ATM, the title is searched by using a bleve MatchPhraseQuery, which fails on searches for iss as in issue and so on.
The MPQ is able to find camel cased words, though, but you will have to input the whole phrase, e.g. foo as in fooBar.
Also, the fuzziness is determined wrong by modules/indexer/internal/bleve/util.go/GuessFuzzinessByKeyword.
bleve only supports a fuzziness of (0,1,2) and some sources state 3. Testing this, however, showed that a
higher fuzziness will return too many results and that a value of 1 seems to be the sweet spot.
The function also return values < 1 for keywords with a length < 4, so this should be addressed, too.
Also, chinese and other languages need to be checked, as the function also bails out on unicode characters
with a code point > 128, returning 0, basically limiting fuzzy searches to more or less standard ASCII.
DB Indexer
The db indexer uses LIKE '%...%' so this should not be changed.
Elasticsearch / Meilisearch
Elasticsearch and meilisearch need to be addressed, too.
Rationale
The rationale for limiting fuzzy search to title only is that one much easier remembers the title than what is inside a wall of text such as the content of comments or the issue itself. Also, the fuzzy search is slower than standard MPQ
silkentrance
commented
5 days ago
Feature Description
Bleve
ATM, the title is searched by using a bleve
MatchPhraseQuery, which fails on searches forissas inissueand so on. The MPQ is able to find camel cased words, though, but you will have to input the whole phrase, e.g.fooas infooBar.Also, the fuzziness is determined wrong by modules/indexer/internal/bleve/util.go/GuessFuzzinessByKeyword. bleve only supports a fuzziness of (0,1,2) and some sources state 3. Testing this, however, showed that a higher fuzziness will return too many results and that a value of 1 seems to be the sweet spot. The function also return values < 1 for keywords with a length < 4, so this should be addressed, too. Also, chinese and other languages need to be checked, as the function also bails out on unicode characters with a code point > 128, returning 0, basically limiting fuzzy searches to more or less standard ASCII.
DB Indexer
The db indexer uses LIKE '%...%' so this should not be changed.
Elasticsearch / Meilisearch
Elasticsearch and meilisearch need to be addressed, too.
Rationale
The rationale for limiting fuzzy search to title only is that one much easier remembers the title than what is inside a wall of text such as the content of comments or the issue itself. Also, the fuzzy search is slower than standard MPQ
Resources
Original issues
Pull Request