model-writtenrun 7302 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut opened the thread with macOS linker warnings caused by Homebrew Perl injecting a deployment-target flag, and proposed stripping -mmacosx-version-min from PL/Perl's linker options in #1.
- Tom Lane supported removing that flag but argued for broader filtering of unwanted compiler and linker options in #2; Peter instead used ExtUtils::Embed's construction of ldopts to subtract $Config{ldflags} together with $Config{ccdlflags} in #3, eliminating the special -arch filter too.
- Tom approved the broader solution with a commit-message typo correction in #4, while Andres Freund found that it repaired GCC builds against system Perl on AIX and also worked on Solaris in #7.
- Further pruning of library flags was left out: Andres found that AIX with GCC still needed pthread linkage in #7, Peter raised variant libperl names in #8, and Andres endorsed the existing improvement without further pruning in #10.
- Peter committed the revised implementation from #3 as b4e936859d, retaining its code and correcting the commit-message typo.
- After landing, Andres raised the removal of rpath settings on NetBSD and SUSE in #12; Tom examined NetBSD's flags and concluded that PostgreSQL's own configuration should supply the necessary paths in #13, so no additional change followed.
- Peter reported that the Cygwin buildfarm animal could no longer load PL/Perl because its magic block was missing in #14, and Tom suspected the loss of --export-all-symbols in #15.
- Tom investigated whether Perl headers interfered with export annotations in #17, but Andrew Dunstan's preprocessor output showed both functions retaining visibility annotations in #21; subsequent symbol inspection and linker-flag experiments were coordinated in #22, #23, and #24.
- Andrew reported that --enable-auto-image-base did not help but --export-all-symbols did in #28; Tom supplied a Cygwin-only Makefile workaround in #29, which Andrew and Andres endorsed in #30 and #31.
- Peter landed Tom's inline workaround, with a discussion-link comment, as 1c3aa54502, restoring PL/Perl on Cygwin while leaving the underlying explanation uncertain.
- In the final message, Andres connected the Cygwin failure to a similar PL/Python issue involving mixed visibility attributes and __declspec(dllexport), and requested a test using an explicit PGDLLEXPORT definition without the workaround in #33.
- The subsequent MinGW export-definition change in 0e497eadb1 came from the matching patch in #516923 in thread #46682, correcting the earlier attribution to this discussion; this thread's filtering change and Cygwin workaround had landed, while the proposed Cygwin export-definition experiment remained unanswered.