model-writtenrun 8631 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut demonstrated that ALTER TABLE ... SET STORAGE left existing indexes with inconsistent storage settings that pg_dump could not reproduce, and proposed propagating changes to those indexes #1.
- Peter refined the patch to identify index columns through pg_index.indkey and added regression tests; Kuntal Ghosh verified the fix but found that shrinking test_decoding values removed coverage for tuples larger than the maximum heap tuple size #2 #3.
- Peter proposed supporting independent index storage through ALTER INDEX and pg_dump changes, while Kuntal warned that enforcing matching storage could break existing workloads and Alvaro Herrera favored supporting independent settings #4 #5 #6.
- Peter deferred ALTER INDEX because it required substantial pg_dump infrastructure and possible psql changes, instead preserving the decoding test through direct catalog manipulation while retaining consistent storage propagation #8.
- Peter committed the final patch unchanged as 501e41dd3c and confirmed backpatches to PostgreSQL 12 and 11; older branches were left unchanged because their differently structured catalog code made backpatching too costly, and independent index-storage DDL remained deferred #12 #8.