model-writtenrun 7543 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Thomas Munro opened the thread with the announcement that latch wakeups had moved from SIGUSR1 to dedicated SIGURG, with signal setup centralized in InitializeLatchSupport to support future handler-free implementations #1.
- Tom Lane questioned whether postmaster should install dummy_handler instead of ignoring SIGURG to preserve signals before children install their handlers; Thomas initially agreed to make the handling consistent while doubting that latches could lose a necessary wakeup #2 #3.
- Thomas attached a patch changing postmaster's handler and the epoll initialization, but questioned its value because latch users explicitly advertise that they are waiting before receiving SIGURG #4.
- After examining InitializeLatchSupport, Tom agreed that platform differences favored retaining SIG_IGN and proposed an explanatory comment instead; Thomas accepted that direction #5 #6.
- Tom also questioned modifying UnBlockSig, prompting Thomas to explain that signalfd requires SIGURG to remain blocked whereas kqueue can receive ignored signals; their subsequent discussion clarified that Linux and BSD-derived systems treat blocked, ignored signals differently #5 #6 #7 #8.
- Thomas committed the comment-only resolution as 8e861eaae8, adopting the explanation proposed in #5 and leaving the attached behavioral patch uncommitted.
- The thread ended with Thomas clarifying that his platform comparison came from source code and that Debian's sigpending manual incorrectly described Linux behavior; the comment fix had landed, with no further patch proposed #9.