I think you're right, we should mention this.
How about changing that bullet like this?:
* All filters now return a DocIdSet instead of java.util.BitSet,
making filters more efficient and flexible. Searching with a
Filter is now more efficient: now the filter is applied to a
document before scoring is done.
Mike
Nadav Har'El wrote:
On Tue, Sep 30, 2008, Michael McCandless wrote about "Re: draft 2.4
announcement":
* All filters now return a DocIdSet instead of java.util.BitSet,
making filters more efficient and flexible.
Hi Michael, I was looking at how filtering works, and I noticed
there is,
I believe, another very big change.
Previously, if I remember correctly, when a Filter was added to a
query,
it worked on the HitCollector level, with each matching document being
checked against the filter. The new code is more efficient when the
filter
only contains a small number of "on" documents, because now the
scorer's
skipTo is actually used.
Isn't this a change worth mentioning, or am I just imagining things?
(If I
am, sorry.. Interestingly, I found in CHANGES.txt a mention of
LUCENE-1169
which is about fixing a bug in that new code, but I don't see any
mention
of that new code being written).
Thanks,
Nadav.
--
Nadav Har'El | Sunday, Oct 5 2008, 6
Tishri 5769
IBM Haifa Research Lab
|-----------------------------------------
|Anyone who quotes me in their
sig is an
http://nadav.harel.org.il |idiot. -- Rusty Russell's sig.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
