Add a test for reappearing lost PVs causing endless metadata correction

updates. (A problem Milan fixed recently.)

test/t-unlost-pv.sh

@@ -0,0 +1,28 @@
+. ./test-utils.sh
+
+aux prepare_vg 3
+
+lvcreate -m 1 -l 1 -n mirror $vg
+lvchange -a n $vg/mirror
+
+check() {
+vgscan 2>&1 | tee vgscan.out
+grep "Inconsistent metadata found for VG $vg" vgscan.out
+vgscan 2>&1 | tee vgscan.out
+not grep "Inconsistent metadata found for VG $vg" vgscan.out
+}
+
+# try orphaning a missing PV
+disable_dev $dev1
+vgreduce --removemissing --force $vg
+enable_dev $dev1
+check
+
+exit 0 # FIXME, bug demonstration code follows
+# try to just change metadata; we expect the new version (with MISSING_PV set
+# on the reappeared volume) to be written out to the previously missing PV
+vgextend $vg $dev1
+disable_dev $dev1
+lvremove $vg/mirror
+enable_dev $dev1
+check