Comment 8 for bug 260140

I have a branch that implements what I think is a "good enough" fix. This branch does an initial query to count the number of BranchRevisions for a product or project. If this is less than some magic number, we do two queryies, one to get the revision ids, and another to get the revisions. This will force the use of indices. If there are more than this magic number, then postgresql "should" use the fast query plan.

bzr+ssh://devpad.canonical.com/code/tim/launchpad/revision-feed-speed