On Mon, Jan 09, 2006 at 09:54:44PM -0500, Tom Lane wrote:
> Extrapolating from the observation that the heuristics don't work well
> on your data to the conclusion that they don't work for anybody is not
> good logic. Replacing that code with a flat 50% is not going to happen
> (or if it does, I'll be sure to send the mob of unhappy users waving
> torches and pitchforks to your door not mine ;-)).
I'm not convinced but of course I cannot exclude that some people may be
depending on this very heuristic. But I consider this being
bug-compatible, I've an hard time to be convinced that such heuristic
isn't going to bite other people like it did with me.
> I did just think of something we could improve though. The pattern
> selectivity code doesn't make any use of the statistics about "most
> common values". For a constant pattern, we could actually apply the
> pattern test with each common value and derive answers that are exact
> for the portion of the population represented by the most-common-values
> list. If the MCV list covers a large fraction of the population then
> this would be a big leg up in accuracy. Dunno if that applies to your
> particular case or not, but it seems worth doing ...
Fixing this with proper stats would be great indeed. What would be the
most common value for the kernel_version? You can see samples of the
kernel_version here http://klive.cpushare.com/2.6.15/ . That's the
string that is being searched against both PREEMPT and SMP.
BTW, I also run a LIKE '%% SMP %%' a NOT LIKE '%% SMP %%' but that runs
fine, probably because as you said in the first email PREEMPT is long
but SMP is short.
Thanks!