Re: FETCH FIRST clause WITH TIES option

From: Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>
To: Surafel Temesgen <surafel3000(at)gmail(dot)com>
Cc: David Steele <david(at)pgmasters(dot)net>, Michael Paquier <michael(at)paquier(dot)xyz>, Robert Haas <robertmhaas(at)gmail(dot)com>, andrew(at)tao11(dot)riddles(dot)org(dot)uk, PostgreSQL Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: FETCH FIRST clause WITH TIES option
Date: 2019-04-06 22:39:26
Message-ID: 20190406223926.oy7qpvz327mnwwnv@development
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi Surafel,

I've been looking at the patch with the intent to commit it, but once
again I ran into stuff that seems suspicious to me but am not familiar
with enough to say if it's OK. I'm sorry about that :-(

First, a couple of tweaks in the attached v9 of the patch:

1) I've removed the costing changes. As Tom mentions elsewhere in this
thread, that's probably not needed for v1 and it's true those estimates
are probably somewhat sketchy anyway.

2) v8 did this (per my comment):

- ExecSetTupleBound(compute_tuples_needed(node), outerPlanState(node));
+ if(node->limitOption == EXACT_NUMBER)
+ ExecSetTupleBound(compute_tuples_needed(node), outerPlanState(node));

but I noticed the comment immediately above that says

* Notify child node about limit. Note: think not to "optimize" by
* skipping ExecSetTupleBound if compute_tuples_needed returns < 0. We
* must update the child node anyway, in case this is a rescan and the
* previous time we got a different result.

which applies to WITH_TIES too, no? So I've modified compute_tuples_needed
to return -1, and reverted this change. Not sure, though.

3) I'm a bit confused by the initialization added to ExecInitLimit. It
first gets the tuple descriptor from the limitstate (it should not do so
directly but use ExecGetResultType). But when it creates the extra slot,
it uses ops extracted from the outer plan. That's strange, I guess ...

And then it extracts the descriptor from the outer plan and uses it when
calling execTuplesMatchPrepare. But AFAIK it's going to be compared to
the last_slot, which is using a descriptor from the limitstate.

IMHO all of this should use descriptor/ops from the outer plan, no? It
probably does not change anything because limit does not project, but it
seems confusing.

cheers

--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Attachment Content-Type Size
fetch_first_with_ties_v9.patch text/plain 28.5 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2019-04-06 23:36:32 Re: jsonpath
Previous Message Alexander Korotkov 2019-04-06 22:10:06 Re: jsonpath