Re: contrib modules and relkind check

On Fri, Mar 10, 2017 at 8:59 AM, Amit Langote
<Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote:
> Hi Stephen,
>
> On 2017/03/10 6:48, Stephen Frost wrote:
>> Amit, Michael,
>>
>> * Amit Langote (Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp) wrote:
>>> On 2017/03/09 11:51, Michael Paquier wrote:
>>>> OK, I am marking that as ready for committer.
>>>
>>> Thanks.
>>
>> Thanks for this, I've pushed this now. I do have a few notes about
>> changes that I made from your patch;
>>
>> - Generally speaking, the user-facing functions come first in these .c
>> files, with a prototype at the top for the static functions defined
>> later on in the file. I went ahead and did that for the functions you
>> added too.
>>
>> - I added more comments to the regression tests, in particular, we
>> usually comment when tests are expected to fail.
>>
>> - I added some additional regression tests to cover more cases,
>> particularly ones for things that weren't being tested at all.
>>
>> - Not the fault of your patch, but there were cases where elog() was
>> being used when it really should have been ereport(), so I changed
>> those cases to all be, hopefully, consistent throughout.
>
> Thanks a lot for all the improvements and committing.

Thanks. Shouldn't this fix be back-patched? pg_visibility should fail
properly for indexes and other relkinds even in 9.6. pgstattuple can
also trigger failures. It would be confusing for users to face "could
not open file" kind of errors instead of a proper error message. Note
that I am fine to produce those patches if there is a resource issue
for any of you two.

+-- an actual index of a partitiond table should work though
Typo here => s/partitiond/partitioned/
--
Michael