model-writtenrun 6024 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- The thread began with Andres Freund's announcement that the libpq_pipeline test now used TESTDIR to avoid Windows PATH-separator problems #1; Andrew Dunstan then pointed out that MSVC builds executables elsewhere, supplied an MSBUILDDIR-based fix #2, and confirmed that it passed the test #3.
- Andres Freund argued that platform-specific executable lookup belonged outside individual tests #4, while Andrew Dunstan explained that relying on an installed libpq_pipeline executable conflicted with its intended status as an uninstalled test program #5.
- Andrew Dunstan introduced a shared TestLib::find_built_program() helper #6; Alvaro Herrera questioned its recursive search and separately identified unchecked command failures #7, while Andrew explained that modules could contain multiple executables and considered globbing instead #8.
- Andres Freund proposed having Makefile.global.in and vcregress.pl put the appropriate module build directory on PATH #9; Andrew Dunstan implemented that approach #10, and Andres endorsed its direction #11.
- Andrew Dunstan expanded the PATH changes to additional test invocation modes, corrected the remaining executable reference, and moved the module directory to the end after encountering an initdb executable-location failure, reporting successful Linux testing #12.
- Tom Lane warned that appending the directory could select a program from another PostgreSQL version and proposed placing it immediately after the installation directory #13; Andrew Dunstan adopted that ordering and reported successful MSVC and Linux testing in the thread's final message #14.
- The final attachment #14 landed unchanged as f4ce6c4d3a, with a backport to PostgreSQL 14; the separate run_command() failure-handling issue raised in #7 was not addressed by these patches.