model-writtenrun 4368 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund reported concurrent LDAP and SSL tests choosing the same port in #1 and proposed several remedies, including PID-bearing lockfiles to coordinate port allocation.
- Andrew Dunstan selected lockfile reservations in #2, proposing an environment override so separate buildfarm builds could share reservations, and identified the need for Meson or Make to supply the build directory in #3.
- Andrew's initial patch #4 reserved ports under an exclusive file lock and checked whether the reserving process remained alive; #5 corrected directory handling while preserving that design.
- Andres questioned Windows behavior in #6; Andrew explained the constants and process-existence check in #7, then added reservation-file cleanup at process exit in #8.
- Andres supported proceeding but requested clearer documentation, better error messages, explicit unlocking and reliable directory selection in #9; Andrew proposed exporting top_builddir in #10, which Andres endorsed in #11.
- Andrew committed 9b4eafcaf4 from #8 with top_builddir integration, explicit unlocking, fixed-width PID writes and diagnostic improvements, announcing the push in #12 and retaining the lowercase variable name for consistency with existing installcheck recipes.
- Andrew confirmed backpatching in #14; afterward Andres reported uninitialized-PID warnings in #15, Tom Lane identified empty reservation files as the likely cause in #16, and Andrew reported the warning fixed in #18.
- Months later, Peter Eisentraut reported in #19 that PGXS installcheck used top_builddir inside the installation tree, causing portlock creation to fail without write permission or write into an inappropriate location, and initially proposed pointing it at the current directory.
- Peter replaced that proposal with removal of the unnecessary top_builddir export in #21, allowing reservations to fall back under tmp_check; Andrew accepted this solution in #22.
- Peter committed #21 unchanged apart from rebase context as 5d42a97519 and confirmed application to all active branches in #23, leaving no outstanding patch or unresolved request at the thread's end.