Perhaps we should use a factory method on SolrIndexSearcher (that way we can wrap the reader and pass that to the SolrIndexSearcher constructor?)
-Yonik On Mon, May 25, 2009 at 1:24 PM, Mark Miller <[email protected]> wrote: > Yonik Seeley wrote: >> >> On Mon, May 25, 2009 at 12:35 PM, Peter Wolanin >> <[email protected]> wrote: >> >>> >>> Playing with this last night, the error seems to occur for me only >>> with an index created using a solr build from a couple weeks ago >>> that's opened with a new solr build. The error goes away if I issue a >>> delete query for *:* and re-index the content. >>> >> >> That's really strange... that really shouldn't be the case. Perhaps >> reopen() is acting a little different with the older index? >> Please keep that index around if possible so we can get to the bottom of >> this! >> >> Mark - could it be the changes from delegation to inheritance for >> SolrIndexSearcher? >> super(reader) is called before we get a chance to to wrap the reader >> and make it a SolrIndexReader... >> >> -Yonik >> http://www.lucidimagination.com >> > > It could be. Good point - SolrIndexSearcher gets the plain reader and it > used to get SolrIndexReader. My fault, because I did most of the thinking on > that pre SolrIndexReader patch. So I guess, Lucene would feed the Solr > DocSet a regular reader instead of SolrIndexReader - no good. But, what I > still don't understand is how that plays into Peter's comment above and my > inability to easily replicate the problem. I'll keep digging, but it sounds > like I may just revert back to delegation or override getIndexReader to > return the SolrIndexReader - if thats done, we should prob deprecate > getReader. We'd obviously like to fully understand whats happening in any > case though. Until I can tickle the error, I can only guess myself. > > - Mark > > > -- > - Mark > > http://www.lucidimagination.com > > > >
