Add sql-lint: style linter for PostgreSQL SQL files (#49)

Add a linter, vendored as a git submodule at `.vendor/linter`
(https://github.com/Postgres-Extensions/linter) . Add to CI,
and fix all linting issues.

Author: jnasbyupgrade

.github/workflows/ci.yml

@@ -308,6 +308,26 @@ jobs:
           # (pgxntool marks installcheck `.IGNORE`), so failures would pass silently.
           make verify-results
 
+  # Style linter (https://github.com/Postgres-Extensions/linter, vendored at
+  # .vendor/linter). Deliberately checked out WITHOUT submodules -- `make
+  # lint` is the same command a developer runs locally, and lint.mk
+  # self-initializes the submodule on first use (see its comment). Using the
+  # exact same entry point here is what actually proves that self-init works,
+  # rather than papering over it with a submodules: true checkout. The
+  # linter's own test suite (fixtures + scanner edge cases) is that repo's
+  # own CI's job, not this one's. No PostgreSQL needed -- sql-lint is a
+  # standalone Perl script -- so this doesn't use the pgxn-tools container.
+  lint:
+    needs: [changes]
+    if: needs.changes.outputs.docs_only != 'true'
+    name: 🧹 SQL Lint
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out the repo
+        uses: actions/checkout@v6
+      - name: Lint SQL
+        run: make lint
+
   # Proves cat_tools survives a BINARY pg_upgrade (in-place catalog migration to a
   # newer PostgreSQL major), not just an in-place extension update. Each leg is a
   # SINGLE jump: install an old cat_tools on an old cluster, plant a dependency
@@ -872,7 +892,7 @@ jobs:
   # the heavy jobs gated off by the `changes` job on a docs-only push), and
   # fails if any failed or were cancelled.
   all-checks-passed:
-    needs: [changes, test, pg-upgrade-test, pg-upgrade-stepwise, extension-update-test, pg-tle-test, pg-tle-upgrade-test]
+    needs: [changes, test, lint, pg-upgrade-test, pg-upgrade-stepwise, extension-update-test, pg-tle-test, pg-tle-upgrade-test]
     if: always()
     runs-on: ubuntu-latest
     steps:

.gitmodules

@@ -0,0 +1,3 @@
+[submodule ".vendor/linter"]
+	path = .vendor/linter
+	url = https://github.com/Postgres-Extensions/linter.git

.vendor/linter

@@ -92,3 +92,16 @@ $(DESTDIR)$(datadir)/extension/cat_tools--0.2.0.sql:
 .PHONY: clean_old_version
 clean_old_version:
 	pgxn uninstall --unstable 'cat_tools=0.2.0'
+
+# Style linter (see https://github.com/Postgres-Extensions/linter, vendored
+# at .vendor/linter -- lint.mk is the thin local hand-off, see its comment).
+# Scoped to the actively-maintained source rather than the default
+# `sql/ test/`: version-specific install/update scripts under sql/ are
+# frozen once released (SQL file conventions rule 5 in CLAUDE.md — never
+# hand-edited again), so linting them would produce permanent, unfixable
+# findings and make `make lint` unusable as a CI gate. Lint the current
+# source instead; a version file still under active development can be
+# linted directly, e.g.
+# `.vendor/linter/sql/bin/sql-lint sql/cat_tools--0.3.0.sql.in`.
+LINT_TARGETS = sql/cat_tools.sql.in sql/omit_column.sql test/
+include lint.mk

Makefile

@@ -0,0 +1,11 @@
+# lint.mk — thin wrapper; the whole local footprint for consuming
+# https://github.com/Postgres-Extensions/linter. Everything else lives in
+# the .vendor/linter submodule; see its README for available targets/rules.
+#
+# Self-initializing (via the rule below) so `make lint` works right after a
+# plain `git clone`, with no --recurse-submodules needed, and so CI can rely
+# on the exact same entry point a developer would use locally.
+.vendor/linter/lint.mk:
+	git submodule update --init -- .vendor/linter
+
+include .vendor/linter/lint.mk

lint.mk

@@ -199,23 +199,51 @@ define _apply_version_seds
 		{print}' $@.tmp > $@.tmp2 && mv $@.tmp2 $@.tmp
 endef
 
-# The recipe builds $@.tmp then moves it into place atomically.
+# ----------------------------------------------------------------------------
+# Two .sql.in sources feed the same .sql.in -> .sql transform below (wrap with
+# @generated@ markers, then _apply_version_seds), producing two different
+# targets:
+#
+#   sql/cat_tools.sql.in (hand-maintained master, committed)
+#     --pattern rule-->  sql/cat_tools.sql
+#
+#   sql/cat_tools.sql.in
+#     --copy rule, tags @generated@ as "VERSIONED FILE!"-->
+#       sql/cat_tools--X.Y.Z.sql.in (committed per-version snapshot; frozen
+#       once released, see CLAUDE.md's SQL file conventions)
+#         --override rule, same transform as the pattern rule-->
+#           sql/cat_tools--X.Y.Z.sql = $(EXTENSION_VERSION_FILES)
+#           (what CREATE EXTENSION actually installs)
+#
+# The bottom (override) rule can't just be left to the sql/%.sql pattern rule
+# matching it: control.mk (auto-generated by pgxntool from cat_tools.control,
+# see pgxntool/base.mk) already defines its OWN recipe for
+# $(EXTENSION_VERSION_FILES) -- straight from cat_tools.sql, no .sql.in layer,
+# no version seds -- and GNU Make always prefers an explicit rule over a
+# pattern rule for the same target. Overriding it here is the only way to
+# route that target through the same .sql.in / version-sed pipeline as
+# everything else; its "overriding recipe" warning is expected. Both final
+# steps build $@.tmp then move it into place atomically.
+#
 # TODO: refactor the version handling into a function.
+# ----------------------------------------------------------------------------
+
+# @generated@ becomes the "-- GENERATED FILE! DO NOT EDIT!" marker below via a
+# plain, unanchored substring match -- it also fires on a handful of
+# coincidental @generated@ occurrences inside real-code comments in
+# cat_tools.sql.in, which is harmless (already inside a -- comment).
 sql/%.sql: sql/%.sql.in pgxntool/safesed
 	(echo @generated@ && cat $< && echo @generated@) | sed -e 's#@generated@#-- GENERATED FILE! DO NOT EDIT! See $<#' > $@.tmp
 	$(_apply_version_seds)
 	mv $@.tmp $@
 
-# Make the current version's .sql.in by copying the base source; the pattern
-# rule above then turns it into the final .sql with SED substitutions applied.
-# (EXTENSION_VERSION_FILES is just sql/cat_tools--<current version>.sql)
+# Appends " VERSIONED FILE!" after every @generated@ occurrence; the pattern
+# rule above resolves the leading @generated@ either way, so the tag rides
+# through into the final marker text untouched.
 $(EXTENSION_VERSION_FILES:.sql=.sql.in): sql/cat_tools.sql.in cat_tools.control
-	cp $< $@
+	sed -e 's/@generated@/@generated@ VERSIONED FILE!/' $< > $@
 
-# Override control.mk's rule (which builds EXTENSION_VERSION_FILES straight from
-# cat_tools.sql, skipping SED) so we build from the .sql.in above instead, with
-# version-conditional substitutions applied. GNU Make's "overriding recipe"
-# warning for this target is expected.
+# See the overview above for why this duplicates the pattern rule's recipe.
 $(EXTENSION_VERSION_FILES): $(EXTENSION_VERSION_FILES:.sql=.sql.in) pgxntool/safesed
 	(echo @generated@ && cat $< && echo @generated@) | sed -e 's#@generated@#-- GENERATED FILE! DO NOT EDIT! See $<#' > $@.tmp
 	$(_apply_version_seds)