Re: Re: ¿¿¿past chunk end??? - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Re: ¿¿¿past chunk end???
Date
Msg-id 10773.1162448368@sss.pgh.pa.us
Whole thread Raw
In response to Re: ¿¿¿past chunk end???  ("luis garcia" <ldgarc@gmail.com>)
List pgsql-hackers
"luis garcia" <ldgarc@gmail.com> writes:
> Well about the changes I have made, they are working just fine. All the
> structures length have been updated and as I said we added 7 new fields
> to pg_class structure and there was no problem. The problem here is with
> relfrequency only.

pg_class is notoriously tricky to add fields to --- there are more
undocumented dependencies than you might think.  If you check the
archives, there was a thread only a week or so back with someone who
was one or two edits short of getting such a change to work.

But what I'm wondering in your case is whether your code thinks it can
set relfrequency and/or the preceding field relfrequencytype to null.
You can't use a C struct to address any fields beyond the first
possibly-nullable field, because the constant offsets computed by the
compiler won't work.  Also, initdb makes some assumptions derived from
this fact to decide which system-catalog columns to mark NOT NULL.
If "\d pg_class" shows not-null column markings that are at variance
with what you intend, you have a problem.

            regards, tom lane


pgsql-hackers by date:

Previous
From: Oleg Bartunov
Date:
Subject: Re: [GENERAL] Index greater than 8k
Next
From: "Simon Riggs"
Date:
Subject: Re: [PATCHES] Writing WAL for relcache invalidation:pg_internal.init