Add new functions, enums

[jnasbyupgrade]

• New enum types for pg_proc fields: routine_type, routine_argument_mode, routine_volatility, routine_parallel_safety
• New relation functions: relation__is_temp, relation__is_catalog, relation__column_names
• New relkind entries and permission tests
• Version bump to 0.3.0
• Updated pgxntool to 2.0.2 (adds pg_tle support, run-test-build.sh, verify-results-pgtap.sh, and more)
• CI expanded to PG 10–18; upgrade test matrix accounts for PG11+ attmissingval and PG12+ oid visibility breakage in pre-0.2.2 scripts
• Historical SQL install files (0.1.0, 0.2.1, 0.2.2) now tracked in git

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

CLAUDE.md (1)

26-38: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Rename extension update terminology in Makefile
The Makefile comment and upgrade_scripts_out still use “upgrade” for extension version changes; rename them to “update scripts” / update_scripts_out to match the terminology rule.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@CLAUDE.md` around lines 26 - 38, Update the Makefile terminology from
“upgrade” to “update” for extension version-change scripts: rename the
upgrade_scripts_out variable to update_scripts_out and revise its related
comment to say “update scripts,” preserving all existing generation behavior and
references.

Source: Learnings

sql/cat_tools--0.2.2--0.3.0.sql.in (1)

12-21: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Stale function reference reintroduces its own "deprecated" warnings during the upgrade.

c_simple_args is computed via cat_tools.function__arg_types_text(args). Later in this same script (line 572-586) that exact function is CREATE OR REPLACEd into a deprecated wrapper that does RAISE WARNING 'function__arg_types_text() is deprecated...' and delegates to routine__parse_arg_types_text. Every __cat_tools.create_function(...) call after that point (roughly a dozen+ remaining statements: cat_tools.regprocedure, object__catalog, _pg_sv_column_array, relation__is_temp/is_catalog/column_names, trigger__parse, etc.) will now silently route through the deprecated wrapper and spam a "deprecated" warning during the extension's own ALTER EXTENSION ... UPDATE, confusing users who never called the deprecated function themselves.

The fresh-install script (cat_tools.sql.in/cat_tools--0.3.0.sql.in) avoids this because its create_function already uses cat_tools.routine__parse_arg_types_text(args) directly (bootstrapped via the temporary stub). This update script should do the same.

🐛 Proposed fix

 DECLARE
-  c_simple_args CONSTANT text := cat_tools.function__arg_types_text(args);
+  c_simple_args CONSTANT text := cat_tools.routine__parse_arg_types_text(args);

Note: at line 21's point in the script, cat_tools.routine__parse_arg_types_text doesn't exist yet either — it's only created later via __cat_tools.create_function itself (line 527+) — so this fix needs the same bootstrap-stub trick used in the fresh-install script (create a temporary pass-through cat_tools.routine__parse_arg_types_text(text) stub before __cat_tools.create_function is defined, matching cat_tools.sql.in lines 74-81).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@sql/cat_tools--0.2.2--0.3.0.sql.in` around lines 12 - 21, Update
__cat_tools.create_function to use cat_tools.routine__parse_arg_types_text(args)
instead of the deprecated cat_tools.function__arg_types_text(args). Before
defining __cat_tools.create_function, add the temporary pass-through
routine__parse_arg_types_text(text) bootstrap stub used by the fresh-install
script, then preserve the existing later replacement with the full
implementation.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Outside diff comments:
In `@CLAUDE.md`:
- Around line 26-38: Update the Makefile terminology from “upgrade” to “update”
for extension version-change scripts: rename the upgrade_scripts_out variable to
update_scripts_out and revise its related comment to say “update scripts,”
preserving all existing generation behavior and references.

In `@sql/cat_tools--0.2.2--0.3.0.sql.in`:
- Around line 12-21: Update __cat_tools.create_function to use
cat_tools.routine__parse_arg_types_text(args) instead of the deprecated
cat_tools.function__arg_types_text(args). Before defining
__cat_tools.create_function, add the temporary pass-through
routine__parse_arg_types_text(text) bootstrap stub used by the fresh-install
script, then preserve the existing later replacement with the full
implementation.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: fd44ba13-5ca3-459c-be71-1192a00be8a0

📥 Commits

Reviewing files that changed from the base of the PR and between e3db247 and af30da3.

⛔ Files ignored due to path filters (2)

• test/build/expected/build.out is excluded by !**/*.out
• test/expected/relation__.out is excluded by !**/*.out

📒 Files selected for processing (9)

• .github/workflows/ci.yml
• CLAUDE.md
• Makefile
• sql/cat_tools--0.2.2--0.3.0.sql.in
• sql/cat_tools--0.3.0.sql.in
• sql/cat_tools.sql.in
• test/.gitignore
• test/gen-relkinds.sh
• test/sql/relation__.sql

[jnasbyupgrade]

@claude please review this PR.

[claude]

Claude finished @jnasbyupgrade's task in 3m 11s —— View job

---

Reviewing PR #16

• Gather context (PR description, prior CodeRabbit reviews, diff)
• Review current diff against origin/master
• Check outstanding issues from prior CodeRabbit rounds
• Post review feedback

CI is currently green on the latest commit (af30da3). Below is a review focused on what's still outstanding at HEAD — CodeRabbit already caught and the author already fixed several rounds of issues (the routine__.out ambiguous-overload bug, the relkind mapping swap, the upgrade-script REVOKEs, README typos). This pass verifies current file state rather than repeating what's already resolved.

Confirmed, not yet fixed

1. Spurious deprecation warnings during every 0.2.2→0.3.0 update — sql/cat_tools--0.2.2--0.3.0.sql.in:21

c_simple_args CONSTANT text := cat_tools.function__arg_types_text(args);

__cat_tools.create_function() calls this on every invocation. Later in the same script (line ~579) cat_tools.function__arg_types_text is redefined as the deprecated wrapper:

RAISE WARNING 'function__arg_types_text() is deprecated, use routine__parse_arg_types_text instead';
RETURN cat_tools.routine__parse_arg_types_text(arguments);

There are 7 more __cat_tools.create_function(...) calls after that redefinition (lines 589, 613, 672, 688, 702, 716, 732). Every one of them will now trigger this warning during ALTER EXTENSION cat_tools UPDATE, even though the extension itself — not a user — is the caller. This was flagged in the most recent CodeRabbit pass (2026-07-13) and is still present. Fix: switch line 21 to call cat_tools.routine__parse_arg_types_text(args) directly (bootstrapping a temporary stub first, the same way cat_tools.sql.in does, since that function doesn't exist yet at this point in the script).

2. README.asc:23 still says the old minimum version

Works on Postgres 9.3 and above.

This PR requires PG12+ (per HISTORY.asc and the CI matrix). Flagged in two earlier review rounds and still unfixed.

3. META.in.json/META.json line 70 stale comment

"X_comment": "Requires named argument support in SQL functions, which started in 9.2",

This no longer matches the "PostgreSQL": "12.0" prereq a few lines below (that part was fixed). The explanatory comment wasn't updated to match.

4. No permissions: block in .github/workflows/ci.yml
The workflow has no top-level permissions:, so GITHUB_TOKEN defaults to broad write scope for a workflow that only needs to read the repo and run tests. Low severity, but a quick win — permissions: {contents: read} at the top level. Note: per this repo's Claude Code setup, I can't push a fix for this myself since I don't have permission to modify files under .github/workflows/.

Confirmed, minor/edge-case, low priority

5. trigger__parse() regex injection via relation name — sql/cat_tools.sql.in:1808-1809

v_on_clause := ' ON ' || r_trigger.tgrelid::pg_catalog.regclass || ' ';
v_array := regexp_split_to_array( v_work, v_on_clause );

regclass output for a quoted identifier can contain regex metacharacters (., +, (, etc. are all valid in quoted identifiers), which would make regexp_split_to_array mis-split or error. This is a narrow edge case (unusual table names) and was flagged as out-of-diff in the first review round; still present in both cat_tools.sql.in and the duplicated parser in sql/cat_tools--0.2.2--0.3.0.sql.in. Worth a follow-up but not blocking.

Verified as already resolved (no action needed)

• routine__.out ambiguous-overload test failure — fixed by removing the ambiguous routine__type overload rather than casting in the test; ok 2 passes now.
• relation__kind/relation__relkind swapped c/f/m mapping — fixed, fresh install now matches the update path.
• Upgrade script missing REVOKE ... FROM PUBLIC before granting _cat_tools helpers — fixed.
• README typos (stray backtick, "it's" → "its") — fixed.
• The "split enum ADD VALUE into its own step" suggestion from the 2026-07-12 pass — not actually a bug: PG12 specifically lifted the restriction on using a newly-added enum value later in the same transaction (this is why PG10/11 support was dropped, per CLAUDE.md's CI policy section), so using relation_type/relation_relkind new values later in the same update script is safe on the supported PG12+ floor.

Not verified (no local Postgres available in this environment)

I didn't have a way to actually run make installcheck / the pgTAP suite here, so the above is based on static reading of the SQL, not an executed test run. CI already ran and passed at af30da3, which covers the actual test execution.