model-writtenrun 2884 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed making Perl warnings fatal throughout build scripts and tests in #1 so developer mistakes would reliably fail, while identifying an undefined getprotobyname() result in CI and unresolved warnings in the legacy MSVC build system.
- Peter added a separate socket workaround in #2, replacing getprotobyname() with protocol 0; Andrew Dunstan supported it in #3 and explained in #5 why it also addressed the Windows concern raised by Michael Paquier in #4.
- Alvaro Herrera proposed a central warning-policy module to simplify rollback in #6, but Andrew preferred direct declarations with local exceptions in #7.
- Peter committed the socket portion of #2 unchanged as 01226c682c and announced in #8 that the broader warning policy would wait for resolution of the MSVC issues.
- Dagfinn Ilmari Mannsåker explained how a shared pragma module could work in #9, but Andrew considered it excessive and noted perlcritic implications in #10; Peter subsequently marked the main patch returned with feedback pending MSVC removal in #11.
- After the MSVC removal, Peter resumed the remaining work in #12, refreshed the warning-policy patch, added a fix for an invalid restart('fast') call, and reported that the series passed CI on all platforms.
- The restart-call prerequisite from #12 landed unchanged as a7ebd82b9e, followed by the warning policy as c538592959 with the same conversion extended to five pg_combinebackup tests; Peter confirmed completion in #13.
- Bharath Rupireddy reported in #14 that fatal warnings exposed an undefined return status when psql() timed out; Peter proposed guarding status decoding with if (defined $ret) in #15 to preserve the distinct undef timeout result.
- Bharath accepted the solution in #16 and attached it in #17; Peter committed it unchanged as 686db12de3 and confirmed the fix in #18.
- Anton Voloshin found two subsequently added tests missing fatal warnings and supplied a two-file correction in #19, along with an internal CI-checking script and a question about adapting it for Cirrus.
- Alexander Korotkov committed the recovery-test hunk from #19 as e1555645d7 and confirmed it in #20; Peter committed the remaining autovacuum-test hunk as 014720c6d9 and confirmed completion in #21.
- By the final message #21, all attached changes had landed, while Anton's proposal in #19 to enforce the warning policy through Cirrus remained unanswered.