Re: [HACKERS] Last thoughts about LONG

On Sat, 11 Dec 1999, Jan Wieck wrote:

> Well, the implementation details. I really would like some
> little comments to verify it's really complete before
> starting.

Before I start the nagging, please be aware that I'm not as smart as I
think I am. Long datatypes of some sort are clearly necessary -- more
power to you.

> - A new field "rellongrelid" type Oid is added to pg_class.
> It contains the Oid of the long-value relation or the
> invalid Oid for those who have no LONG attributes.

I have a mixed feeling about all these "sparse" fields everywhere. Doing
it completely formally, this seems to be a one-to-many relation, so you
should put the referencing field into the pg_long table or whatever
structure you use, pointing the other way around. This is probably slower,
but it's cleaner. As I mentioned earlier, this whole arrangement will
(hopefully) not be needed for all too long, and then we wouldn't want to
be stuck with it.

> - At CREATE TABLE, a long value relation named
> "_LONG<tablename>" is created for those tables who need it.

Please don't forget, this would require changes to pg_dump and psql. Also,
the COPY command might not be able to get away without changes, either.

In general, it wouldn't surprise me if some sections of the code would go
nuts about the news of tuples longer than BLCKSZ coming along. (Where
"nuts" is either 'truncation' or 'segfault'.)

I guess what I'm really saying is that I'd be totally in awe of you if you
could get all of this (and RI) done by Feb 1st. Good luck.

--
Peter Eisentraut Sernanders vaeg 10:115
peter_e(at)gmx(dot)net 75262 Uppsala
http://yi.org/peter-e/ Sweden