Re: Commitfest 2023-03 starting tomorrow! - Mailing list pgsql-hackers

So of the patches with no emails since August-December 2022:

* New hooks in the connection path

* Add log messages when replication slots become active and inactive
  - Peter Smith and Alvaro Herrera have picked up this one

* Remove dead macro exec_subplan_get_plan
  - Minor cleanup

* Consider parallel for LATERAL subqueries having LIMIT/OFFSET
 - No response since Sept 2022. There was quite a lot of discussion
and Tom Lane and Robert Haas expressed some safety concerns which were
responded to but I guess people have put in as much time as they can
afford on this. I'll mark this Returned with Feedback.

* pg_rewind WAL deletion pitfall
 - I think this is a bug fix to pg_rewind that maybe should be Ready
for Committer?

* Simplify find_my_exec by using realpath(3)
 - Tom Lane is author but I don't know if he intends to apply this in
this release

* Move backup-related code to xlogbackup.c/.h
  - It looks like neither Alvaro Herrera nor Michael Paquier are
particularly convinced this is an improvement and nobody has put more
time in this since last October. I'm inclined to mark this Rejected?

* Avoid hiding shared filesets in pg_ls_tmpdir (pg_ls_* functions for
showing metadata ...)
  - According to the internet "As part of their 39 month old
development and milestones, your patch should be able to see like an
adult (20/20 vision), be able to run, walk, hop and balance himself
while standing with one foot quite confidently." Can it do all those
things yet?

  - Should this be broken up into smaller CF entries so at least some
of them can be Ready for Committer and closed?

> * Fix bogus error emitted by pg_recvlogical when interrupted
  - Is this a minor cleanup?

> * Check consistency of GUC defaults between .sample.conf and pg_settings.boot_val
  - It looks like this was pretty active until last October and might
have been ready to apply at least partially? But no further work or
review has happened since.

> * Code checks for App Devs, using new options for transaction behavior
  - This is an interesting set of features from Simon Riggs to handle
"dry-run" style SQL execution by changing the semantics of BEGIN and
END and COMMIT. It has feedback from Erik Rijkers and Dilip Kumar but
I don't think it's gotten any serious design discussion. I posted a
quick review myself just now but still the point remains.

I think features supporting a "dry-run" mode would be great but
judging by the lack of response this doesn't look like the version of
that people are interested in.

I'm inclined to mark this Rejected, even if that's only by default. If
someone is interested in running with this in the future maybe
Returned with Feedback would be better even there really wasn't much
feedback. In practice it amounts to the same thing I think.

> * Lockless queue of waiters based on atomic operations for LWLock
 - Is this going in this CF? It looks like something we don't want to
lose though

> * Fix assertion failure with next_phase_at in snapbuild.c
  - It's a bug fix but it doesn't look like the bug has been entirely fixed?

> * Add sortsupport for range types and btree_gist
  - It doesn't l look like anyone has interested in reviewing this
patch. It's been bouncing forward from CF to CF since last August. I'm
not sure what to do. Maybe we just have to say it's rejected for lack
of reviewers interested/competent to review this area of the code.

> * asynchronous execution support for Custom Scan
  - This is a pretty small isolated feature.

> * CREATE INDEX CONCURRENTLY on partitioned table
  - I'm guessing this patch is too big and too late to go in this CF.
And it sounds like there's still work to be done? Should this be
marked RwF?

> * Partial aggregates push down

  - I'm not sure what the state of this is, it's had about a year and
a half of work and seems to have had real work going into it during
all that time. It's just a big change. Is it ready for commit or are
there still open questions? Is it for this release or next release?

> * Non-replayable WAL records through overflows and >MaxAllocSize lengths

  - Andres says it's a bug fix

-- 
greg