a bug in fhv: first word from next field is appended in multiValue fields

[waziqi89]

This happens when fetch fragments using NRTFastVectorHighLighter(fhv). To reproduce this:

using fhv to highlight a multivalue field the fragment is at the end of the field the fragment is from a non-top-matched field the non-top-matched field is stored in front of the top-matched field

This is a bug from lucene-core, and not fixed in the latest version. (no one reports it)

The bug is caused by the BaseFragmentBuilder. When processing a multivalue field with the discreteMultiValue toggle on (do not create fragments across the field), the discreteMultiValueHighlighting will be called to reprocess the fragInfos with respect to the multivalue field. However, when creating the new fragInfos, it doesn't subtract 1 from fragEnd, ending up the fragment actually ends at the first char from the next field. later on, when creating the fragment string, a boundary scanner will be used to avoid chopping the word at the end of the fragment, so it will continue expanding the end till the end of first word from the next field.

This isn't a concern for the top field, because it is always processed first, plus a coincident of a buffer design and a out-of-boundary fallback mechanism.

[waziqi89]

created a fix PR in Lucene-core https://github.com/apache/lucene/issues/12221