"Thomas G. Lockhart" <lockhart@alumni.caltech.edu> writes:
> The *only case* I've noticed so far which does better in v6.3.2 than
> "v6.4today" (not yet v6.4alpha :) is the one involving OIDs:
> regression=> explain select * from tenk1 where oid = 3000;
> Index Scan on tenk1 (cost=2.05 size=1 width=100)
Actually, I'm aware of another one: comparisons using "> constant"
or "< constant" seem more likely to use an index in 6.3.2 than they
do in the sources I have. I have examples involving both datetime
and int4 columns where "where x = constant" will be implemented by
index scan, but "where x > constant" will not. Explicit casts of
the righthand side make no difference. And it works fine in 6.3.2.
I'm a couple of weeks behind the CVS tree, so I was going to wait
until I'd confirmed it with up-to-the-minute sources before complaining.
But if you're proceeding on the assumption that the "oid = integer" case
is the only thing that's broken, you may be misled.
The fact that casting doesn't affect this makes me think it is a
different problem than the must-cast-to-get-an-index-scan cases
we've discussed so far.
regards, tom lane
