Re: array_agg versus repeated execution of aggregate final functions

Merlin Moncure <mmoncure(at)gmail(dot)com> writes:
> On Sat, Jun 20, 2009 at 1:31 PM, Tom Lane<tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> There are basically two ways that we could respond to this:
>>
>> (1) Decide that the bug is array_agg_finalfn's, and make the necessary
>> fixes so that it doesn't modify or free its argument ArrayBuildState.
>> This would amount to legislating that aggregate final functions cannot
>> scribble on their inputs *ever*, rather than the compromise position
>> we thought we had adopted last fall, namely that they can do so when
>> called by an Agg node but not when called by WindowAgg.

> This definitely sounds like the right answer.

Yeah. On reflection, even if we tried to go with #2, we'd still need
half of the #1 fix for array_agg_finalfn --- the detoasting business is
a hazard for WindowAgg's usage anyway.

regards, tom lane