On 12/18/17 00:49, Michael Paquier wrote:
> I have a problem with this approach which is basically something
> similar to what I complained in the thread about typed and partition
> tables for identity columns
> (https://www.postgresql.org/message-id/20171023074458.1473.25799@wrigleys.postgresql.org).
> In my opinion, this ALTER TABLE handling should be done by treating
> identity columns a way similar to default expressions in
> transformColumnDefinition(), by storing the FuncExpr node at parsing
> time instead of storing the information needed to rebuild it when
> executing the query. In short the mapping should get closer to what
> default does with nextval or serial.
The serial case works because it stores the sequence *name* in the
default value in the catalog. That doesn't work because for the
identity case we don't store the expression in the catalog. The
proposed patch works by storing the sequence *name* in the internal
structures so that it can be used in place of the stored default value.
So I think this approach is pretty consistent.
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services