2016-08-29 4:51 GMT-03:00 Michael Paquier <michael.paquier@gmail.com>:
>
>> I see the current behavior is documented, and I do understand why global
>> objects can't be part of the extension, but for indexes it seems to violate
>> POLA a bit.
>>
>> Is there a reason why we don't want the extension/index dependencies?
>
> I think that we could do a better effort for indexes at least, in the
> same way as we do for sequences as both are referenced in pg_class. I
> don't know the effort to get that done for < 9.6, but if we can do it
> at least for 9.6 and 10, which is where pg_dump is a bit smarter in
> the way it deals with dependencies, we should do it.
ATM I don't have a strong opinion one way or the other regarding the
dependency of indexes and extensions. I believe we have to put more
thought into it, and at the end we might just leave it as it is.
What I do believe is that this requires a separate thread, and if
agreed, a separate patch from this issue.
I'm going to prepare another patch where I'm going to strip the tests
for external indexes which are failing now. They actually fail
correctly as the table they depend on will not be dumped, so it's the
developer/DB designer who has to take care of these things.
If in the near or not so near future we provide a patch to deal with
these missing dependencies, we can easily patch pg_dump so it deals
with this correctly.
Regards,
--
Martín Marqués http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services