was in two minds  
anyway when writing that section.

MySQL has some ORDER BY options as well, and I don't want to start a  
precedent saying that you should pre-process by popping the last  
argument to get a separator value, and treat the rest as the  
expression list.

- Steve

Received on Tuesday, 22 December 2009 09:42:06 UTC