Uwe, This is beautiful! Especially for conversion from Trie to Point fields is going to be extremely handy. I am going to have to check this out further. Thank you for the tip!
Rahul On Mon, Jan 17, 2022 at 10:23 AM Uwe Schindler <u...@thetaphi.de> wrote: > By the way > > Hi, one thing that always works to "forcefully" upgrade without > reindexing. You > > just merge the old index into a completely new index not by coping > files, but by > > sending their SegmentReaders to addIndex, stripping all metadata from > them > > with some trick: > > > https://lucene.apache.org/core/8_11_0/core/org/apache/lucene/index/SlowCo > > decReaderWrapper.html in combination with > > < > https://lucene.apache.org/core/8_11_0/core/org/apache/lucene/index/Index > > Writer.html#addIndexes-org.apache.lucene.index.CodecReader...-> > > > > One way to do this is the following: > > - Open old index using DirectoryReader.open(): reader = > > DirectoryReader.open(...old directory...) > > - Create a new Index with IndexWriter writer: writer = new > IndedxWriter(...new > > directory...) > > - Call > > > writer.addIndexes(reader.leaves().stream().map(IndexReaderContext::reader). > > map(SlowCodecReaderWrapper::wrap).toArray(CodecReader[]::new)); > > This trick also works if you want to transform indexes. I wrote some code > that on the-fly rewrites old NumericField to PointField. The trick is to > add another FilterLeafReader (before wrapping with SlowCodecReaderWrapper), > that detects legacy numeric fields, removes them fromm metadata and feeds > them as new stream of flat BKD points enumerated by the TermsEnum (which > works because order is same and hierarchy is generated by the receiving > IndexWriter) to a new field with PointField metadata. This is a bit hacky > but works great. > > Uwe > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org > For additional commands, e-mail: java-user-h...@lucene.apache.org > >
