model-writtenrun 5400 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund opened the thread by arguing that manually arranging catalog fields around AIX alignment rules was unsustainable, proposing either proper alignment support or dropping AIX in #1.
- Peter Geoghegan questioned AIX's continuing value, while Tom Lane cited portability bugs it had exposed and favored retaining platforms while maintainers remained willing to support them; Andres emphasized the costs imposed on other developers in #2, #3, #6, and #7.
- Thomas Munro distinguished AIX's current hardware from discontinued HP-UX hardware in #9, and Tom Lane supported dropping HP-UX and Itanium while retaining PA-RISC's unusual spinlocks as useful portability coverage in #10 and #13.
- Thomas initially proposed removing HP-UX, aCC, Itanium, and PA-RISC together in #14, but Tom's working NetBSD/PA-RISC installation prompted the revised patch in #21 to preserve PA-RISC and simplify its GCC spinlock code.
- The AIX discussion also explored requiring eight-byte NameData sizes, introducing an explicitly aligned catalog floating-point type, or adding a distinct int64 alignment category; compiler support and tuple-deforming overhead remained concerns in #8, #16, #17, #18, #19, and #20.
- Andres requested separate HP-UX and Itanium removals and preservation of the PA-RISC instruction comments in #22, which Thomas implemented as the two-patch submission #24.
- Andres approved the split patches while finding a stale MSVC comment in #25, Peter Eisentraut requested preservation of the /dev/tty explanation in #28, and Tom reported successful NetBSD build and regression testing in #30.
- Robert Haas and Andres continued examining AIX alignment alternatives in #31, #32, and #33, with concern about both ensuring consistent typedef use and maintaining performance.
- Thomas landed the HP-UX component of #24 as 9db300ce6e with the /dev/tty comment preserved, and the Itanium component as 0ad5b48e58 with the stale MSVC comment cleaned up, announcing both in #34.
- Thomas then proposed removing unusable VAX support and a leftover Alpha reference while updating the CPU list in #34; Tom requested more precise variant wording and Greg Stark confirmed the VAX floating-point obstacle in #35 and #36.
- Robert's further AIX investigation found that redefining double alignment around MAXIMUM_ALIGNOF could enable the aligned-float8 approach but would break on-disk and pg_upgrade compatibility on AIX, leaving that work unresolved in #37.
- Thomas expanded the platform documentation to include DragonFlyBSD and illumos and refer generally to current OS versions in #38; after Tom's final wording suggestion in #39, this landed as 92d70b77eb, announced in #40.
- Tom next argued that untested M68K, M88K, M32R, and SuperH support should be removed unless someone supplied buildfarm coverage, and Robert endorsed removing unsupported claims in #41, #43, and #44.
- Thomas supplied the additional architecture removals and a separate historical-platform paragraph in #46, allowing that SuperH could return with maintenance and testing; Tom reviewed the submission in #47.
- The removal component landed unchanged as 718aa43a4e, and the historical paragraph landed with Tru64 UNIX replacing OSF as 14168d3c62; Thomas confirmed both in #48.
- At the end of #48, all submitted platform-cleanup work had landed and further historical-documentation contributions were welcome, while the original AIX alignment problem remained without an attached or committed fix in this thread.