Philip Warner wrote:
> At 03:37 AM 7/11/2003, Peter Eisentraut wrote:
>
>> It would
>> already help a lot if we just added the table name, or something that
>> was
>> until a short time before the action believed to be the table name, or
>> even only the table OID, before (or after) the $1.
>
>
> Can we allow/bypass the pg_* restriction, and call it
> pg_<table-oid>_<n>, and for pedants like me, add a DB setting that
> says 'enforce unique constraints' ala the spec to avoid manually
> created constraints being non-unique?
>
> Alternatively, I would be happy *not* to enforce constraint name
> uniqueness (and break the spec) so long as we also break the spec and
> add table OID (or something else) to the information schema (table
> name would be OK so long as renaming the table won't break anything).
The first seems impractical for reasons given by Tom. Why provide an
option for behaviour we fear could deadlock etc.?
Regarding the second option, I don't understand what virtue there is in
breaking the spec more, rather than embedding the table name in the
constraint name.
(If we do that, if the table is renamed my instinct would be to rename
constraints in the renamed table with autogenerated names, but I haven't
looked into it).
cheers
andrew