Message-ID: <d7ff8f5a-e64c-4bc9-a902-9795d3645df1@postgrespro.ru>
Date: Thu, 10 Oct 2024 13:41:01 +0300
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: BUG #18652: Planner can not find pathkey item to sort for query
 with expression and expression index
To: Andrei Lepikhov <lepihov@gmail.com>, Tender Wang <tndrwang@gmail.com>
Cc: Richard Guo <guofenglinux@gmail.com>, exclusion@gmail.com,
 pgsql-bugs@lists.postgresql.org, Tom Lane <tgl@sss.pgh.pa.us>
References: <18652-deaa782ebcca85d1@postgresql.org>
 <2527690.1728510183@sss.pgh.pa.us>
 <CAMbWs49Pcnm_cX=7azYY9d8+qRrOOKDjJDCa0KGRk9kWWMOz3A@mail.gmail.com>
 <ea7e53bf-5aad-4a0a-8f73-a21dd8f6969b@gmail.com>
 <CAHewXNmFmXFxK7bsgAhqv_ToGxK4y7niNVL3GAzgWEOKuiAg5w@mail.gmail.com>
 <8026fe90-e661-4448-b9cc-bcbb65629b7b@gmail.com>
Content-Language: en-US
From: Alena Rybakina <a.rybakina@postgrespro.ru>
In-Reply-To: <8026fe90-e661-4448-b9cc-bcbb65629b7b@gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: 
 <https://www.postgresql.org/message-id/d7ff8f5a-e64c-4bc9-a902-9795d3645df1%40postgrespro.ru>
Precedence: bulk

On 10.10.2024 13:23, Andrei Lepikhov wrote:
> On 10/10/24 16:02, Tender Wang wrote:
>>
>>
>> Andrei Lepikhov <lepihov@gmail.com <mailto:lepihov@gmail.com>> 于2024年 
>>      > I think this is because find_computable_ec_member() only tries to
>>      > match Vars.  Maybe we should teach it to also match OpExprs?
>>     Looking into that case, I don't understand only one thing:
>>     generate_orderedappend_paths decided to try MergeAppend; the
>>     create_append_path routine added the Sort cost,  but the Sort node
>>     itself wasn't added. Maybe the origin problem is the lack of
>>     feasibility
>>     examinations?
>>
>>
>> Yeah, I'm also curious why only cost Sort but not adding Sort node in 
>> create_merge_append_path().
>> The comments say that " We'll need to insert a Sort node, so include 
>> cost for that".  Does another place
>> insert the Sort node?
> Before inserting the Sort node, we must identify the column 
> corresponding to each path key. And here is the problem: IndexScan has 
> two elements in the target list, but SeqScan has only one. It already 
> looks strange to me. How do we UNION two sources with different 
> numbers of resulting columns? It seems to me we have a bug under the 
> Append node.
>
Maybe a mistake exists at the stage of creating a sequential scan? 
Because the subpath of the MergeAppend path does not have any elements 
in pathkeys, although the pasthtarget has one element and this 
expression wasn't involved anywhere?

I honestly don’t understand at what stage this expression i+1 appears:

SELECT i + 0 AS c FROM (SELECT i FROM t UNION ALL SELECT i + 1 FROM t)

-- 
Regards,
Alena Rybakina
Postgres Professional