Bruce Momjian <pgman@candle.pha.pa.us> writes:
> No, I don't remember hearing this discussed and I don't think most
> people would want libpq spilling to disk by default.
Far more useful would be some sort of streaming API to let the
application process the rows as they arrive, or at least fetch the rows
in small batches (the V3 protocol supports the latter even without any
explicit use of a cursor). I'm not sure if this can be bolted onto the
existing libpq framework reasonably, but that's the direction I'd prefer
to go in.
regards, tom lane