Description
When a document doesn't have the field, we fill with zero. when a segment doesn't have the field, we also fill with zero.
however, when the analyzer doesn't return any terms for the field, we still call similarity.computeNorm(0)... with the default similarity this encodes infinity... -1
in such a case, it doesnt really matter what the norm is, since it has no terms. But its more efficient for e.g. compression if we consistently use zero.