model-writtenrun 2605 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed eight components converting macros to static inline functions in dependency order, leaving by-value assignment macros unchanged and postponing tuple-header macros that accepted multiple tuple types #1.
- Amul Sul endorsed the series and suggested simpler conditional returns, declaration spacing and a switch-based fetch_att(), while Alvaro Herrera proposed local PageHeader variables to avoid repeated casts #2 #3; Peter incorporated the latter changes but retained the conditional-return style in the revision #4.
- Peter Geoghegan supported the conversion because inline functions improve maintainability and make defensive assertions easier without multiple-evaluation hazards #5.
- The revised block.h component #4 landed with redundant-parenthesis cleanup as d63d957e33.
- The bufpage.h component #4 landed as 2cd2569c72 with more accessors converted, unused validity macros removed and additional callers corrected, while pruning helpers stayed macros because of their access/transam.h dependency.
- The itemptr.h component #4 landed unchanged as aeb767ca0b, and the bufmgr.h component landed with an additional RelationGetNumberOfBlocks() conversion as 9c727360bc.
- The WAL filename helpers from #4 landed with comment and expression cleanup as 507ba16b28.
- The tupmacs.h component #4 landed partially as f58d7073b7, retaining fetchatt() as a macro, dropping its associated GiST renaming and adding frontend guards; the itup.h component then landed with a frontend guard for index_getattr() as 14a8bd9827.
- Peter later submitted the previously unsent RelationCloseSmgr() conversion and questioned whether it should instead be an external function #6; Amul favored inlining and moving the assertion outside the conditional #7, which landed as 3edc71ec04 and was confirmed by Peter #8.
- After more than two years, Peter returned to the deferred tuple-header work with const prerequisites, straightforward htup_details.h and itup.h conversions, and a separate experimental C11 generic-selection patch that he explicitly did not intend to commit #9.
- The const prerequisites landed unchanged as b15b8c5cf8, and the tuple-accessor conversions landed with a parameter rename as 34694ec888, both from #9; Peter confirmed the landing while excluding the generic-selection experiment #10.
- Maxim Orlov then proposed forcing inlining for 55 htup_details.h functions because they occur in hot paths, while leaving the larger fastgetattr() and heap_getattr() functions alone for debugging #11.
- Peter's final response reported that -Winline analysis showed no relevant function-size problem and argued against forcing inlining without evidence #12; the always-inline proposal remained uncommitted in the supplied record, the generic-selection experiment was deliberately excluded #9, and the disposition of the original pgstat.h component #4 is not established by the supplied commits or discussion.