Re: Dealing with empty aggregates?

[Josh Stone <jistone at redhat dot com>]

On 05/19/2011 03:05 PM, Jim Keniston wrote:
> On Thu, 2011-05-19 at 15:46 +0100, Bryn M. Reeves wrote:
>> On 05/19/2011 03:36 PM, Frank Ch. Eigler wrote:
>>> The canonical construct doesn't require a hand-maintained flag:
>>>
>>> probe timer.ms(1000) {
>>>     if (@count(stuff))
>>>        printf("Sum of stuff: %d\n", @sum(stuff))
>>>     else
>>>        printf("No stuff.")
>>> }
>>
>> Exactly what I was looking for, thanks! A count is not especially useful for the
>> stuff I'm looking at (nfsd operations fwiw) and it didn't occur to me to gate on
>> count (obvious really since it's the only one defined on an empty aggregate.. duh!).
>>
>> Cheers,
>> Bryn.
> 
> Another way: If all you're interested in is the sum, then you can
> initialize it with
> 	stuff <<< 0
> and then it'll never be empty.

We've debated empty-@sum behavior in the past, and IMO it makes perfect
sense to let @sum(EMPTY) = 0.  Frank has agreed to let this change, with
appropriate --compatible fallback behavior.  If there are no other
objections, then I'll go ahead and relax this.

Thanks,
Josh