Bruce Momjian <maillist@candle.pha.pa.us> writes:
>>>> That is fixed now.
>>
>> Is it? postgres.c treats an all-whitespace input as an empty query,
>> but if you pass it a comment and nothing else it will cycle the parser/
>> planner/executor, and I'm not sure every phase of that process behaves
>> reasonably on empty input. Also, that path will not produce the
>> "empty query" response code that you get from all-whitespace input.
>> I *think* libpq doesn't depend on that anymore, but other frontend
>> libraries might...
> postgres -D /u/pg/data test
> POSTGRES backend interactive interface
> $Revision: 1.130 $ $Date: 1999/09/29 16:06:10 $
backend> -- test
backend>
> Is that what you mean?
OK, so the parser/planner/executor can cope with dummy input. That's
good. There's still the problem of returning an 'empty query' response
to the frontend. I think you'd probably need to hack up postgres.c
so that when the querytree list produced by the parser is NIL, the
IsEmptyQuery flag gets set --- this could be done instead of, rather
than in addition to, the current test for an all-whitespace input
buffer.
regards, tom lane