model-writtenrun 5589 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Feb - May 2015 · 14 entries
- Alvaro Herrera opened the thread with a 37-patch series covering CREATE, ALTER, and other utility commands, using event-trigger capture, JSON representations, and SQL reconstruction; he proposed replaying regression commands and comparing database dumps to detect missing deparser support #1.
- Andres Freund supported landing independent prerequisites, and Alvaro reported pushing object-identity fixes while identifying the difficulty of recovering transformed ALTER COLUMN TYPE USING expressions #2 #6.
- Stephen Frost's review prompted ObjectAddress return values instead of separate object and attribute identifiers, while the event-trigger prerequisites gained explicit documentation that GRANT, COMMENT, and SECURITY LABEL support applies to database-local objects #7 #9 #12 #13.
- Andres Freund supplied extensive infrastructure and deparser review plus fix-up patches, then revised function SET/RESET handling to use parse-tree information so CREATE OR REPLACE would retain the intended semantics; Alvaro folded these changes into the next full series #16 #19 #26.
- Stephen Frost advocated reusable catalog-based CREATE reconstruction, but Andres Freund distinguished that feature from replaying executed commands; Alvaro proposed retaining the current event-trigger interfaces while leaving room for future catalog-only callers, which Stephen accepted #17 #21 #23 #27.
- Alvaro separated and expanded the ObjectAddress prerequisites, Stephen reviewed them, and Alvaro reported committing them together except for ALTER TABLE, which still needed revision #28 #29 #31.
- The remaining series consolidated command patches, centralized ordinary command collection, and expanded supported commands; Stephen reviewed the result, and Alvaro subsequently adopted ObjectAddress results for ALTER TABLE subcommands and added replay-and-dump regression testing #33 #34 #35.
- Alvaro reported landing the text-search secondary-object and ALTER TABLE prerequisites, while Ryan Pedela and Robert Haas argued against prohibiting DDL inside event triggers because useful applications depended on it #37 #36 #38.
- In April, Alvaro proposed shipping command capture independently and moving JSON deparsing into an extension for further development, initially connecting that extension through CommandDeparse_hook #40 #41.
- After David Steele questioned how unavailable JSON should be represented, Alvaro replaced the single hook with a pg_ddl_command pointer type that multiple extension functions could consume independently #42 #43.
- David Steele tested the new interface with pg_audit and supported committing the core patches; Amit Kapila reviewed state handling and collection details, and Dimitri Fontaine endorsed the extensible representation design #44 #45 #46.
- Alvaro narrowed the proposed commit to command capture and a test module, deferred supporting APIs with no core callers, and allowed collection of inherited-table operations so consumers could decide which commands to ignore #47.
- The last pre-commit patch added documentation and broader tests, and David Steele reported successful review and testing shortly before the commit #50 #51.
- Alvaro committed b488c580ae from #50, adding pg_ddl_command I/O rejection functions, catalog corrections, and further documentation and tests; the commit shipped command capture without the deparser, and Alvaro requested stronger future-command test coverage #52.
Jul - Aug 2015 · 4 entries
- In July, Oleksandr Shulgin revived the uncommitted extension-support patch, rebased it, and began moving its helpers into the extension after discovering that the earlier supporting APIs had never landed #53.
- Oleksandr's testing exposed a typmod formatting bug and operator-resolution failures during view deparsing; Jim Nasby suggested explicitly qualifying the operator, and Alvaro confirmed that this fixed the reported resolution failure #54 #55 #56.
- Further testing revealed replay errors caused by lost original index names and repeated ALTER TABLE subcommands; Alvaro suggested filtering inherited operations in the deparser and requested separate patches for independent changes #59 #60 #61 #62.
- Oleksandr split the remaining work into five patches and added IndexStmt.conname to preserve index identity, while leaving repeated ALTER COLUMN TYPE output and hierarchical representation of internally generated commands unresolved #63.
Apr 2022 · 2 entries
- After nearly seven years, Ajin Cherian revived the remaining work for renewed interest in DDL logical replication and posted a rebase of the first support patch; Peter Eisentraut objected that the submission lacked sufficient explanation, documentation, and tests #64 #66.
- The thread ends with Ajin's rebase of three older patches, omitting the testing and text-search mapping patches because the associated extension was absent; he acknowledged retaining the old commit messages, and no further landing of this remaining deparser work is reported #67 #68.