check_text, for the oracles check_num cannot express (#418)

test/harness_selftest.sh

@@ -285,6 +285,17 @@ check "check_num still fails two unequal numbers" \
 check "check_num accepts a decimal and a sign" \
 	"$(_probe check_num "decimal" "-1.5" "-1.5")" "0"
 
+check "check_text refuses two empty strings, where plain check passes" \
+	"$(_probe check_text "empty vs empty" "" "")" "1"
+check "check_text refuses one empty side" \
+	"$(_probe check_text "one empty" "abc" "")" "1"
+check "check_text compares two md5 hashes, which check_num cannot" \
+	"$(_probe check_text "md5" "9dd4e461268c8034f5c8564e155c67a6" "9dd4e461268c8034f5c8564e155c67a6")" "0"
+check "check_text still fails two different strings" \
+	"$(_probe check_text "differ" "abc" "def")" "1"
+check "check_num refuses an md5, which is why check_text exists" \
+	"$(_probe check_num "md5" "9dd4e461268c8034f5c8564e155c67a6" "9dd4e461268c8034f5c8564e155c67a6")" "1"
+
 check "check_ratio refuses an empty measurement" \
 	"$(_probe check_ratio "empty" "" "100" "0.5")" "1"
 check "check_ratio refuses a zero denominator rather than dividing by it" \

test/lib.sh

@@ -337,6 +337,33 @@ pgc_is_number() {	# $1 -> 0 when $1 is a number
 	return 0
 }
 
+# check_text LABEL GOT WANT -- check, with both sides required to be non-empty.
+#
+# check_num covers a measurement that is a NUMBER. Plenty of oracles are not: an
+# md5 over an ordered result, a plan node name, a returned string. check_num
+# rejects those outright -- it refuses two identical md5 hashes, because an md5
+# is not a number -- so a suite comparing one has nothing to reach for and falls
+# back to plain check, where "" equals "" and prints PASS (#418).
+#
+# That is the same defect, on the larger half: 35 places in this tree compare an
+# md5(string_agg(...)) oracle, and every one of them is a down cluster or an
+# errored query away from comparing nothing with nothing.
+#
+# Deliberately weaker than check_num: it asserts presence, not shape. A caller
+# that knows the shape should say so, and native_index_projection.sh's agree()
+# additionally requires 32 hex characters before it trusts either side.
+check_text() {
+	local name="$1" got="$2" want="$3"
+	if [ -z "$got" ] || [ -z "$want" ]; then
+		PGC_CHECKS=$((PGC_CHECKS + 1))
+		PGC_FAIL=1
+		echo "FAIL  $name: a side is empty, so nothing was compared:" \
+			"got [$got] want [$want]"
+		return 1
+	fi
+	check "$name" "$got" "$want"
+}
+
 # check_num LABEL GOT WANT -- check, with both sides required to be numbers.
 check_num() {
 	local name="$1" got="$2" want="$3"

test/native_index_projection.sh

@@ -93,9 +93,9 @@ check "partial index finds the row inside its predicate" \
 # prints PASS (#418). This oracle is the strongest assertion in the file, so it is the
 # worst one to have silently comparing two empty strings.
 #
-# Local guard on purpose. #418 proposes `check_num` in `test/lib.sh` and
-# @ChronicallyJD owns it; this file should adopt that helper when it lands and drop
-# the check below.
+# The emptiness half is `check_text` in `test/lib.sh` now (#418, #422), so this file no
+# longer carries its own version of it. The shape half stays here: an md5 is 32 hex
+# characters, and this file knows that where the shared helper only knows "not empty".
 agree() {  # $1 label, $2 predicate, $3 selected expression
 	local viaix viaseq
 	viaix=$(qset "SET enable_seqscan=off; SET enable_bitmapscan=off" \
@@ -105,11 +105,11 @@ agree() {  # $1 label, $2 predicate, $3 selected expression
 	              "SELECT md5(string_agg(t::text, ',' ORDER BY t))
 	                 FROM (SELECT $3 AS t FROM w WHERE $2) s")
 	if ! grep -qE '^[0-9a-f]{32}$' <<<"$viaix" || ! grep -qE '^[0-9a-f]{32}$' <<<"$viaseq"; then
-		check "$1 (both sides must be a real result, not empty)" \
+		check "$1 (both sides must be a real md5, not empty or an error)" \
 			"index=[$viaix] seq=[$viaseq]" "two md5 hashes"
 		return
 	fi
-	check "$1" "$viaix" "$viaseq"
+	check_text "$1" "$viaix" "$viaseq"
 }
 agree "plain index agrees with a sequential scan"   "k BETWEEN 1000 AND 9999"        "k"
 agree "late-column index agrees"                    "c18 > repeat('x',80)||'99000'"  "k"