Fixed race where insertion into implicit producer hash could accidentally skip a tombstoned slot, forcing an empty slot to be sacrificed for a reused producer, which in extreme cases can cause the hash to break its 0.75 load factor invariant (see #288)

Author: cameron314

concurrentqueue.h

@@ -3418,14 +3418,13 @@ class ConcurrentQueue
 						index = hashedId;
 						while (true) {
 							index &= mainHash->capacity - 1;
-							probedKey = mainHash->entries[index].key.load(std::memory_order_relaxed);
 							auto empty = details::invalid_thread_id;
 #ifdef MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED
 							auto reusable = details::invalid_thread_id2;
-							if ((probedKey == empty    && mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_relaxed, std::memory_order_relaxed)) ||
-								(probedKey == reusable && mainHash->entries[index].key.compare_exchange_strong(reusable, id, std::memory_order_acquire, std::memory_order_acquire))) {
+							if (mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_seq_cst, std::memory_order_relaxed) ||
+								mainHash->entries[index].key.compare_exchange_strong(reusable, id, std::memory_order_seq_cst, std::memory_order_relaxed)) {
 #else
-							if ((probedKey == empty    && mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_relaxed, std::memory_order_relaxed))) {
+							if (mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_seq_cst, std::memory_order_relaxed)) {
 #endif
 								mainHash->entries[index].value = value;
 								break;
@@ -3506,15 +3505,13 @@ class ConcurrentQueue
 				auto index = hashedId;
 				while (true) {
 					index &= mainHash->capacity - 1;
-					auto probedKey = mainHash->entries[index].key.load(std::memory_order_relaxed);
-					
 					auto empty = details::invalid_thread_id;
 #ifdef MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED
 					auto reusable = details::invalid_thread_id2;
-					if ((probedKey == empty    && mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_relaxed, std::memory_order_relaxed)) ||
-						(probedKey == reusable && mainHash->entries[index].key.compare_exchange_strong(reusable, id, std::memory_order_acquire, std::memory_order_acquire))) {
+					if (mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_seq_cst, std::memory_order_relaxed) ||
+						mainHash->entries[index].key.compare_exchange_strong(reusable, id, std::memory_order_seq_cst, std::memory_order_relaxed)) {
 #else
-					if ((probedKey == empty    && mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_relaxed, std::memory_order_relaxed))) {
+					if (mainHash->entries[index].key.compare_exchange_strong(empty,    id, std::memory_order_seq_cst, std::memory_order_relaxed)) {
 #endif
 						mainHash->entries[index].value = producer;
 						break;
@@ -3553,9 +3550,8 @@ class ConcurrentQueue
 			auto index = hashedId;
 			do {
 				index &= hash->capacity - 1;
-				probedKey = hash->entries[index].key.load(std::memory_order_relaxed);
-				if (probedKey == id) {
-					hash->entries[index].key.store(details::invalid_thread_id2, std::memory_order_release);
+				probedKey = id;
+				if (hash->entries[index].key.compare_exchange_strong(probedKey, details::invalid_thread_id2, std::memory_order_seq_cst, std::memory_order_relaxed)) {
 					break;
 				}
 				++index;