model-writtenrun 5879 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed replacing regression-test path substitution with environment variables and a new psql \getenv command, preferring this over command-line variable assignments because it avoided shell-quoting and linking complications; he attached a proof of concept in #1.
- Tom expanded the prototype into four patches in #2, using \set to concatenate and quote paths, constructing the tablespace path inside the test, filtering unstable path-bearing output, and separating script conversion, file relocation, and infrastructure removal.
- Corey Huinker supported the series and suggested lowercase psql variables to distinguish them from environment variables in #3; Tom proposed PG_ prefixes for the environment variables in #4, and Corey endorsed that addition in #5.
- Corey also suggested importing groups of environment variables in #5, but Tom considered that premature in #6; Corey accepted that decision and reported that the attached series passed make check-world after application with patch -p1 in #7.
- Peter Eisentraut pointed out the existing shell-echo workaround in #8, but Tom rejected it for Windows portability in #9; Andrew Dunstan endorsed adding \getenv in #10.
- Tom committed the unchanged \getenv component of #2 as 33d3eeadb2, including its documentation and regression test.
- Tom committed the script-conversion component of #2 as d1029bb5a2, incorporating lowercase psql variables and PG_ environment-variable prefixes.
- Tom combined the file-relocation and infrastructure-removal components of #2 into dc9c3b0ff2, completing the attached series without substantive changes to those components.
- Tom then implemented the dblink cleanup proposed but left outside the attachments in #2, merging its paths initialization into the main test in 0e6e7f0806; the commit identifies broader main-regression-test refactoring as future work.
- By the final message, the attached series and dblink follow-up had landed; Greg Stark warned in #11 that bulk environment-variable imports could introduce security problems, reinforcing the concern about the unimplemented extension rather than reporting a defect in the committed command.