[re-sent]

Andrew,

This is one of the cases where it's hard to improve something without making
something other worse :-) The new v2.0 cost estimation algorithm considers
indices on s.type_id and l.action_id worth using, whilst FB 1.5 ignored
them. And in this case 1.5 was more correct, unfortunately. I don't see any
easy way to fix this situation, so the only good solution for 2.0 would be
"hints" like:

s.type_id+0 = i.type_id
-- and
l.action_id+0 = a.id

that disable usage of the "bad" indices.

We'll try to fix/enhance the cost estimation rules in the next version, so
it would perform better in such cases.


Dmitry