[Lldb-commits] [lldb] [lldb] Improve locking in PathMappingLists (NFC) (PR #114576)

[Jonas Devlieghere via lldb-commits]

================
@@ -48,7 +48,10 @@ PathMappingList::PathMappingList(const PathMappingList &rhs)
 
 const PathMappingList &PathMappingList::operator=(const PathMappingList &rhs) {
   if (this != &rhs) {
-    std::scoped_lock<std::recursive_mutex, std::recursive_mutex> locks(m_mutex, rhs.m_mutex);
+    std::scoped_lock<std::mutex, std::mutex> pairs_locks(m_pairs_mutex,
+                                                         rhs.m_pairs_mutex);
+    std::scoped_lock<std::mutex, std::mutex> callback_locks(
+        m_callback_mutex, rhs.m_callback_mutex);
----------------
JDevlieghere wrote:

Right, it took me a while to convince myself that wouldn't just cause the same problem. The only way to avoid a deadlock is always acquire the lock in the same order and because we always acquire the `m_callback_mutex` before doing callback, and that's the only one that could be locked while acquiring the `m_pairs_mutex` (i.e. there's no code path besides that and `operator= ` that acquires both locks), that both mutex are always acquired in order.

https://github.com/llvm/llvm-project/pull/114576