Bruce Momjian wrote:
> Kevin Brown wrote:
> > The two approaches aren't necessarily mutually exclusive (though SQL99
> > compliance on constraint names would obviously make it unnecessary to
> > specify a tablename along with a constraint name), so I see little
> > problem here. But the current arrangement is obviously untenable,
> > because it allows you to create a situation (multiple constraints by
> > the same name) that you can't reasonably extricate yourself from.
>
> Well, it seems if we want to continue to allow the same constraint name
> to be used by different tables in the same schema, we have to print the
> tablename in the error message. Would someone actually be looking for a
> standards-compliant error string? We have already extended the standard
> --- either we revert that, or we have to go the entire way and print the
> table name.
If PG were configurable in terms of how it manages constraint names,
then it would depend on how the DBA had the database configured. With it
configured to disallow name collisions, it would obviously be unnecessary
to report the table name, though I still think it would be useful (if
only because it gives a little extra context to work with). But if it's
configured to allow name collisions, then it doesn't make sense not to
print the table name in an error message, because that's the only way to
guarantee that the DBA can identify which constraint is being referred to.
The problem as things stand now is that even if we printed the table name
involved, the DBA is placed in a difficult position if the constraint in
question isn't uniquely named -- which is the only case where printing
the table name would really matter. That's because he can't actually
refer to the constraint in any unique way short of playing with the
system tables; he'd have to rename the constraint first before being
able to really do something with it (is this even possible for him to
do without manipulating system tables? Is there an ALTER CONSTRAINT?).
--
Kevin Brown kevin@sysexperts.com