Re: [Tracker] cvs is a bit hard to test.

[Jamie McCracken <jamiemcc blueyonder co uk>]

Anders Aagaard wrote:

> > In the meantime can you use the libextractor that comes with gentoo. 
> > During compile, if tracker sees that libextractor is already installed 
> > then it wont build the inlined version by default.
>
> heh, yeah, that's the problem, that one doesn't work either ;)

please report the bugs to them then.

> Nice, but does that currently support full text searches?  One of the 
> things I'd absolutely love to do with tracker is to search using 
> something like "type:video funny", to find funny video's (type video can 
> use mime types to find all the extensions).  Or even "match:*.avi funny".

should be possible in the near future. for now you would have to use rdf 
query to do that.


> All I found on the subject was this:
> http://www.sqlite.org/cvstrac/wiki/wiki?p=FullTextIndex


We will be using qdbm to create an inverted word index.

This is just a file based hash table that takes a word and returns 
DocId's and associated scores which we can stick in a temporary table in 
sqlite for further processing.

Im confident we will get a faster and more lightweight tracker by doing 
this along with bonus features like stemming and score weighting on 
different types of metadata (none of which is supported in mysql).

If in the future, sqlite can deliver equivalent functionality without 
sacrificing speed/memory then we can migrate towards that. But I need a 
scalable solution today that can handle tons of emails as well as files


--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/