Once upon a time (Thu, 25 Sep 2003 01:28:29 -0400)
Tom Lane <tgl@sss.pgh.pa.us> uttered something amazingly similar to:
> Robert Creager <Robert_Creager@LogicalChaos.org> writes:
> >> Seems like the correct behavior to me. When CONSTRAINT is given, the
> >> name is not optional AFAICS.
>
> > True, but the second CREATE DOMAIN, which is a valid form, fails when
> > preceded by the invalid form. This was the error I was referring to.
> > Not the fact that I didn't know how to use domains ;-) If the
> > transaction is bad after the first invalid CREATE DOMAIN, the error
> > should say so, rather than waiting for a second valid statement to
> > come along, shouldn't it?
>
> Uh, which part of "queries ignored until end of transaction block"
> didn't you understand?
I understand every word of it. Why doesn't that error come up after the first
CREATE DOMAIN error, since I was in a transaction. Then it would of been
perfectly clear that the transaction I was in needed to be rolled back. Rather
that error doesn't occur until the second valid CREATE DOMAIN is executed.
I was not aware of what Stephan indicated: "All errors are considered
unrecoverable ones by PostgreSQL so after any error the transaction is in an
effectively unusable state and should be rolled back". I know better now. It
still seams that the "queries ignored until end of transaction block" should of
been thrown after the first invalid CREATE DOMAIN. That's all.
>
> > begin;
> > create domain test as integer constraint check( value > 0 );
> > -- ERROR: syntax error at or near "check" at character 42
> > create domain test as integer check( value > 0 );
> > -- ERROR: current transaction is aborted, queries ignored until end
> > of transaction block
Cheers,
Rob
--
07:16:42 up 54 days, 23:48, 5 users, load average: 2.17, 2.06, 2.01