model-writtenrun 4937 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Sep - Dec 2022 ยท 17 entries
- Simon Riggs proposed a Transaction Management chapter covering identifiers, locking, subtransactions, and prepared transactions, with related command and function documentation, while explicitly requesting content review before markup polishing #1.
- Erik Rijkers corrected typos, Justin Pryzby supplied detailed wording feedback, Alvaro Herrera requested an XID epoch explanation, and Robert Treat clarified subtransaction rollback and prepared-transaction wording; Simon incorporated these reviews through #9 #2 #4 #6 #8.
- After a month, Bruce Momjian substantially edited the proposal, separated two-phase transaction usage into the WAL chapter, added documentation cross-references, and announced an intention to backpatch all supported releases #10.
- Simon supplied separate RELEASE SAVEPOINT and ROLLBACK TO SAVEPOINT clarifications, including a nested-savepoint error example and an explanation that rollback starts a new subtransaction #13 #16.
- Alvaro supported the chapter and savepoint example while suggesting clearer terminology, and Bruce merged the chapter and command changes with revised explanations of subcommits and parent identifiers #14 #20.
- Robert warned that describing savepoints as destroyed could imply loss of their data changes, prompting Bruce to rewrite the RELEASE SAVEPOINT description; Simon then objected to 'destroy' and proposed explicit subcommit wording that Robert refined #21 #22 #23 #24.
- Following confusion over the focused attachment, Simon supplied a complete cumulative patch incorporating the savepoint revisions so reviewers could examine the whole proposal #25 #26 #28.
- Bruce replaced user-facing subcommit terminology with release-and-merge semantics, arguing that command reference pages should describe behavior rather than implementation details; Simon preferred matching code terminology, leaving a substantive disagreement over documentation scope #31 #39 #45.
- Laurenz Albe's extensive review exposed an incorrect table-lock wait description, questioned the WAL placement of prepared-transaction internals, and requested clearer identifier explanations and a more accurate chapter title; Bruce adopted Transaction Processing and moved the two-phase section back into the new chapter #29 #30 #32 #40.
- Further discussion clarified that ROLLBACK AND CHAIN creates a transaction that is not aborted, explained the I/O cost of excess subtransactions, and standardized terminology markup #34 #35 #36 #41 #46.
- Laurenz and Alvaro refined savepoint wording and the distinction between explicitly started and implicit transactions; Bruce incorporated these changes and Erik's plural corrections in the final polishing rounds #47 #48 #49 #50 #51 #52.
- Laurenz reported no remaining problems apart from a blank line, and Justin demonstrated that commands in an aborted transaction produce an error rather than the documented warning; Bruce removed that claim in the final attachment #53 #54 #56.
- Bruce landed the final attachment as 66bc9d2d3e, crediting Simon and Laurenz as authors, and announced backpatches through PostgreSQL 11; the landing omitted the proposed COMMIT chained-transaction paragraph without an explanation in the thread #56 #57.
- Peter Eisentraut requested reversing the backpatches because major chapter numbering should remain stable within a release, while Bruce defended making useful documentation immediately available and David G. Johnston supported keeping changes of this scope on the development branch #58 #60 #61 #64.
- Alvaro argued that the new chapter should receive further editing before release, and Tom Lane found a copied mvcc.sgml filename in its header; both supported removing the chapter from stable branches #65 #66.
- Bruce responded with 4dd00b042f, which fixes the header on master while its stable-branch counterparts revert the additions, and confirmed that the chapter remained on master #67.
- The thread ended with the documentation retained for the development release and all stable-branch backpatches reversed; Alvaro clarified that documentation fixes and smaller additions can still justify backpatching #67 #68.