model-writtenrun 6058 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr - Nov 2019 · 8 entries
- Kyotaro Horiguchi proposed replacing XLogReader's page-read callback with an XLREAD_NEED_DATA return protocol, initially using continuation macros to preserve the existing control flow while acknowledging the added complexity #1.
- Antonin Houska reviewed the macro machinery, naming and buffer contract, then supplied a comment fix; Kyotaro incorporated the feedback and moved continuation state into function-local static variables #2 #4 #5.
- Andres Freund questioned whether the size of the change justified its benefits, and Kyotaro defended direct caller-controlled reads while remaining open to withdrawing the proposal #7 #9.
- Heikki Linnakangas supported the new interface but rejected the continuation macros and static locals, supplying an on-top rewrite with an explicit four-state switch, caller-owned buffers and caller-side read-error handling #10.
- Heikki subsequently introduced XLogBeginRead, integrated record seeking into the state machine and reorganized header validation, but Kyotaro missed that submission and continued from Heikki's first rewrite #12 #25.
- Kyotaro consolidated that earlier rewrite into a smaller series, reduced state variables, restored early record-header validation and isolated buffer allocation as an optional patch; the next revision fixed timeline following and added removal of duplicate globals #13 #14.
- Further revisions repaired missing first-page validation when crossing segments and removed duplicated XLogNeedData logic #16 #17.
- Alvaro Herrera caught a rebase that inadvertently undid recent pg_waldump changes, and Kyotaro corrected it along with comments and formatting #20 #22.
Jan 2020 · 3 entries
- Heikki revived the separate XLogBeginRead idea as an independently useful preparatory change, receiving approval from Alvaro and a detailed check from Kyotaro #23 #24 #26.
- Heikki committed that preparation as 38a957316d from #23, additionally fixing XLogFindNextRecord to leave the reader positioned at the discovered record rather than restore its previous pointers #28.
- Craig Ringer warned about difficult timeline-switching and cross-segment cases, while Heikki distinguished the small preparation from the riskier callback conversion; Kyotaro then rebased the remaining series onto the landed API #27 #29 #30.
Mar - Jul 2020 · 2 entries
- After further upstream callback changes, Kyotaro retained only a cleanup callback in XLogReaderState and passed segment callbacks explicitly to WALRead; Craig argued for pluggability, while Kyotaro explained that tightly coupled caller state and error handling motivated the separation #34 #35 #36.
- Takashi Menjo found a WAL_DEBUG build failure, inconsistent buffer allocation and a mistaken initializer change; Kyotaro fixed these and replaced the optional allocation change with configurable WAL block-size support, while questioning its immediate usefulness #38 #42.
Mar - Apr 2021 · 4 entries
- Following another long interval and a rebase, Thomas Munro tested and reformatted the core three patches, reported compiler warnings and left the optional block-size extension unreviewed and outside his submission #43 #46.
- Thomas and Andres requested separate requested and available byte counts and an input-setting helper; Kyotaro implemented these and also removed the remaining XLogFindNextRecord callback, while broader decoder renaming was deferred #47 #48 #52 #53.
- Thomas found a record-positioning bug and remaining warnings, then squashed the series and supplied a fixup that corrected these issues and renamed the input helper to XLogReaderSetInputData #53 #54.
- Thomas committed 323cbe7c7d from #54, with only additional comment correction and removal of debugging changes, and announced the landing in #55.
Jun - Jul 2021 · 2 entries
- Kyotaro later reported that the callback-removal commit had been reverted, expressed uncertainty about reproducing a possible defect on the affected platform and returned the work to review #57.
- Kyotaro resubmitted the consolidated patch rebased onto current master, after which Thomas directed him to an earlier problem report that he had missed #59 #60 #61.
Sep - Oct 2021 · 1 entry
- The final submission returns to the earlier five-patch series, completes XLogFindNextRecord's state machine, adds a page-skipping test and introduces XLogTerminateRead while adapting to overwritten continuation records; the callback removal remains proposed after its revert, with no second landing shown in the thread #62.