Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
> > On 10/24/2010 08:12 PM, Tom Lane wrote:
> >> This shows that the bitmapset optimization really is quite effective,
> >> at least for cases where all the enum labels are sorted by OID after
> >> all. That motivated me to change the bitmapset setup code to what's
> >> attached. This is potentially a little slower at initializing the
> >> cache, but it makes up for that by still marking most enum members
> >> as sorted even when a few out-of-order members have been inserted.
>
> > That's nice. It's a tradeoff though. Bumping up the cost of setting up
> > the cache won't have much effect on a creating a large index, but could
> > affect to performance of retail comparisons significantly. But this is
> > probably worth it. You'd have to work hard to create the perverse case
> > that could result in seriously worse cache setup cost.
>
> Well, notice that I moved the caching into typcache.c, rather than
> having it be associated with query startup. So unless you're actively
> frobbing the enum definition, that's going to be paid only once per
> session.
Thanks for modifying pg_upgrade so it works with this new format. The
changes look good and cleaner than what I had to do for 9.0.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ It's impossible for everything to be true. +