As I try to implement this I am beginning to see some serious issues with the heuristic cited in this document. The problem is sub-expression Cursors in the OR Cursor can be on anything: a logical expression or on an attribute Index for a simple leaf assertion. When on an index, the results of the Cursor are sorted by the key (value of the attribute) of the index and not by ID.
The children of an OR Cursor can be on any attribute index. This means the Cursor results are not sorted by id as supposed by this heuristic. Instead they're sorted by attribute value. This will lead to inconsistent results.
This heuristic is useless :(. What a waste! It would have been so efficient.
On Thu, Mar 27, 2008 at 3:59 AM, Emmanuel Lecharny <email@example.com> wrote:
I have read and checked the wiki page. I have nothing to add
Cool. BTW I documented that sweet optimization you cited a week ago for rapid shorting of child expression evaluations. I think this will be a significant optimization.
FYI, I'm slowly coming back but I still have some side things to deal with.
No worries. BTW I'm thinking we can just do a quick ADS release with only the top handful of issues that are critical for this 1.5.2 feature release.