On Fri, May 1, 2009 at 9:48 AM, Marvin Humphrey <[email protected]> wrote: > I hope we can achieve the same ends in Lucy with at most 2 classes: > "SortCollector" and "TopDocCollector".
Hi Marvin --- I haven't had time to read the recent messages closely, but I took a few minutes to try to catch up with this one. The small-scale logic makes sense to me (rationale for jump tables etc) but I'm not quite getting the bigger picture. Could you contextualize a little? My instinct is that you're putting a little too much knowledge of the internals here into SortCollector. Maybe this is merely naming, though. Is there a difference between 'Score' and 'Float'? Between Ord8 and Int8? And will there be a way to just fall back on String comparisons, bypassing the Sort cache? I'd like to view the cache as an optimization, like adding an index to database column[1]. But it feels like you are making it a requirement. As always, I'm worried that maybe I won't be able to abuse your interface in ways it was not designed. :) In particular, I wonder about integrating hits from multiple networked machines and being able to iteratively return all the sorted results for big databases. Nathan Kurz [email protected] [1] It does feel like you're redeveloping a lot of SQLite here!
