Uphold/StopWhenUnneeded/BindsTo: requeue when job finishes

When a unit is upheld and fails, and there are no state changes in
the upholder, it will not be retried, which is against what the
documentation suggests.

Requeue when the job finishes. Same for the other two queues.

src/core/job.c

@@ -1051,6 +1051,12 @@ finish:
                         job_add_to_gc_queue(other->job);
                 }
 
+        /* Ensure that when an upheld/unneeded/bound unit activation job fails we requeue it, if it still
+         * necessary. If there are no state changes in the triggerer, it would not be retried otherwise. */
+        unit_submit_to_start_when_upheld_queue(u);
+        unit_submit_to_stop_when_bound_queue(u);
+        unit_submit_to_stop_when_unneeded_queue(u);
+
         manager_check_finished(u->manager);
 
         return 0;

test/units/testsuite-57-retry-fail.service

@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+[Unit]
+Description=Failed Dependency Unit
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=/bin/sh -c "if [ -f /tmp/testsuite-57-retry-fail ]; then exit 0; else exit 1; fi"
+Restart=no

test/units/testsuite-57-retry-upheld.service

@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+[Unit]
+Description=Upheld Unit
+Requires=testsuite-57-retry-fail.service
+After=testsuite-57-retry-fail.service
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=/bin/echo ok

test/units/testsuite-57-retry-uphold.service

@@ -0,0 +1,7 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+[Unit]
+Description=Upholding Unit
+Upholds=testsuite-57-retry-upheld.service
+
+[Service]
+ExecStart=/bin/sleep infinity

test/units/testsuite-57.sh

@@ -26,6 +26,33 @@ done
 
 systemctl stop testsuite-57-uphold.service
 
+# Idea is this:
+#    1. we start testsuite-57-retry-uphold.service
+#    2. which through Uphold= starts testsuite-57-retry-upheld.service
+#    3. which through Requires= starts testsuite-57-retry-fail.service
+#    4. which fails as /tmp/testsuite-57-retry-fail does not exist, so testsuite-57-retry-upheld.service
+#       is no longer restarted
+#    5. we create /tmp/testsuite-57-retry-fail
+#    6. now testsuite-57-retry-upheld.service will be restarted since upheld, and its dependency will
+#       be satisfied
+
+rm -f /tmp/testsuite-57-retry-fail
+systemctl start testsuite-57-retry-uphold.service
+
+while ! systemctl is-failed testsuite-57-retry-fail.service ; do
+    sleep .5
+done
+
+systemctl is-active testsuite-57-retry-upheld.service && { echo 'unexpected success'; exit 1; }
+
+touch /tmp/testsuite-57-retry-fail
+
+while ! systemctl is-active testsuite-57-retry-upheld.service ; do
+    sleep .5
+done
+
+systemctl stop testsuite-57-retry-uphold.service testsuite-57-retry-fail.service testsuite-57-retry-upheld.service
+
 # Idea is this:
 #    1. we start testsuite-57-prop-stop-one.service
 #    2. which through Wants=/After= pulls in testsuite-57-prop-stop-two.service as well