Re: add "WITH OIDS" to CREATE TABLE AS

Neil Conway <neilc@samurai.com> writes:
> No equivalent functionality has been added to SELECT INTO: there
> isn't a convenient syntax for it that I could see, and in any case
> CREATE TABLE AS has always offered a superset of the functionality of
> SELECT INTO. Therefore, I don't view this as a problem.

I agree on that.

> The implementation is a tad messy (it would be nice if CREATE TABLE AS
> were a distinct node, to avoid needing to clutter up SelectStmt
> further).

Yeah, I have been wanting for awhile to redesign the parse
representation of CREATE TABLE AS/SELECT INTO.  It's not obvious exactly
what to do though.

> I also needed to add an additional production to avoid a
> shift/reduce conflict in the parser (see the XXX note in the patch
> itself).

This is a fairly standard way of avoiding conflicts --- looks fine to
me.

One thing that is *not* fine is something that I see had snuck past me
in the previous WITH OIDS patch.  It is not okay for gram.y or scan.l
to be looking at GUC variables --- that creates synchronization issues.
(Consider a single querystring containing a SET followed by something
that depends on the SET value.  The whole string will be parsed before
the SET is applied.)  The references to default_with_oids have to be
postponed to analyze.c's processing.  Compare the way that inhOpt has
an INH_DEFAULT setting --- you probably need the same kind of solution
for the WITH OIDS options.

            regards, tom lane