Re: BUG #12752: Regression in ALTER TABLE RENAME COLUMN

Tom Lane wrote:
> gianni(dot)ciolli(at)2ndquadrant(dot)it writes:
> > It appears that the bug discussed here has regressed since 9.1:
>
> > http://www.postgresql.org/message-id/200110121753.f9CHrnl10542@candle.pha.pa.us
>
> We gave up trying to make index column names match the underlying table in
> 9.0; so no currently supported PG release does what you wish, and it's
> unlikely that any future one will either. (This is called out as an
> incompatibility in the 9.0 release notes, FWIW.)

I think it's misleading that we name index columns following table
columns in the first place .. IMO it would be better to use some dummy
name, like the ones we assign to dropped columns. That would be far
less tempting for tool writers.

> > Context information: the bug breaks Londiste, which uses
> > pg_attribute.attname to match columns on the primary key index with the
> > corrisponding columns on the table.
> > I found the bug after a customer reported Londiste issues on a table whose
> > primary key had been renamed.
>
> Sorry, but that's a Londiste bug not a Postgres bug.

The fix most likely involves using indkey to join table columns to index
columns in Londiste's internal queries.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services