On 2 March 2012 20:45, Robert Haas <robertmhaas@gmail.com> wrote:
> I decided to investigate the possible virtues of allowing "text" to
> use the sortsupport infrastructure, since strings are something people
> often want to sort.
I should mention up-front that I agree with the idea that it is worth
optimising text sorting because it is a very common thing to have to
do, and therefore the standard for inclusion ought to be lower. I
don't intend to talk about tapesort though - that isn't really fair,
not least because I have some serious doubts about the quality of our
implementation. Furthermore, I think that it is logical that doing
things like resolving collations occur within a preparatory function
in advance of sorting, rather than redundantly doing that for each and
every comparison.
Why have you made the reusable buffer managed by sortsupport
TEXTBUFLEN-aligned? The existing rationale for that constant (whose
value is 1024) does not seem to carry forward here:
* This should be large enough that most strings will be fit, but small* enough that we feel comfortable putting it on
thestack.
ISTM it would be on average worth the hit of having to repalloc a few
more times for larger strings by making that buffer much smaller
initially, and doubling its size each time that proved insufficient,
rather than increasing its size to the smallest possible
TEXTBUFLEN-aligned size that you can get away with for the immediately
subsequent memcpy. Realistically, any database I've ever worked with
would probably be able to fit a large majority of its text strings
into 16 chars of memory - you yourself said that sorting toasted text
isn't at all common.
--
Peter Geoghegan http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training and Services