Re: Extended Statistics set/restore/clear functions. - Mailing list pgsql-hackers

From Michael Paquier
Subject Re: Extended Statistics set/restore/clear functions.
Date
Msg-id aXca7jnwtzFSYr4s@paquier.xyz
Whole thread Raw
In response to Re: Extended Statistics set/restore/clear functions.  (Corey Huinker <corey.huinker@gmail.com>)
Responses Re: Extended Statistics set/restore/clear functions.
List pgsql-hackers
On Sun, Jan 25, 2026 at 05:20:13AM -0500, Corey Huinker wrote:
> Ok, here's my attempt at splitting things out:
>
> 0001 - Add attribute stats test for multiranges

This was indeed independent of the rest, qualifying as a test defect
on v18, so applied first.

> 0002 - Basic restore functionality ndistinct

This has been combined with portions of 0007 to make sure that some
docs are there, with irrelevant portions related to other stats kinds
removed.

> 0003 - add dependencies

This one is simple.  It still needs a couple of adjustments (elog
messages on one line, one more test).

> 0004 - add mcv
> 0005 - add expressions

These would be next in line.

> 0006 - Add in the before/after diff test on a stats object (in theory this
> could be added after 0002)

I think that this would mean 0005 now.

> 0007 - add SGML

Partially done.

> 0008 - restore MAINTAIN test (in theory this could be merged into 0002)

I was hesitating to do exactly that, but just split things at the end.
This patch was actually incorrect, with regress_test_extstat_clear
as role name getting partially used.  The output was wrong.

Another side note: I am not sure how the CI or the buldfarm did not
notice that, but I've found a bug in the tuple initialization logic
while reviewing more the code, fixed in 114e84c532d1.  Perhaps my
build environment was more efficient in detecting that.
--
Michael

Attachment

pgsql-hackers by date:

Previous
From: Bertrand Drouvot
Date:
Subject: Re: Flush some statistics within running transactions
Next
From: Amit Langote
Date:
Subject: Re: unnecessary executor overheads around seqscans