On 14 June 2012 17:35, Robert Haas <robertmhaas@gmail.com> wrote:
> The problem with pre-detoasting to save comparison cycles is that you
> can now fit many, many fewer tuples in work_mem. There might be cases
> where it wins (for example, because the entire data set fits even
> after decompressing everything) but in most cases it seems like a
> loser.
If I had to guess, I'd say you're probably right about that -
optimising sorting toasted text doesn't seem like a terribly sensible
use of your time.
What about the strxfrm suggestion of Greg's? You might find that the
added benefit of being able to avail of a highly optimised strcmp()
tipped the balance in favour of that idea, beyond the simple fact that
there's only a linear number of what you might loosely call "strcoll_l
units of work" rather than as many as O(n ^ 2). Furthermore, I'd
speculate that if you were to interlace the strxfrm() calls with
copying each text string, somewhere like within a specialised
datumCopy(), that would make the approach more efficient still, as you
specify a location for the blob in the just-palloc()'d leading-key
private memory directly, rather than just using memcpy.
--
Peter Geoghegan http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training and Services