[KinoSearch] kinosearch 0.20 usable?

[Gert Brinkmann]

Marvin Humphrey wrote:

> It's frequent enough to be unacceptable: 1 in 20, 1 in 30.  What happens is
> that when the Highlighter trims an excerpt, the keyword that made the excerpt
> relevant gets trimmed sometimes, yielding an excerpt that is not obviously
> relevant.

This reminds me of a bug in the current Solr 1.3 that we are using. 
There the problem is that prefix-queries like "house*" do not return a 
highlight snippet. This was the reason why I did look for another 
solution like kinosearch. But as we have found a workaround for this bug 
we are trying to continue with Solr first. I keep kinosearch in my mind 
to probably give it a try sometimes later if at least 0.20 is released.

> I've put a ton of effort into modularizing the OO design and the file format
> so that supporting back compat will be easier going forward.  There will be a
> big push sometime soon to write up a draft Apache Lucy File Format

Is kinosearch using lucy somehow? Is the lucy format different to the 
lucene format? Isn't there already a lucy/lucene file format 
specification as this project does exist for longer time?

> Specification version 0.1, which KS will conform to.  If you'd like to
> contribute by reviewing that document, I'd appreciate it; the better a job we
> do, the sooner we'll "get the engine stable".

Well, this sounds to me as Linus Torvalds asked me to review a linux 
kernel documentation. What type of documentation will this be? Isn't it 
highly covering internal IR-theoretical things (that I do not know much 
about)?

Gert