Set ON_ERROR_STOP in test/install/load.sql

[jnasbyupgrade]

test/install/load.sql now sets \set ON_ERROR_STOP on, so psql aborts (exit 3) at the first error in it and pg_regress reports not ok 1 - ../install/load (e.g. "1 of 3 tests failed") instead of the error passing silently. This makes errors visible; it does not add a gate.

No CI step fails on that not ok:

• Fresh mode (pg-build-test, ci.yml:166) ends in make installcheck || status=$?; exit $status, and pgxntool/base.mk:314 has .IGNORE: installcheck, so it exits 0 even on a real diff. This step gates no pg_regress failure at all, in any test file.
• make test-update (ci.yml:196) and existing mode (bin/test_existing, make test) go through pgxntool/base.mk:376, which fails only if regression.diffs exists. pg_regress can never diff load.sql (its expected and result paths coincide; see test/install/.gitignore), so an error there does not fail them either.

Also notes at the first \if that it needs psql 10+.

Checked on PG17: make test and existing-mode run-suite pass; with a bogus SELECT appended to the end of the file, the install step reports not ok and make test still exits 0.

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 7ce0da0e-5507-4393-8df1-b66f58b3d67e

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Claude finished @jnasbyupgrade's task in 16s —— View job

---

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

• \set ON_ERROR_STOP on sits right after \set ECHO none, before any statement that could fail. That is the right place.
• The change only adds a psql variable and comments. It doesn't alter behavior in the existing and fresh/update modes when nothing errors, and the PR's evidence (a bogus statement gives not ok) shows it works.
• The \if comment says psql 10+ is safe because the floor is PG12. I didn't verify the PG12 floor against the repo's metadata.

[github-actions]

Claude finished @jnasbyupgrade's task in 15s —— View job

---

Code review

No issues found in PR #54. I checked the diff for bugs and CLAUDE.md compliance.

• \set ON_ERROR_STOP on sits right after \set ECHO none. That is before any SQL runs, so the DO blocks and the mode checks in the file are all covered.
• The comment on the \if line says the PG12 floor makes psql 10+ safe. That matches .github/workflows/ci.yml, which sets the CI matrix floor at PG12.
• The PR description is candid that this does not make make test exit nonzero, and that whether a not ok fails the CI job depends on pg-build-test. I did not check that either.

I did not run the tests.

[jnasbyupgrade]

Claude here. Decisions needed from the maintainer before this merges:

1. Accept parity, or add a real gate? This PR makes an error in test/install/load.sql visible (not ok 1 - ../install/load, psql exit 3), but make test still exits 0: pgxntool/base.mk:376 fails only when regression.diffs exists, and this file is compared against itself, so none is ever produced. pg_count_nulls' equivalent change behaves the same way. Options for a real gate: (a) a pgxntool change so make test propagates a pg_regress failure (upstream of this repo); (b) a CI step that runs test/install/load.sql with ON_ERROR_STOP or greps the TAP output for ^not ok; (c) a dedicated test whose expected file differs from its result file.
2. Separate PR: move the fresh-mode CI leg to make test? Today it runs only pg-build-test (ci.yml:166), which ends in make installcheck; pgxntool/base.mk:314 marks installcheck as .IGNORE, so that step exits 0 even on a real test diff. The update-mode and existing-mode legs do gate on diffs. pg_count_nulls' fresh leg runs make test.