model-writtenrun 2777 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Sam James reported that PostgreSQL 17.1 failed with GCC 15's new C23 default because PostgreSQL tried to typedef bool, and Tom Lane reproduced the failure and traced it to configure's obsolete stdbool.h conformance test #1 #2.
- Thomas Munro explained that C23 promotes bool and similar macros to keywords; he and Tom settled on trusting stdbool.h while initially retaining the fallback for systems whose bool is wider than one byte #3 #5 #6 #8.
- Thomas submitted a minimal backpatchable conformance-check fix and a separate macOS/PPC cleanup, then Tom supported deleting the untested wider-bool fallback altogether if no current platform needed it #9 #10.
- Peter Eisentraut tested the bool patch with two existing C23 fixes and obtained clean builds on PostgreSQL 16 and later, but found older branches blocked by node-tree-walker function signatures and noted their differing C-language baselines #11 #13.
- The conformance-check fix landed separately from the cleanup; the supplied representative 0305dd3ade adapts the first patch in #9 for PostgreSQL 11 by retaining a header-presence check, as Thomas explained after pushing #14.
- Peter questioned further backport work when older node-tree-walker code could not support C23, and Thomas and Tom agreed to support C23 from PostgreSQL 16 onward while keeping the already-landed bool fixes and considering automatic compiler fallback for older branches #15 #16 #17.
- Thomas proposed requesting C17 only when a compiler selected a newer standard, preserving existing flags otherwise; Tom endorsed that design, and Thomas submitted branch-specific patches tested on PostgreSQL 9.2 and 15 with GCC and Clang #18 #19 #21.
- The discussion prompted Peter's backports renaming the C23 keyword constexpr in predtest.c, recorded as ad89c8bda1 and ae145d1de6; these reused existing work rather than an attachment in this thread #11 #22.
- Peter also backported the explicit amcostestimate function-pointer prototype in pathnodes.h as f3834c12af and f7a929a007, completing the additional C23 changes identified during testing #13 #17.
- Thomas committed the conditional C17 fallback from #21 as f00c401c65 for older branches and reported that their buildfarm results were turning green, while newer branches still encountered a GCC internal compiler error #22.
- Thomas returned to the remaining bool cleanup with a broader patch removing fallback definitions, configuration checks and compatibility workarounds; Tom visually reviewed it, and it landed unchanged as 97525bc5c8 #23 #24.
- Andres Freund reproduced the remaining compiler crash with stricter GCC checking, found that reduced debug information avoided it, and confirmed that an upstream patch identified by Sam fixed the compiler problem #26 #27 #28 #29.
- Alligator remained affected, so Tom recommended keeping it exposed to the compiler bug to detect the upstream fix; Robins Tharakan temporarily reduced its build frequency, then restored it after the December 12 GCC build cleared the failure #30 #31 #32 #33 #34.
- After compilation recovered, Robins reported a PostgreSQL 17 recovery-test failure during concurrent branch runs; Tom suspected a timing race around stopping replication processes, while Andrew Dunstan recommended the buildfarm's coordinated parallel-run mechanism #35 #36 #37 #38.
- The PostgreSQL C23 changes had landed and the upstream compiler failure had cleared, but buildfarm issues remained under discussion: Robins planned configuration updates, and the final message attributed a missing shared-memory segment to possible systemd IPC removal without confirming a resolution #41 #42.