Jeff Davis wrote:
> All I mean is that the second argument to
> COPY should produce/consume bytes and not records. I'm not discussing
> the internal implementation at all, only semantics.
>
> In other words, STDIN is not a source of records, it's a source of
> bytes; and likewise for STDOUT.
>
In the context of the read case, I'm not as sure it's so black and
white. While the current situation does map better to a function that
produces a stream of bytes, that's not necessarily the optimal approach
for all situations. It's easy to imagine a function intended for
accelerating bulk loading that is internally going to produce a stream
of already processed records. A good example would be a function that
is actually reading from another database system for the purpose of
converting its data into PostgreSQL. If those were then loaded by a
fairly direct path, that would happen at a much higher rate than if one
had to convert those back into a stream of bytes with delimiters and
then re-parse.
I think there's a very valid use-case for both approaches. Maybe it
just turns into an option, so you can get a faster loading path record
at a time or just produce a stream characters, depending on what your
data source maps to better. Something like this:
COPY target FROM FUNCTION foo() WITH RECORDS;
COPY target FROM FUNCTION foo() WITH BYTES;
Would seem to cover both situations. I'd think that the WITH BYTES
situation would just do some basic parsing and then pass the result
through the same basic code path as WITH RECORDS, so having both
available shouldn't increase the size of the implementation that much.
--
Greg Smith 2ndQuadrant Baltimore, MD
PostgreSQL Training, Services and Support
greg@2ndQuadrant.com www.2ndQuadrant.com