Bruce Momjian <pgman@candle.pha.pa.us> writes:
> This doesn't look good. If we throw a WARNING, why do we not insert
> anything into pg_description. Seems we should throw an error, or do the
> insert with a warning.
Throwing an error breaks existing pg_dump files. Doing the insertion is
simply wrong: it will allow the former breakage to be perpetuated
forward by dump/reload. Thus the current behavior is an unfortunate but
necessary compromise ... at least until we have better support for
comments on databases.
regards, tom lane
