Zeugswetter Andreas SB <ZeugswetterA@wien.spardat.at> writes:
>> But you don't really need to look at the index (if it even exists
>> at the time you do the ANALYZE). The extent to which the data is
>> ordered in the table is a property of the table, not the index.
> Think compound, ascending, descending and functional index.
> The (let's call it) cluster statistic for estimating indexscan cost can only
> be deduced from the index itself (for all but the simplest one column btree).
If you want to write code that handles those cases, go right ahead ;-).
I think it's sufficient to look at the first column of a multicolumn
index for cluster-order estimation --- remember all these numbers are
pretty crude anyway. We have no such thing as a "descending index";
and I'm not going to worry about clustering estimation for functional
indexes.
regards, tom lane