On Thu, 4 Oct 2007, Tom Lane wrote:
> There's some limited smarts in there about deciding that leading columns
> of an index don't matter to the sort ordering if they're constrained to
> just one value by the query. But it doesn't catch the case you need,
> which is that columns of an ORDER BY request are no-ops when they're
> constrained to just one value.
Oh, no, that explains it perfectly, because that's precisely the case I
have - I dropped the columns from the ordering, but not the where clause.
Thanks, now I understand the current behavior.