It looks like [[Lucene]] doesn’t like searching through indices that have been built in lots of little chunks. I get proper results if I build an index of 300 blog posts like this:

- open index
- for post in posts:
    - add post to index
- close index

… but if I do it like this, it seems to stop indexing them after the first couple of hundred:

- for post in posts:
    - open index
    - add post to index
    - close index

Update: It looks like the reason for this behaviour was that I was passing a true value as the create parameter to IndexWriter’s constructor in the test. I’m seeing the same behaviour in my application code, though, which sets it to false. Odd.

Update 2 (2003-06-12): Figured the application out now (finally)! The problem was a combination of the use of different analyzers in indexing and search and a broken custom query object that was blowing away some of the hits.

Downhill uses the [[Blogging Ecosystem]] data to find the shortest path between two given weblogs. Nice.