Remove redundant client_min_messages handling from install script

[jnasbyupgrade]

CREATE EXTENSION already raises client_min_messages to WARNING for the install script (only-raising, so a stricter caller is respected) and restores it afterward — so the script's own setting was redundant, and being blind it actually lowered a stricter caller's level during install. Remove it.

[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: Pro Plus

Run ID: a2250ea6-5c19-499a-a58d-dcae5594dfdd

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

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[jnasbyupgrade]

@claude please review this PR.

[claude]

Claude finished @jnasbyupgrade's task in 5m 2s —— View job

---

Claude finished the review

• Gather context (PR diff, CLAUDE.md conventions, Makefile generation rules)
• Understand the request (code review of sql/cat_tools.sql.in changes)
• Review the guard logic and its interaction with CREATE EXTENSION/transaction semantics
• Check repo conventions (generated .sql.in tracking rules)
• Post review feedback

Review of sql/cat_tools.sql.in

The overall approach (save the caller's client_min_messages, only raise it to WARNING when the caller's setting is more verbose, restore afterward) is the right idea. Two issues found, one of which is significant:

1. Generated per-version file not regenerated (blocking)

Per this repo's own convention (CLAUDE.md, "SQL file conventions" rule 4), sql/cat_tools--0.2.2.sql.in is generated by make from sql/cat_tools.sql.in but must still be tracked. This PR only touches sql/cat_tools.sql.in — sql/cat_tools--0.2.2.sql.in still contains the old, unguarded SET LOCAL client_min_messages = WARNING; (verified by reading the file on this branch). Since PGXN packaging installs from the tracked --0.2.2.sql.in rather than running make first, the fix won't actually ship until this file is regenerated and committed.

Fix this →

2. The save/restore mechanism can still leak past install (the bug the PR is trying to fix)

sql/cat_tools.sql.in lines ~1-27 stash the caller's client_min_messages into a custom GUC (cat_tools.client_min_messages) via SET LOCAL (transaction-scoped, correct), then raise via a plain SET client_min_messages = warning; (non-local). At the end of the file (~line 1566), the restore does:

SELECT pg_catalog.set_config('client_min_messages', pg_catalog.current_setting('cat_tools.client_min_messages'), false);

false means non-local (session-level) SET.

Per Postgres GUC-stacking rules, when a plain (non-local) SET happens after a SET LOCAL within the same transaction level, the non-local SET wins and the value propagates past COMMIT. Concretely:

BEGIN;
SET LOCAL client_min_messages = error;   -- caller wants this only for this transaction
CREATE EXTENSION cat_tools;              -- our script sees 'error', skips raising (already >= warning)
                                          -- final restore does a *non-local* SET of 'error'
COMMIT;
-- client_min_messages is now permanently 'error' for the rest of the session,
-- instead of reverting to whatever it was before BEGIN.

This reproduces exactly the "leaking the change past install" problem the PR's own comment says it's avoiding — just for a caller-supplied SET LOCAL instead of our own raise.

Suggested simplification: since a DO block without an EXCEPTION clause doesn't open a subtransaction, a guarded SET LOCAL by itself is sufficient and needs no save/restore at all:

DO $$
DECLARE
  c_levels CONSTANT text[] := ARRAY['debug5','debug4','debug3','debug2','debug1','log','notice','warning','error'];
  v_current CONSTANT text := pg_catalog.current_setting('client_min_messages');
BEGIN
  IF pg_catalog.array_position(c_levels, pg_catalog.lower(v_current))
     < pg_catalog.array_position(c_levels, 'warning')
  THEN
    SET LOCAL client_min_messages = warning;
  END IF;
END
$$;

SET LOCAL is inherently transaction-scoped, so it automatically reverts to whatever was in effect (session-level or an ambient caller's SET LOCAL) once that transaction ends — correctly handling both the plain top-level CREATE EXTENSION case and the "runs inside a larger transaction" case this PR is explicitly trying to support, with no custom GUC, no explicit restore statement, and less code.

Fix this →

Minor

• No test coverage was added for this behavior (e.g. a TAP test that runs CREATE EXTENSION inside an explicit transaction with a stricter client_min_messages and asserts it isn't clobbered, or is restored correctly after commit).
• The rest of the logic (level ordering array, array_position comparison, LOG sorting between DEBUG1/NOTICE) is correct and matches documented client_min_messages semantics.

[jnasbyupgrade]

Folding the install-script change (removing the redundant SET LOCAL client_min_messages) into #16 (commit b6fc753) rather than carrying a separate PR. On new_functions the build sanity test needed the suppression moved into test/build/build.sql, since it \i's the install script directly. Note: the pgxntool subtree updates that were also on this branch were NOT folded — those are a separate concern from the titled change.