[LUCENE-7626] IndexWriter shouldn't accept broken offsets

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 7.0
• Lucene Fields: New
• Assignee: Michael McCandless
• Reporter: Michael McCandless
• Created: 11/Jan/17 10:14
• Updated: 28/Aug/22 15:08
• Resolved: 13/Jan/17 22:46

Description

I think we should do this in 7.0 (not 6.x).

Long ago we stopped accepting broken offsets (where the start offset
for a token is before the start offset of the last token) in postings
(LUCENE-4127), but we are still lenient with term vectors.

I think we should also check for term vectors: this would let users
know that their analysis chain is producing offsets that cannot be
used properly at search time.

Issue Links

• blocks: LUCENE-7963 Useless getAttribute() in DefaultIndexingChain causes performance drop