Re: Implementation of global temporary tables?

Andres Freund <andres@anarazel.de> writes:
> On 2015-07-15 16:24:52 +0100, Simon Riggs wrote:
>> It may be possible to do this, though I'm sure there's a wrinkle somewhere.
>> But there doesn't seem to be a need to overload the main feature request
>> with additional requirements. Doing that is just scope creep that prevents
>> us getting features out. Nice, simple patches from newer developers. Later
>> tuning and tweaking from more expert community members.

> I think that's generally a fair point. But here we're discussing to add
> a fair amount of wrinkles with the copy approach. The fact alone that
> the oid is different will have some ugly consequences.

> So we add complexity, just to shift it into different places later? I'm
> not sure that's a good idea.

With all due respect, there are features that are beyond the abilities of
some "newer developers", and reducing the scope isn't a good way to fix
that.  It just leaves a bigger mess to be cleaned up later.

I think Andres' idea of a per-backend filenode mapping table might work.
The existing relfilenode mapper solves a somewhat related problem, namely
how do you replace the filenode for shared system catalogs whose pg_class
entries can't be changed.
        regards, tom lane