Alexey Klyukin wrote:
> On Jan 22, 2010, at 4:38 PM, Robert Haas wrote:
>
>
>> On Fri, Jan 22, 2010 at 9:13 AM, Alexey Klyukin <alexk@waki.ru> wrote:
>>
>>> I think elog(WARNING) is less surprising for the end-user, unless there's an objection strong enough to include it
intothe documentation :)
>>>
>> I think the main possible objection would what Simon just wrote on the
>> other thread - that it's been this way for a while, and while someone
>> might think that a different decision about how to handle it would
>> have been better, there may be people counting on the current behavior
>> who will have to spend time and perhaps money making changes if we
>> change it.
>>
>
> Well, then we have to choose between a fixed number of unhappy users in the past and potentially increasing number of
unhappyusers in the future (if we admit the fact that this behavior is illogical). IMO if something behaves
counterintuitivelyto most users the behavior should be at least documented, if not fixed.
>
>
>
Well, as Tim Bunce pointed out, if we get his on_init patch users would
be able to choose which behaviour they wanted. So we don't necessarily
have to choose between what people think conforms to POLA and backwards
compatibility.
Right now I'm a bit hung on that patch because of the "lost GUC
placeholder" issue mentioned elsewhere - everything I have thought of so
far that might overcome it has been unspeakably ugly :-(
cheers
andrew