Howard Rogers <hjr@diznix.com> writes:
> OK, Tom: I did actually account for the number of rows difference
> before I posted, though I accept I didn't show you that. So here goes:
> ...
> Both queries return zero rows. One takes an awful lot longer than the
> other. The only difference between them is that one searches for
> 'ftx1' and the other searches for 'ftx0'.
Well, this still doesn't tell us anything about what I think the
critical point is, namely how many actual matches there are for
ftx1 versus ftx0. Could we see counts for *just* those words without
the other conditions?
> So, I would still like to know if this performance difference when
> encountering alpha-numeric "words" is dictionary-related,
AFAIK there is no significant difference between treatment of pure alpha
and mixed alphanumeric "words", at least not once you get past
to_tsquery. I'm still expecting this is just a matter of how many index
entries match. It's barely possible that you've got a dictionary
configuration that makes the to_tsquery() function itself a lot slower
in the alphanumeric case, but that should affect ftx1 and ftx0 equally.
regards, tom lane