Greek search does not return highlighted results

[hcayless]

See http://dlibdev.nyu.edu/navigator/portal/ddbdp-results.psml?queryterm=strathg&boolean=THEN&queryterm=&slop=10&boolean=THEN&queryterm=&slop=10&offset=0&beta=on&place=&pubSeries=&pubVol=&apis=&after=&afterEra=CE&before=&beforeEra=CE&query=query(sub(beta(%22strathg%22)%2CIA))&submitted=X

Highlights only appear for exact matches on "strathg"

[jsosin]

Compare:
http://dlibdev.nyu.edu/navigator/portal/ddbdp-results.psml?queryterm=strathg&boolean=THEN&queryterm=&slop=10&boolean=THEN&queryterm=&slop=10&offset=0&beta=on&place=&pubSeries=&pubVol=&apis=&after=&afterEra=CE&before=&beforeEra=CE&query=query%28sub%28beta%28%22strathg%22%29%2CIA%29%29&submitted=X

and
http://apptest.cul.columbia.edu:8082/navigator/portal/ddbdp-results.psml?queryterm=strathg&boolean=THEN&queryterm=&slop=10&boolean=THEN&queryterm=&slop=10&offset=0&beta=on&place=&pubSeries=&pubVol=&apis=&after=&afterEra=CE&before=&beforeEra=CE&query=query%28sub%28beta%28%22strathg%22%29%2CIA%29%29&submitted=X

[hcayless]

Found the problem: In the refactoring, the JAR produced by source:/pn-ddbdp-indexers was named after the project name. Presumably, the original jar name was something like ddbdp-indexers.jar. source:/pn/trunk/pn-ddbdp-indexers/src/main/java/org/apache/lucene/search/highlight contains classes with identical names to some of those in the Lucene highlighter package. Because of the name change, the lucene jar was loaded first, and therefore the classes with the same name in the Lucene jar "won". The modified ones were never loaded.

Changing the jar name to something alphanumerically prior to lucene-highlighter-2.4.0.jar causes the application to work properly. Will investigate refactoring so that this kind of thing doesn't happen again, but quick fix (read: hack) would be to change the name of the generated jar.

[hcayless]

Tried several avenues to route around this, without much luck. The WeightedSpanTermExtractor? was modified to call rewrite on the query in certain circumstances. Calling rewrite before it is passed in results in more highlights, but not complete success. In the end, I moved the SpanScorer? and WeightedSpanTermExtractor? classes out of the way (by prefixing 'PN' to the names) and called those classes instead. This removes the classloading precedence problem, but it does mean special attention will have to be payed during future Lucene upgrades. r1748.