model-writtenrun 7949 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed abstract Unix-domain sockets using @-prefixed names in #1, avoiding filesystem setup, lock files, and cleanup while giving up filesystem access permissions; he also supplied bind() hint and obsolete errno-guard cleanups.
- The errno-guard cleanup from #1 landed as 555eb1a4f0, adjusted only for guards already removed from the base, and Peter rebased and resent the remaining two patches in #2.
- Michael Paquier reviewed and tested the remaining work in #3, questioning permission documentation and reporting that an abstract-socket address conflict incorrectly advised deleting a nonexistent file; Peter defended the conventional @ notation in #4, while Andreas Karlsson highlighted network-namespace behavior in #5.
- Michael proposed constructing the hint sentence by sentence in #6, but Peter raised translation concerns and considered removing the Unix hint or replacing its port wording in #7; subsequent discussion with Michael and David G. Johnston explored how socket addresses and PostgreSQL's port setting should be described in #8, #9, #10, and #12.
- Peter narrowed the change in #13 to removing the file-removal instruction while retaining the existing port wording, explaining that abstract sockets have no file to remove and that the filesystem-socket branch was unreachable under the proposed EADDRINUSE restriction; broader wording changes were deferred to a separate thread.
- David questioned whether socket-file replacement exposed a missing collision check in #14, #16, and #18; Peter distinguished lock-file protection from bind() behavior in #15 and #19, while Michael supported the narrowed fix in #17.
- Peter committed both remaining patches from #13 unchanged as c9f0624bc2 for abstract sockets and d5d91acdcc for accurate bind() hints, announcing completion in #21.
- The thread ended with all three patches landed and Peter explaining in #22 that a separate file-existence check would race, that lock files provide the protection, and that such a check would not change the bind() behavior addressed by the patch.