"David G. Johnston" <david.g.johns...@gmail.com> writes:
> On Thursday, October 7, 2021, vignesh C <vignes...@gmail.com> wrote:
>> Should tables be allowed to create in "information_schema" schema, if
>> yes should the tables/publications be dumped while dumping database
>> contents?
> I presume you have to be superuser to do this. If so, this would seem to
> fit under the “we don’t stop you, but you shouldn’t” advice that we apply
> throughout the system, like in say modifying stuff in pg_catalog.
> Information_schema is an internal schema attached to an static for a given
> release.
It is (supposed to be) possible for a superuser to drop information_schema
post-initdb and then recreate it by sourcing the information_schema.sql
file. In fact, I seem to recall that we've recommended doing so in past
minor releases to correct errors in information_schema declarations.
So it's fairly hard to see how we could enforce prohibitions against
changing information_schema objects without breaking that use-case.
On the other hand, just because you did that doesn't mean that you want
information_schema to start showing up in your dumps. Quite the opposite
in fact, because then you'd have problems with trying to load the dump
into a newer PG version that might need different information_schema
contents.
So I agree: there's nothing to be done here, and the proposed scenario
is a case of "superusers should know better than to do that".
regards, tom lane
