model-writtenrun 8528 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bertrand Drouvot separated the heap-rescan bug from streaming SampleScan work in thread #253556, posting a fix and test that recreate the ReadStream when strategy use changes instead of adding a strategy-update API #1.
- Nazir Bilal Yavuz supported the simpler approach for backpatching and advocated regression coverage; Bertrand measured differences below 0.15 microseconds per rescan and concluded performance did not favor either approach strongly #2 #5.
- shihao zhong questioned whether disabling the strategy was reachable in core, and Bertrand explained how an extension could trigger it through table_rescan_set_params(), which Shihao accepted #3 #4 #6.
- Joao Rito verified that the original test failed without the fix and passed with it, including worker and synchronous I/O, while explicitly leaving strategy disabling and other scan types untested #7.
- Melanie Plageman revised the patch to put stream reset, destruction, and creation inside initscan(), avoiding confusing ordering and possible duplicate resets, and simplified the test to use a cursor while retaining a dedicated TAP cluster #8.
- Bertrand reviewed the revision and requested a lifecycle-comment update, then suggested shared I/O statistics for the PostgreSQL 17 backpatch; Nazir endorsed the revision and comment correction #9 #10 #11.
- Melanie committed the revised submission as 556c025db9, adding the requested lifecycle comment and strengthening the test's before-and-after statistics checks, then confirmed that the fix had landed on master and the back branches, completing the thread #12.