Re: pg_dump emits ALTER TABLE ONLY partitioned_table

Tom, Robert,

* Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > I would appreciate help from other contributors and committers on this
> > open item; pg_dump is not my strong point. In the absence of such
> > help, I will do my best with it. I will set aside time this week to
> > study this and send another update no later than Thursday.
>
> The proposed patch seems rather ad-hoc, and I think that it is working
> around a backend behavior that might be broken.

Yeah, I'm not a big fan of the proposed patch either.

> While I admit that I've not been paying close attention to the whole
> table partitioning business, I wonder whether we have any clearly written
> down specification about (a) how much partition member tables are allowed
> to deviate schema-wise from their parent, and (b) how DDL semantics on
> partitioned tables differ from DDL semantics for traditional inheritance.
> Obviously those are closely related questions. But the fact that this
> bug exists at all shows that there's been some lack of clarity on (b),
> and so I wonder whether we have any clarity on (a) either.

Right. I thought I had understood that partition child tables really
can't deviate from the parent table in terms of columns or constraints
but might be allowed to differ with regard to indexes (?).

I'll try looking into this also, I should be able to spend some time on
it this week.

Thanks!

Stephen