model-writtenrun 200 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ke reported that binary tsvector input accepted duplicate lexemes and position zero, demonstrating silent changes or restore failures after text dump/restore and suggesting normalization or rejection in #1.
- Tom Lane proposed requiring strictly increasing lexemes because merging duplicates would also require merging position lists using a different internal representation, and agreed that accepting position zero was an oversight in #2.
- Tom Lane supplied #3, replacing lexeme re-sorting with rejection and checking positions against an initial zero; the patch retained elog errors rather than adopting the report’s suggested SQLSTATE change.
- The thread ended with the patch submission in #3, without subsequent review or testing reports; its proposed policy was to tighten validation only on master to avoid disrupting existing workflows in minor releases.
- The patch subsequently landed as 10cc5aa96a with identical code and minor commit-description edits, completing both validation fixes without backports.