model-writtenrun 6264 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jul - Aug 2021 ยท 14 entries
- Dave Cramer reported that upgrading from PostgreSQL 10.5 to 13.3 preserved pg_stat_statements version 1.5 and its older function definitions, whereas a fresh installation used version 1.8 #1; he subsequently explained that a dump and restore failed while restoring function permissions #5.
- David G. Johnston and Tom Lane explained that retaining extension versions during pg_upgrade was intentional; both supported clearer documentation, and Tom proposed reporting extensions with available updates instead of forcing upgrades #4 #7.
- Dave supplied an inline documentation draft suggesting extension recreation and outlined detecting version differences #8, but discussion established that ALTER EXTENSION UPDATE already provided the update operation and that replacement libraries must match the new server #30 #31.
- Bruce Momjian supplied the first attached documentation patch #35, then refined it through reviews by David and Dave to distinguish installing compatible shared libraries from transferring existing SQL definitions #36 #37 #38 #51.
- Dave objected that 'recreated' implied new definitions and that the update guidance was too weak #54, prompting Bruce to use 'copied' and 'should be upgraded' #56.
- Alvaro Herrera pointed out that modules need not be extensions, extensions need not contain shared libraries, and updates should be described as available rather than mandatory; he also proposed a separate full-text search step #64, which Bruce incorporated into the final documentation patch #73.
- Jan Wieck raised the problem of extension installation scripts producing different objects on different server versions #48, and Julien Rouhaud supplied powa-archivist as a concrete example whose compatibility wrapper could survive an upgrade unnecessarily and harm performance #65 #85.
- Tom proposed the concrete implementation of inspecting pg_available_extensions in the new cluster and producing ALTER EXTENSION UPDATE commands for installed versions differing from the default #79; he also clarified that ordinary dump and restore installs the target server's default extension version #81.
- Julien explained that updating extensions cannot repair server-dependent definitions when the extension version is already current #91; Dave suggested a post-upgrade API, but Julien preferred handling this niche case within the extension #92 #93.
- Bruce posted a separate code patch that checks every database, warns about available extension updates, and writes update_extensions.sql for the administrator to execute #94, leaving automatic execution out of the implementation.
- Jan proposed encoding the server version in extension version numbers #95, but Tom rejected imposing a universal numbering scheme and recommended shipping a new default extension version with an ordinary update script #97; Jan agreed that server-dependent object definitions should use distinct extension versions #98.
- Bruce committed the final documentation patch #73 unchanged as 5090d709f1 and announced its application to all supported versions #99.
- Bruce committed the warning and script-generation patch #94 unchanged as e462856a7a, announcing backpatches through PostgreSQL 9.6 #100.
- Justin Pryzby then spotted a copied comment that incorrectly described the extension query as finding hash indexes #101; Bruce corrected it in 9e51cc87fd and confirmed the fix in the thread's final message #102, leaving all attached work landed while broader automatic upgrades and a post-upgrade API remained unimplemented.