model-writtenrun 509 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nathan Bossart proposed retiring unsafe, obsolete PQfn() while retaining its exported symbol as an error-returning function; prepared-statement invalidation and a measured PQexecParams slowdown complicated replacing internal large-object fast-path calls #1.
- Christoph Berg found wrappers but no actual external callers and supported retaining the symbol with an error, while Jacob Champion supported retirement and keeping private PQnfn() temporarily but questioned the shortcomings of the recommended alternatives #2 #3 #4.
- Jelte Fennema-Nio supplied an unvetted optimization using one-shot plans for unnamed statements, but Nathan's test improved only from 0.457 to 0.444 seconds against 0.319 seconds for the existing implementation; Nathan also posted his separate prepared-statement and PQexecParams experiments #6 #7.
- Nathan explored server notifications for deallocated prepared statements and posted a protocol 3.3 implementation with libpq callbacks, prepared large-object queries, and a PQexecParams fallback for older servers #5 #9 #10.
- Jacob argued for a protocol extension instead of a version bump and favored retaining PQnfn() while the broader design matured; Nathan accepted those directions and moved deallocation notifications to another thread, explaining why pinning, built-in statements, and query inspection seemed unsuitable #11 #12 #13 #14.
- Nathan then narrowed the series to making PQfn() fail and switching its remaining internal callers to PQnfn(), alongside separate byte-order and argument-setup cleanups in fe-lobj.c #16.
- Ilmari requested preserving an explanation in the obsolete-feature appendix, and Nathan incorporated that documentation while retaining the narrowed implementation #17 #18.
- After a two-month gap, Nathan committed the unchanged byte-order cleanup component as 5593aca0c6, rebased the remaining retirement patch, and abandoned the argument helpers because they did not seem worth the effort #19.
- Nathan refined documentation and comments in the final attachment, then landed PQfn() retirement with further light wording and comment edits as 40af05bb61 and announced completion #20 #21.
- At the thread's close, retirement and byte-order cleanup had landed, while replacing large-object fast-path calls remained future work, deallocation notifications continued separately, and neither the argument-helper cleanup nor Jelte's experimental optimization landed through this thread #14 #19 #21 40af05bb61.