Add fresh/update/existing-mode extension test infrastructure (#18)

Adds the local test/Makefile machinery for testing this extension's own
version-update path (`ALTER EXTENSION ... UPDATE`) — not a PostgreSQL
major-version `pg_upgrade`, which is unrelated, separate, later work:
`TEST_LOAD_SOURCE=fresh|update|existing` plus
`TEST_UPDATE_FROM`/`TEST_UPDATE_TO` make vars, a single committed
installer covering all three modes, a hand-authored `0.1.0` → `stable`
update script, and one permanent pgTAP assertion proving the extension's
own SQL never relies on an ambient `search_path`. CI wiring that
actually drives `update`/`existing` mode in an automated job is a
separate, later PR — CI here just keeps running the normal fresh-mode
suite.

Author: jnasbyupgrade

.github/workflows/ci.yml

@@ -129,6 +129,11 @@ jobs:
         run: pg-start ${{ matrix.pg }}
       - name: Check out the repo
         uses: actions/checkout@v4
+      # pgxntool's test-build feature (test/build/build.sql) syncs
+      # test/build/*.sql into test/build/sql/ via rsync (run-test-build.sh),
+      # which the pgxn/pgxn-tools image doesn't ship.
+      - name: Install rsync
+        run: apt-get install -y rsync
       - name: Test on PostgreSQL ${{ matrix.pg }}
         # `make test` alone never fails this step: pgxntool/base.mk (as
         # vendored here, 2.2.0) marks its underlying installcheck .IGNORE,

Makefile

@@ -1,10 +1,90 @@
+# Committed-once install of the extension (test/install/load.sql), run before
+# the main pgTAP suite in its own pg_regress session so its state persists
+# (committed) into every per-test file. Must be set (and set to exactly
+# "yes"/"no", not auto-detected) BEFORE base.mk is included below, since
+# base.mk reads it at parse time.
+PGXNTOOL_ENABLE_TEST_INSTALL = yes
+
+# Safeguard for `make results`: refuses to copy test/results/*.out over
+# test/expected/*.out while a real regression is showing. This is already
+# pgxntool's own default, but set it explicitly so that stays true even if a
+# future pgxntool default ever changes.
+PGXNTOOL_ENABLE_VERIFY_RESULTS = yes
+
+# TEST_LOAD_SOURCE selects how test/install/load.sql installs the extension:
+#   - fresh (default): CREATE EXTENSION object_reference (current version).
+#   - update: CREATE EXTENSION at TEST_UPDATE_FROM (default 0.1.0, the only
+#     real historical PGXN release) then ALTER EXTENSION UPDATE -- to
+#     TEST_UPDATE_TO if set, otherwise to the current default_version
+#     ("stable"). Running the SAME suite with the SAME expected output
+#     against the updated database verifies it behaves identically to a
+#     fresh install.
+#   - existing: the extension is ALREADY installed in the target database (by
+#     a binary pg_upgrade, or an ALTER EXTENSION UPDATE done outside the
+#     suite). load.sql does not touch it; it only asserts presence + current
+#     version. Pair with CONTRIB_TESTDB=<db> and
+#     EXTRA_REGRESS_OPTS=--use-existing so pg_regress runs against that
+#     database instead of dropping and recreating a throwaway one.
+#
+# The mode (and the update from/to versions) are signalled to load.sql via
+# placeholder GUCs. pg_regress does not forward make variables, but the psql
+# processes it spawns inherit the environment, so PGOPTIONS reaches load.sql.
+#
+# The GUCs are exported UNCONDITIONALLY, so load.sql can read them WITHOUT
+# missing_ok and fail loudly if they did not propagate. Relying on an absent
+# GUC to mean "fresh" is unsafe: a silent break anywhere in the
+# make -> PGOPTIONS -> env -> psql chain would quietly run the wrong mode.
+#
+# TEST_LOAD_SOURCE must be exactly `fresh`, `update` or `existing`; anything
+# else is a hard error at parse time (so e.g. `make test
+# TEST_LOAD_SOURCE=typo` fails fast rather than defaulting).
+TEST_LOAD_SOURCE ?= fresh
+ifeq ($(filter $(TEST_LOAD_SOURCE),fresh update existing),)
+$(error TEST_LOAD_SOURCE must be 'fresh', 'update' or 'existing', got '$(TEST_LOAD_SOURCE)')
+endif
+
+# update-mode version range (read by load.sql only in update mode). Empty
+# TEST_UPDATE_TO means "update to the current default_version" (stable).
+TEST_UPDATE_FROM ?= 0.1.0
+TEST_UPDATE_TO ?=
+
+export PGOPTIONS := $(PGOPTIONS) -c object_reference.test_load_mode=$(TEST_LOAD_SOURCE) -c object_reference.test_update_from=$(TEST_UPDATE_FROM) -c object_reference.test_update_to=$(TEST_UPDATE_TO)
+
+# Convenience wrapper: `make test-update` == `make test TEST_LOAD_SOURCE=update`.
+# Must recurse (a fresh $(MAKE)) rather than depend on `test`, so the
+# parse-time TEST_LOAD_SOURCE conditional above re-evaluates with update set.
+.PHONY: test-update
+test-update:
+	$(MAKE) test TEST_LOAD_SOURCE=update
+
 include pgxntool/base.mk
 
 testdeps: $(wildcard test/*.sql test/helpers/*.sql) # Be careful not to include directories in this
 testdeps: test_factory
 
 install: cat_tools
 
+# 0.1.0 (TEST_UPDATE_FROM's default -- the update-mode floor, see above) needs
+# count_nulls too: its install script's _object_oid.null_count trigger calls
+# count_nulls' not_null_count_trigger(), and object_reference.control's
+# `requires` (cat_tools only -- count_nulls was dropped once the reg*
+# pseudotype removal made that trigger unnecessary) no longer CASCADEs it in.
+# Only needed for update-mode testing against that floor -- current
+# object_reference has no runtime dependency on count_nulls at all -- so this
+# is conditional, not folded into the unconditional `install: cat_tools` above.
+ifeq ($(TEST_LOAD_SOURCE),update)
+install: count_nulls
+endif
+
+.PHONY: count_nulls
+count_nulls: $(DESTDIR)$(datadir)/extension/count_nulls.control
+$(DESTDIR)$(datadir)/extension/count_nulls.control:
+	pgxn install count_nulls
+
+# Clean the cruft pg_regress writes into test/install/ (the self-comparing
+# result .out and its diff), which is listed in test/install/.gitignore.
+extra_clean += $(addprefix test/install/,$(shell grep -v '^\#' test/install/.gitignore 2>/dev/null))
+
 test: dump_test
 extra_clean += $(wildcard test/dump/*.log)
 dump_test: test/dump/run.sh test/helpers/object_table.sql $(wildcard test/dump/*.sql)