"Johann Zuschlag" <zuschlag@online.de> writes:
>> Okay ... are there any rows in pg_operator with OID 280346 or 280347 ?
> Yes, seems so. See the attachment. Again, the negator stuff
> never worked for numeric.
Looks like these are "shell" operator definitions left over from
commutator or negator forward references that were never satisfied.
pg_dump did the right thing to not dump them. I'd say that the backend
should never have accepted a shell operator def with an empty name,
though, which is what you seem to have at OID 280347.
Do you happen to have the exact command that you gave to create
operator 280343 (numeric_neq)? I think what this really boils down
to is insufficient error checking somewhere in CREATE OPERATOR.
regards, tom lane