fix: add thread safety primitives to internal shard list

.cspell.json

@@ -2,6 +2,7 @@
 	"version": "0.2",
 	"language": "en-GB",
 	"words": [
+		"merperson",
 		"EVFILT",
 		"fflags",
 		"udata",

include/dpp/cluster.h

@@ -152,6 +152,11 @@ class DPP_EXPORT cluster {
 	 */
 	std::shared_mutex named_commands_mutex;
 
+	/**
+	 * @brief Mutex for protection of shards list
+	 */
+	mutable std::shared_mutex shards_mutex;
+
 	/**
 	 * @brief Typedef for slashcommand handler type
 	 */
@@ -558,9 +563,9 @@ class DPP_EXPORT cluster {
 	/**
 	 * @brief Get the list of shards
 	 *
-	 * @return shard_list& Reference to map of shards for this cluster
+	 * @return shard_list map of shards for this cluster
 	 */
-	const shard_list& get_shards();
+	shard_list get_shards() const;
 
 	/**
 	 * @brief Sets the request timeout.

src/dpp/cluster.cpp

@@ -229,7 +229,11 @@ void cluster::start(start_type return_after) {
 				if (now >= shard_reconnect_time) {
 					/* This shard needs to be reconnected */
 					reconnections.erase(reconnect);
-					discord_client* old = shards[shard_id];
+					discord_client* old = nullptr;
+					{
+						std::shared_lock lk(shards_mutex);
+						old = shards[shard_id];
+					}
 					/* These values must be copied to the new connection
 					 * to attempt to resume it
 					 */
@@ -238,6 +242,7 @@ void cluster::start(start_type return_after) {
 					log(ll_info, "Reconnecting shard " + std::to_string(shard_id));
 					/* Make a new resumed connection based off the old one */
 					try {
+						std::unique_lock lk(shards_mutex);
 						if (shards[shard_id] != nullptr) {
 							log(ll_trace, "Attempting resume...");
 							shards[shard_id] = nullptr;
@@ -255,6 +260,7 @@ void cluster::start(start_type return_after) {
 						shards[shard_id]->run();
 					}
 					catch (const std::exception& e) {
+						std::unique_lock lk(shards_mutex);
 						log(ll_info, "Exception when reconnecting shard " + std::to_string(shard_id) + ": " + std::string(e.what()));
 						delete shards[shard_id];
 						delete old;
@@ -340,6 +346,7 @@ void cluster::start(start_type return_after) {
 				if (s % maxclusters == cluster_id) {
 					/* Each discord_client is inserted into the socket engine when we call run() */
 					try {
+						std::unique_lock lk(shards_mutex);
 						this->shards[s] = new discord_client(this, s, numshards, token, intents, compressed, ws_mode);
 						this->shards[s]->run();
 					}
@@ -455,6 +462,7 @@ void cluster::shutdown() {
 		next_timer = {};
 	}
 
+	std::unique_lock lk(shards_mutex);
 	/* Terminate shards */
 	for (const auto& sh : shards) {
 		delete sh.second;
@@ -581,6 +589,7 @@ void cluster::set_presence(const dpp::presence &p) {
 	}
 
 	json pres = p.to_json();
+	std::shared_lock lk(shards_mutex);
 	for (auto& s : shards) {
 		if (s.second->is_connected()) {
 			s.second->queue_message(s.second->jsonobj_to_string(pres));
@@ -610,15 +619,16 @@ std::string cluster::get_audit_reason() {
 }
 
 discord_client* cluster::get_shard(uint32_t id) const {
+	std::shared_lock lk(shards_mutex);
 	auto i = shards.find(id);
 	if (i != shards.end()) {
 		return i->second;
-	} else {
-		return nullptr;
 	}
+	return nullptr;
 }
 
-const shard_list& cluster::get_shards() {
+shard_list cluster::get_shards() const {
+	std::shared_lock lk(shards_mutex);
 	return shards;
 }