The attached patch implements the pg_attribute.attstattarget scheme
discussed on -hackers today. A value of "-1" in this column indicates
that ANALYZE should use the compiled-in default value. This allows
pg_dump to record and restore any changes made by the DBA to this
value using ALTER TABLE ALTER COLUMN SET STATISTICS.
I made another small behavioral change: IMHO, silently changing
stat target values we deem inappropriate (< 0, > 1000) without
warning the user is a bad idea. I changed the < 0 case to be a
elog(ERROR) and added an elog(WARNING) for the > 1000, and
documented the upper bound.
Unless anyone sees any problems, please apply.
Neil Conway <neilconway(at)rogers(dot)com>
PGP Key ID: DB3C29FC
pgsql-patches by date
|Next:||From: Bruce Momjian||Date: 2002-07-20 04:58:35|
|Subject: Re: Fix for regression caused by heap tuple header changes|
|Previous:||From: Tom Lane||Date: 2002-07-20 03:26:58|
|Subject: Re: lock listing |