model-writtenrun 5147 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund reported that C++17 rejected register declarations exposed by PostgreSQL headers and proposed removing them in #1; Tom Lane raised possible historical inline-assembly requirements but supported trying removal and watching the buildfarm in #2.
- Andres argued that existing assembly already worked without register and identified another occurrence in pg_regexec in #3, while Fabien COELHO cautioned that removal would lose protection against taking variable addresses in #4.
- After encountering the issue again during Meson work, Andres rejected the address-taking concern, removed register from both headers and regexec.c, and dropped the CI warning suppression in #5; he also explained that -Wregister could not enforce this in ordinary C compilation.
- The attachment in #5 exactly reproduces 03bf971d2d, whose recorded commit time precedes its posting; Peter Geoghegan and Tom subsequently approved the change in #6 and #7, and Andres announced completion in #8.
- In February 2024, Christoph Berg reported that C++ extension builds against PostgreSQL 15 and earlier failed on Ubuntu 24.04, described a preprocessor workaround, and requested backpatching in #9; Tom tentatively supported that possibility because no regressions had been reported in PostgreSQL 16 in #10.
- In October 2024, Tristan Partin renewed the backpatch request after another PostgreSQL 15 extension failure in #11, but Tom argued that older minor releases would still require workarounds and recommended defining register away or using -Wno-register in #12.
- The thread ends with the development-branch fix landed and no backpatch shown: Tristan explained that a backpatch would still help his latest-minor-only support policy, but confirmed that -Wno-register had resolved his build problem in #13.