model-writtenrun 541 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Vladlen Popolitov reported that the tzcode update broke timezone-data generation on filesystems without hard links and proposed allowing zic to fall back on EINVAL, reporting successful exFAT, NTFS and ext4 tests #1.
- Tom Lane consulted tzdb upstream and questioned PostgreSQL's Windows error translation; Vladlen confirmed that CreateHardLinkA returns ERROR_INVALID_FUNCTION on exFAT, which becomes EINVAL, and explained why Windows symlink support cannot provide the fallback #2 #3 #4.
- The investigation also exposed different unsupported-link errors on Linux and BSD, and Tom confirmed that NetBSD distinguishes EOPNOTSUPP from ENOTSUP while seeking an upstream resolution #4 #5.
- Vladlen proposed translating ERROR_INVALID_FUNCTION specifically inside the Windows link wrapper, and Tom endorsed this because changing the global error mapping could affect unrelated calls and the localized fix would remain compatible with upstream zic changes #6 #7.
- Vladlen submitted a revised patch combining that Windows fix with Linux and BSD fallback handling in zic, reporting an Ubuntu exFAT failure and successful testing of the changes #8.
- Tom committed the Windows-only fix as 481052c775, corresponding to the Windows subset of #8 with explanatory and stylistic edits, although he had implemented it from the inline proposal before seeing the attachment #6 #9.
- Tom deferred the Linux and BSD changes until a future tzcode synchronization because upstream had already implemented a fix and he wanted to avoid a divergent local patch; Vladlen accepted the Windows resolution and mentioned a separate junction-point buffer issue being discussed elsewhere #9 #10.