Peter Eisentraut <peter_e@gmx.net> writes:
> Since these functions will primarily be used in building a sort of
> information schema and for querying system catalogs, we should use the
> approach that is or will be used there: character type values contain the
> table name already case-adjusted.
Weren't you just arguing that such cases could/should use the OID, not
the name at all? ISTM the name-based variants will primarily be used
for user-entered names, and in that case the user can reasonably expect
that a name will be interpreted the same way as if he'd written it out
in a query.
The nextval approach is ugly, I'll grant you, but it's also functional.
We got complaints about nextval before we put that in; we get lots
fewer now.
regards, tom lane