new Group BY code

Started by Vadim B. Mikheevalmost 28 years ago4 messages
#1Vadim B. Mikheev
vadim@sable.krasnoyarsk.su
1 attachment(s)

I hope that GROUP BY fixed.
Unfortunately, I can't get any errors under FreeBSD
when using GROUP BY with aggregates. I only got core
for something like this: select y, z from tgrp group by y, z.
Two different funcs are used for these two cases and so
please test new code on your platforms.
Please test with different -S XXX to check that all works
for both in-memory and on-disk sorts (you'll see pg_psort.xxxx
files in last case).

New code are for 6.3. This is not patch (sorry, I lost orig version
of execnodes.h...) and has to be un-tarred in src dir.

If someone hasn't 6.3 - ok, I'll post patch for 6.2.1 tomorrow.

CVS updated.

Vadim

Attachments:

newGroup.tgzapplication/octet-stream; name=newGroup.tgzDownload
#2Bruce Momjian
maillist@candle.pha.pa.us
In reply to: Vadim B. Mikheev (#1)
Re: [HACKERS] new Group BY code

This is a multi-part message in MIME format.
--------------0478E3EE3CFC638DB65D6A96
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

I hope that GROUP BY fixed.
Unfortunately, I can't get any errors under FreeBSD
when using GROUP BY with aggregates. I only got core
for something like this: select y, z from tgrp group by y, z.
Two different funcs are used for these two cases and so
please test new code on your platforms.
Please test with different -S XXX to check that all works
for both in-memory and on-disk sorts (you'll see pg_psort.xxxx
files in last case).

New code are for 6.3. This is not patch (sorry, I lost orig version
of execnodes.h...) and has to be un-tarred in src dir.

If someone hasn't 6.3 - ok, I'll post patch for 6.2.1 tomorrow.

6.3 is only ten days away. I suggest you forget a 6.2.1 patch if it is
any trouble.

--
Bruce Momjian
maillist@candle.pha.pa.us

#3Vadim B. Mikheev
vadim@sable.krasnoyarsk.su
In reply to: Bruce Momjian (#2)
Re: [HACKERS] new Group BY code

Bruce Momjian wrote:

If someone hasn't 6.3 - ok, I'll post patch for 6.2.1 tomorrow.

6.3 is only ten days away. I suggest you forget a 6.2.1 patch if it is
any trouble.

I'm not sure that I will upgrade my production system to 6.3 in near feature ->
I need in 6.2.1 patch for myself. And - it's easy.

Vadim

#4Ronald Baljeu
rjb@xs4all.nl
In reply to: Vadim B. Mikheev (#1)
Re: [HACKERS] new Group BY code

I hope that GROUP BY fixed.
Unfortunately, I can't get any errors under FreeBSD
when using GROUP BY with aggregates. I only got core
for something like this: select y, z from tgrp group by y, z.
Two different funcs are used for these two cases and so
please test new code on your platforms.
Please test with different -S XXX to check that all works
for both in-memory and on-disk sorts (you'll see pg_psort.xxxx
files in last case).

New code are for 6.3. This is not patch (sorry, I lost orig version
of execnodes.h...) and has to be un-tarred in src dir.

If someone hasn't 6.3 - ok, I'll post patch for 6.2.1 tomorrow.

Vadim,

I downloaded the snapshot of Feb 18 and tried it. I noticed it's much
harder indeed to reproduce the bug (as compared to 6.2.1p6), but
occasionaly the bug appeared.

After installing your code the bug disappeared completely. I also tried
a couple of group-by queries on my production database, and tried both
in-memory sorting and on-disk sorting, but everything seems ok now.

So I guess congratulations are in order.

Thanks!

Cheers,
Ronald