Robert Haas <robertmhaas@gmail.com> writes:
> On Tue, Apr 14, 2009 at 10:27 AM, Kevin Grittner
>> Yeah, if it would be too heavy to add a timestamp column or two to
>> pg_class and maybe one or two others, why is it better to add a whole
>> new table to maintain in parallel -- with it's own primary key,
>> foreign keys (or similar integrity enforcement mechanism), etc.
> Making pg_class and pg_proc tables larger hurts run-time performance,
> potentially. Making a separate table only slows down DDL operations,
> which are much less frequent.
And even more to the point, adding columns to the core system tables
means you pay the performance cost *even when not using the feature*.
We normally expect that inessential features should avoid making a
performance impact on those who have no use for them.
regards, tom lane