[Xapian-discuss] How to implement range searches?

[Olly Betts]

On Sun, Nov 16, 2008 at 01:26:12PM +0100, Torsten Foertsch wrote:
> http://www.xapian.org/docs/queryparser.html states: "The QueryParser can 
> be configured to support range-searching using document values."
> 
> How to do that using the Perl API?

Much the same as in C++, except you can't currently create your own
subclasses in Perl - you are limited to the standard ones:

http://xapian.org/docs/valueranges.html

> I want to be able to search for "a:10..15 AND b:qaa..raa", that is find 
> all documents where the value a is between 10 and 15 and b in the range 
> qaa, qab, ..., qba, qbb, ..., qzz, raa.
> 
> Is that possible?

Yes (intermediate values such as qaabcde will also match the latter).

> Do I understand it correctly that the range search is applied after the 
> actual search?

Not really.  The matching all happens together, but the matcher is
fairly smart about what order to do things in within that.

> Does it mean the query above would lead to a complete 
> scan over all documents since there is no search word in the query?

Yes, and this can be a bit slow for such cases in 1.0.x.

> Or is there an index also on the value fields of the documents?

SVN trunk's chert backend stores values differently (much like how we
store term postings) and is much quicker.

> Background: I want to store in the index 2 values. One is a string 
> handle that says what process has brought the document into the index 
> or updated it recently. The other is a UNIX timestamp.

Does a range search really make sense on the first?  If not, then
indexing as a boolean term may be more sensible.  You can check for one
of several values still.

Cheers,
    Olly