* When compiled NOT with unicode, there are a few problems
* Query: +test^10 returns strange results
* Indexing many documents returns different files than java version. newsegmentname func is called to many times?
* Boost query uses float, which is not accurate (2.3 for example could return 2.9999992)
* Still some memory leaks
* Memory leak when adding 2 fields of the same name
