The attached patch removes GUC's USERLIMIT variable category, changing
all the affected variables to be plain SUSET, as per recent discussion.
I also modified postgres.c so that variable settings coming from the
client connection request packet (eg, from PGOPTIONS on the client side)
are processed only after we have determined whether the user is a
superuser. This allows a superuser to set SUSET options from PGOPTIONS,
which has never worked before.
I consider this to be a lot cleaner and more flexible than what we have
now. Comments?
regards, tom lane