I've got a problem with a query running against Lucene 7.3 where the boolean AND is not being applied.
The fields involved are: 1. _missing_, which contains a token for each field missing in the document 2. phoneNationalNumberQueryNgrams, which contains a phone number. The index analyzer is keyword, and the query analyzer is a right edge ngram with a minimum length of 6 (There are other fields, but this is the minimal reproducer) I have in the index one document containing: _missing_ = phoneCountryCode phoneNationalNumberQueryNgrams = "987654321" These queries find the document as expected: 1. _missing_:phoneCountryCode 2. phoneNationalNumberQueryNgrams(987654321) 3. phoneNationalNumberQueryNgrams(0987654321) 4. _missing_:phoneCountryCode AND phoneNationalNumberQueryNgrams:(987654321) These queries fail to find the document as expected: 1. phoneNationalNumberQueryNgrams(87654321) 2. phoneNationalNumberQueryNgrams(54321) 3. _missing_:phoneCountryCode AND phoneNationalNumberQueryNgrams:(87654321) 4. _missing_:phoneCountryCode AND phoneNationalNumberQueryNgrams:(54321) However this query finds the document: (Note this is the boolean AND of one term that does match the document and one that does not) 1. _missing_:phoneCountryCode AND phoneNationalNumberQueryNgrams:(54321) In fact any combination of a field term that does match, and a phoneNationalNumberQueryNgrams term that is below the min ngram length appears to result in a match. At a guess, this looks like a term where the query analyzer produces no tokens results in that term being eliminated from the overall expression (i.e. it doesn't evaluate to FALSE in a boolean position). The workaround is to include another non-eliminatable term in the query: _missing_:phoneCountryCode AND (phoneNationalNumberQueryNgrams:(54321) OR FALSE) Is this an expected behavior, or a bug? Is there any downside to the workaround? Tim
