On 2015-04-20 12:49:39 +0300, Heikki Linnakangas wrote:
> With the patch, pg_class.relam column references to the pg_seqam table for
> sequences, but pg_indexam for indexes. I believe it's the first instance
> where we reuse a "foreign key" column like that. It's not a real foreign
> key, of course - that wouldn't work with a real foreign key at all - but
> it's a bit strange. That makes me a bit uncomfortable. How do others feel
> about that?
Hm. I'd modeled it more as an extension of the 'relkind' column
mentally. I.e. it further specifies how exactly the relation is
behaving. Given that the field has been added to pg_class and not
pg_index, combined with it not having index in its name, makes me think
that it actually was intended to be used the way it's done in the patch.
It's not the first column that behaves that way btw, at least pg_depend
comes to mind.
Greetings,
Andres Freund