Do you have any alternatives to Vlad's index idea for keeping VARBINARY
values distinguishable in a multi-segmented index? If not, and all
other ideas will continue to produce wrong LIMIT results in the future,
then I think this is what we should do long term.
But if you start on this now we will have our priorities mixed up. The
only thing this buys us is the ability to sort (1) VARBINARY (2)
strings of zeros correctly in a (3) multisegmented index, which only
reveals itself in a (4) LIMIT query. That is a very narrow
justification for such a huge change to Falcon.
Both changes seem to require a new index version, but Vlad's new
multisegmented index structure may even require an index3Page.cpp. I
The argument for doing it now is that we are pre-beta and now is the
time to change the index structure.
Any comments? Now or later?
-------- Original Message --------
Subject: Index Discussion
Date: Wed, 18 Feb 2009 17:16:56 -0500
From: Jim Starkey <jstarkey@stripped>
To: Falcon <dbg-falcon-private-ext@stripped>
It's been a great discussion. Almost everyone has been right at least
once and wrong at least once. We've got a neat new idea to chew on
(thanks, Vlad, sorry, Kevin).
There are a number of lessons here:
1. There are still lots of good ideas out there
2. New good ideas take some force and persistence to get an honest
3. Curmudgeons can learn new tricks.
Thanks, guys (and Ann).
President, NimbusDB, Inc.