Incorrect out-of-memory errors being reported

[dirkfeytons]

Similar to #2525 I encountered a situation where incorrectly an out-of-memory was reported: a client connected to the Mosquitto broker, published a message and disconnected without properly waiting for the puback. The broker would then log a message saying the client disconnected due to out-of-memory while the correct error is "broken pipe".

I can fix this with a patch like the following (against 2.0.15):

diff --git a/src/handle_publish.c b/src/handle_publish.c
index 111f0316..8736812d 100644
--- a/src/handle_publish.c
+++ b/src/handle_publish.c
@@ -332,9 +332,9 @@ int handle__publish(struct mosquitto *context)
                        rc2 = sub__messages_queue(context->id, stored->topic, stored->qos, stored->retain, &stored);
                        /* stored may now be free, so don't refer to it */
                        if(rc2 == MOSQ_ERR_SUCCESS || context->protocol != mosq_p_mqtt5){
-                               if(send__puback(context, mid, 0, NULL)) rc = 1;
+                               rc = send__puback(context, mid, 0, NULL);
                        }else if(rc2 == MOSQ_ERR_NO_SUBSCRIBERS){
-                               if(send__puback(context, mid, MQTT_RC_NO_MATCHING_SUBSCRIBERS, NULL)) rc = 1;
+                               rc = send__puback(context, mid, MQTT_RC_NO_MATCHING_SUBSCRIBERS, NULL);
                        }else{
                                rc = rc2;
                        }

While this seems to fix my particular scenario with minimal changes I think more improvements can be done: it looks to me like rc2 can be eliminated (in the above piece of code at least). And if I look further in this file I think there are other cases where rc = 1 is done instead of propagating the return value of a helper function.
Grepping the code for rc = 1; reveals some other places that seem to me to warrant further investigation.