Subject: Re: Partitioned tables and covering indexes
To: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>,
 Alexander Korotkov <a.korotkov@postgrespro.ru>
Cc: Jaime Casanova <jaime.casanova@2ndquadrant.com>,
 Pg Hackers <pgsql-hackers@postgresql.org>
References: 
 <CAJGNTeO=BguEyG8wxMpU_Vgvg3nGGzy71zUQ0RpzEn_mb0bSWA@mail.gmail.com>
 <48893c03-d1c8-62ef-b37d-b2f7dbdb8245@lab.ntt.co.jp>
 <CAPpHfdujMkQmhhzKG0sPB=27V9GXJeSVBGHt1rhHCPK89OfKfw@mail.gmail.com>
 <a5ca67cf-0448-7875-ae89-64495748c016@sigaev.ru>
 <994fa677-2dac-8fe5-3f39-024cd6406522@lab.ntt.co.jp>
From: Teodor Sigaev <teodor@sigaev.ru>
Message-ID: <5ee72852-3c4e-ee35-e2ed-c1d053d45c08@sigaev.ru>
Date: Wed, 11 Apr 2018 19:39:20 +0300
User-Agent: Mozilla/5.0 (X11; FreeBSD amd64; rv:52.0) Gecko/20100101
 Firefox/52.0 SeaMonkey/2.49.1
MIME-Version: 1.0
In-Reply-To: <994fa677-2dac-8fe5-3f39-024cd6406522@lab.ntt.co.jp>
Content-Type: text/plain; charset=KOI8-R; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Patch makes buildfarm almost red, patch is temporary reverted.

Actually, discovered bug is not related to patch except new test faces with it,
problem is: CompareIndexInfo() checks rd_opfamily for equality for all 
attributes, not only for key attribute.

Obviously, CompareIndexInfo() needs more work:
     for (i = 0; i < info1->ii_NumIndexAttrs; i++)
     {
         if (maplen < info2->ii_KeyAttrNumbers[i])

Seems, we can go out from ii_KeyAttrNumbers array.


Amit Langote wrote:
> Hi.
> 
> On 2018/04/11 0:36, Teodor Sigaev wrote:
>>>  ššš Does the attached fix look correct?š Haven't checked the fix with
>>> ATTACH
>>>  ššš PARTITION though.
>>>
>>>
>>> Attached patch seems to fix the problem.š However, I would rather get
>>> rid of modifyingšstmt->indexParams.š That seems to be more logical
>>> for me.š Also, it would be good to check some covering indexes on
>>> partitioned tables.š See the attached patch.
>>
>> Seems right way, do not modify incoming object and do not copy rather
>> large and deep nested structure as suggested by Amit.
> 
> Yeah, Alexander's suggested way of using a separate variable for
> indexParams is better.
> 
>> But it willš be better to have a ATTACH PARTITION test too.
> 
> I have added tests.  Actually, instead of modifying existing tests, I
> think it might be better to have a separate section at the end of
> indexing.sql to test covering indexes feature for partitioned tables.
> 
> Attached find updated patch.
> 
> Thanks,
> Amit
> 

-- 
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
                                                    WWW: http://www.sigaev.ru/