Still no joy isolating, increasingly convinced the problem is in my code. The overseeks are always in the .frq files, and always modest, e.g. 10% past the end, not 1000%. Always from the ConjunctionScorer -- I can search for every word in the index in isolation without hitting the problem.
Any of that ring bells? Stu On Mon, Jan 7, 2013 at 11:25 AM, Stuart Halloway <stuart.hallo...@gmail.com>wrote: > Beyond it, and it seems to read after that too. I am re-running the > (large, cumbersome) test case I have in hand with the infoStream spout > turned on and will be back with more detail in a bit... > > > On Mon, Jan 7, 2013 at 11:08 AM, Michael McCandless < > luc...@mikemccandless.com> wrote: > >> Hmm good question! >> >> Are you seeing ConjunctionScorer seek to exactly the end-of-file (ie, >> == length), or beyond it? >> >> Mike McCandless >> >> http://blog.mikemccandless.com >> >> On Mon, Jan 7, 2013 at 10:57 AM, Stuart Halloway >> <stuart.hallo...@gmail.com> wrote: >> > I have implemented a custom Lucene directory, and I think I am seeing >> > Lucene's ConjunctionScorer seek off the end of IndexInput, which I have >> > implemented to throw an exception. One example stack trace leading to >> this >> > is: >> > >> > at >> org.apache.lucene.index.SegmentTermDocs.skipTo(SegmentTermDocs.java:197) >> > at org.apache.lucene.search.TermScorer.advance(TermScorer.java:156) >> > at >> > >> org.apache.lucene.search.ConjunctionScorer.doNext(ConjunctionScorer.java:99) >> > at >> > >> org.apache.lucene.search.ConjunctionScorer.<init>(ConjunctionScorer.java:72) >> > at >> > >> org.apache.lucene.search.ConjunctionScorer.<init>(ConjunctionScorer.java:33) >> > >> > Upon reading Lucene's own source, I see some evidence that seeking off >> the >> > end should be legal, e.g. RAMInputStream has a boolean enforceEOF flag >> which >> > is set to false by calls to seek. >> > >> > I am working to make a small reproducible case, but in the meantime >> hoping >> > that somebody has more details about how seek is supposed to be used. >> > >> > On the 3.x line btw, and testing with multiple versions from 3.2 to 3.6. >> > >> > Thanks, >> > Stu >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org >> For additional commands, e-mail: dev-h...@lucene.apache.org >> >> >