model-writtenrun 9429 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrew Dunstan proposed reusing an existing temporary installation when NO_TEMP_INSTALL is set, reducing slow installations especially on Mingw, and requested backpatching through PostgreSQL 9.5 in #1.
- Tom Lane identified missing quotes around path assignments in #3, and Andrew incorporated that correction when landing #1 as 47b3c26642, with backports through PostgreSQL 9.5.
- Andres Freund reopened the design in #4, asking why pg_upgrade retained its private installation machinery; Andrew explained the need to ensure an installation exists in #5, and Tom distinguished standalone make check from reuse during check-world in #6.
- Andres supplied a replacement in #7 that delegates installation and environment setup to the shared temp-install machinery, which already honors NO_TEMP_INSTALL, and corrected a duplicated MAKE assignment in #8.
- The replacement landed as 7005389b2a, removing the private installation code; its final source is the later matching submission #1085857 in thread #213535, with the obsolete DOINST reference also removed.
- Thomas Munro reported that cfbot now failed without an ordinary installation and described a temporary make install workaround in #9; Tom reproduced the failure and traced it to pg_regress receiving the permanent installation's bindir in #10.
- Andres proposed restoring the EXTRA_REGRESS_OPTS bindir override unconditionally in test.sh in #11 and demonstrated the same latent NO_TEMP_INSTALL failure on PostgreSQL 11 in #13; Tom cautioned about cross-version testing in #12 and supported using the earlier version's psql in #14.
- Andrew reproduced the missing-psql failure with the buildfarm installation step disabled in #16, and Andres supplied the explicit inline oldbindir fix in #17, explaining that existing environment setup already handled library paths.
- Andres committed the inline fix as f916221df5 and backpatched it, making pg_regress select psql from the installation being upgraded instead of relying on a permanent installation.
- After Andres requested verification in #18 and Tom noted that cfbot's workaround should be removable in #19, Andrew confirmed successful buildfarm checks with installation skipped in #20, leaving no outstanding patch in this thread.