model-writtenrun 1892 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar - Apr 2024 · 5 entries
- Japin Li reported that PostgreSQL configuration on Illumos failed to find a working 64-bit integer type when CFLAGS contained -Werror in #1, and Tom Lane traced this to a harmless missing-prototype warning promoted to an error in #6.
- Andres Freund proposed warning about or temporarily filtering -Werror in #7, while Tom Lane preferred rejecting it in #11; the thread instead developed a way to eliminate the obsolete integer probes.
- Thomas Munro suggested relying on standardized integer facilities in #12, and Tom Lane supported implementing PostgreSQL's existing names over <stdint.h> while avoiding widespread naming churn in #13.
- Thomas Munro submitted the integer conversion and separate BeOS cleanup in #17, but Japin Li's Illumos testing exposed incorrect printf modifiers and differences in unsigned limit definitions in #18.
- After Peter Eisentraut questioned the need for custom format probing in #19, Thomas Munro switched to <inttypes.h> macros in #20; Japin Li confirmed the formatting fix in #23, and Peter explained the need to keep integer types, limits, and format macros consistent in #24.
Jul 2024 · 4 entries
- Heikki Linnakangas reviewed the series favorably while discussing format-string readability in #25, but Windows' nonstandard modifiers led Thomas Munro to restore an INT64_MODIFIER probe and add a tzcode conversion in #26.
- Tom Lane asked to defer the timezone changes until upstream synchronization and warned against exposing int_fastN_t through PostgreSQL APIs in #27; Thomas Munro agreed to wait in #28.
- Peter Eisentraut supplied detailed corrections for integer-size selection, ECPG, OID_MAX, and tzcode in #29, and advocated standard format macros because gettext can translate them portably, explaining the runtime mechanism in #31.
- Peter Eisentraut extracted the BeOS cleanup from #26 into a2d9a9b95a, adapting it to precede the integer conversion and including related ECPG cleanup, then announced the landing in #32.
Nov - Dec 2024 · 8 entries
- After the gap, Thomas Munro resumed the earlier <inttypes.h> approach in #33, incorporating review feedback and initially overriding Windows macros; after further configure cleanup in #34, he replaced those overrides with snprintf support for Windows modifiers in #35 because gettext expands the system's original macros.
- Peter Eisentraut approved the final integer conversion and identified a misplaced comment in #37; Thomas Munro landed #35 with that correction and small build-file cleanup as 962da900ac, announcing it in #38.
- The landing exposed Perl ABI failures and incorrect off_t sizes on older 32-bit systems, reported in #39 and investigated in #41; Tom Lane diagnosed postgres_ext.h including a system header before pg_config.h could define _FILE_OFFSET_BITS in #43.
- Although restoring the old generated header was initially considered, Tom Lane proposed changing inclusion order in #46, leading to Thomas Munro's repair and optional extra-header cleanup in #48.
- Tom Lane supplied clearer comments and rejected expanding postgres_ext.h's header footprint in #49; Thomas Munro incorporated those decisions in #50, which Tom approved in #53 and which landed unchanged as 71cb352904.
- Thomas Munro explained that Meson's command-line large-file definitions had masked the ordering problem in #52, and Tom Lane raised the remaining concern that extension builds could become ABI-incompatible without those flags in #54.
- Thomas Munro proposed a separate cleanup in #56 to expose int64_t directly in libpq, retaining a deprecated pg_int64 alias but moving it from postgres_ext.h into libpq-fe.h.
- Building on the gettext discussion and the inline demonstration in #35, Peter Eisentraut landed Thomas Munro's pg_checksums-only formatting trial as 15a79c7311, allowing translation-toolchain testing before broader conversion.
Mar - Apr 2025 · 4 entries
- Peter Eisentraut approved the libpq cleanup in #58, and Thomas Munro committed #56 unchanged apart from rebasing as 3c86223c99, announcing completion in #59.
- Nathan Bossart then reported pg_regress compiling against an old installed libpq-fe.h in #60; Thomas Munro and Tom Lane diagnosed mixed include paths in #61 and #62, and Tom's inline fix in #64 landed as f186f90e55, also backpatched to PostgreSQL 17.
- After the formatting trial, Peter Eisentraut landed Thomas Munro's broader PRI macro conversion as a0ed19e0a9 and separately changed pgbench to parse full-width random seeds with SCNu64 in 53a2a1564a; neither change appears as an attached patchset here.
- Thomas Munro reported a related Meson include-order problem in #65 and pointed to separate work, without establishing a complete resolution in this thread.
Sep 2025 · 3 entries
- Peter Eisentraut discovered that moving pg_int64 could produce conflicting typedefs when extensions combined older server headers with newer libpq headers in #66; Thomas Munro discussed include order and alternative compatibility strategies in #67.
- Peter Eisentraut supplied the typedef-restoration patch in #68 and clarified why include-path order did not eliminate the conflict in #69; it landed with a required <stdint.h> include and clearer comment as 409543da54, retaining the int64_t APIs and deprecation.
- The thread ends with Peter Eisentraut confirming the compatibility repair in #71; the core conversion and subsequent repairs had landed, while the tzcode work remained deferred as stated in #33 and the related Meson concerns had moved to separate discussion.