Re: Removing faulty hyperLogLog merge function

On Wed, Apr 27, 2016 at 12:08 PM, Joe Conway <mail(at)joeconway(dot)com> wrote:
> On 04/26/2016 07:23 PM, Robert Haas wrote:
>> On Tue, Apr 26, 2016 at 9:35 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>>>> I'm not prepared to commit this over the objection offered by Tomas
>>>> Vondra on that thread.
>>>
>>> FWIW, I agree with Peter that we should remove this code. We know that it
>>> is buggy. Leaving it there constitutes an "attractive nuisance" --- that
>>> is, I'm afraid that someone will submit a patch that depends on that
>>> function, and that we might forget that the function is broken and commit
>>> said patch.
>>>
>>> Tomas' objection would be reasonable if a fix was simple, but so far as
>>> I can tell from the thread, it's not. In particular, Peter doesn't trust
>>> the upstream patch in question. But whether or not you trust it, doing
>>> nothing is not a sane choice. The reasonable alternatives are to remove
>>> the merge function or sync the upstream patch.
>>
>> Now I agree with that. And now we do not have a 1-1 tie on which
>> alternative to prefer, which is a good start towards a consensus. Any
>> other views?
>
> I haven't followed this issue all that closely, but to me it seems
> pretty clear. If the function is brand new to 9.6, buggy, and not even
> used anywhere, I cannot imagine why we would leave it in the tree.

+1. We should definitely not encourage its use for 3rd-part plugins.
--
Michael