Re: Abbreviated keys for Numeric

From: Peter Geoghegan <pg(at)heroku(dot)com>
To: Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk>
Cc: Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>, Pg Hackers <pgsql-hackers(at)postgresql(dot)org>, Gavin Flower <GavinFlower(at)archidevsys(dot)co(dot)nz>
Subject: Re: Abbreviated keys for Numeric
Date: 2015-03-24 04:24:18
Message-ID: CAM3SWZT-xHa2Er1-dEC9NzU+D==kNj8+sYYbyoSdVckiq0VNdw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Sun, Mar 22, 2015 at 1:01 PM, Andrew Gierth
<andrew(at)tao11(dot)riddles(dot)org(dot)uk> wrote:
> So here's a version 3 patch:

By the way, there was another bug in this that I forgot to point out,
but removed, here:

+ if (nss->estimating)
+ {
+ uint32 tmp = (uint32)result;
+ addHyperLogLog(&nss->abbr_card, hash_uint32(tmp));
+ }

And here:

+ if (nss->estimating)
+ {
+ uint32 tmp = ((uint32)result) ^ (uint32)((uint64) result >> 32);
+ addHyperLogLog(&nss->abbr_card, hash_uint32(tmp));
+ }

(I simply operate on the raw Datum when hashing for hyperLogLog, in a
similar manner to the text opclass, which is safe with 8 byte datums +
pass by value int64).

The problem here is that you're not calling hash_uint32() through
DatumGetUInt32() - hash_uint32() returns Datum. And yes, that matters
- see commit 2c22afaa.

You also need to indent this code correctly.

--
Peter Geoghegan

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Ashutosh Bapat 2015-03-24 04:44:20 Re: Display of multi-target-table Modify plan nodes in EXPLAIN
Previous Message Alvaro Herrera 2015-03-24 02:46:21 Re: Table-level log_autovacuum_min_duration