Re: Polyphase merge is obsolete

Peter Geoghegan <pg@bowt.ie>

From: Peter Geoghegan <pg@bowt.ie>

To: John Naylor <john.naylor@enterprisedb.com>

Cc: Heikki Linnakangas <hlinnaka@iki.fi>, Jaime Casanova <jcasanov@systemguards.com.ec>, vignesh C <vignesh21@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Tomas Vondra <tomas.vondra@2ndquadrant.com>

Date: 2021-10-05T21:44:37Z

Lists: pgsql-hackers

On Tue, Oct 5, 2021 at 10:25 AM John Naylor
<john.naylor@enterprisedb.com> wrote:
> int64 is used elsewhere in this file, and I see now reason to do otherwise.

Right. The point of using int64 in tuplesort.c is that the values may
become negative in certain edge-cases. The whole LACKMEM() concept
that tuplesort.c uses to enforce work_mem limits sometimes allows the
implementation to use slightly more memory than theoretically
allowable. That's how we get negative values.

-- 
Peter Geoghegan

Commits

Fix and clarify function comment on LogicalTapeSetCreate.
- 95f62b16a334 15.2 landed
- 236f1ea84c4c 16.0 landed
Refactor LogicalTapeSet/LogicalTape interface.
- c4649cce39a4 15.0 landed
Replace polyphase merge algorithm with a simple balanced k-way merge.
- 65014000b351 15.0 landed
logtape.c: do not preallocate for tapes when sorting
- 075896496394 14.0 cited