Index trigrams
moio opened this issue · comments
One Lucene limitation when it comes to fuzzy search is that the maximum edit distance is capped to 2 due to algorithmic reasons.
One way to circumvent this is to index trigrams instead of lines, thereby allowing some trigrams to be missing or be different in sequence, rather than searching for lines and allowing fuzzy matching with up to two edits.
This requires changing SourceAnalyzer to decompose text in trigrams and, crucially, rewrite queries so that they target trigrams.
Then performance has to be evaluated (time-wise and search-effectiveness-wise).