Fixing asan error

Author: MaciejKaszynski

score/launch_manager/daemon/src/common/concurrency/BUILD

@@ -49,11 +49,6 @@ cc_library(
 cc_test(
     name = "mpmc_concurrent_queue_test",
     srcs = ["mpmc_concurrent_queue_test.cpp"],
-    # TODO(eclipse-score/lifecycle#241): UBSan reports a misaligned reference while this
-    # test fixture is constructed on ASan's fake stack; the queue itself uses
-    # 64-byte-aligned members, so the report is a fake-stack alignment false
-    # positive rather than a queue bug.
-    tags = ["no-asan"],
     visibility = ["//tests:__subpackages__"],
     deps = [
         ":mpmc_concurrent_queue",

score/launch_manager/daemon/src/common/concurrency/mpmc_concurrent_queue.hpp

@@ -273,22 +273,36 @@ class MPMCConcurrentQueue
         return score::cpp::blank{};
     }
 
+    /// @brief Helper type to align members to the cache lines.
+    /// @details Using padding rather than alignas(CacheLineSize) so that 
+    ///          ASan's fake stack doesn't cause UBSan misalignment.
+    template <class Atomic>
+    struct CacheLinePaddedAtomic : public std::atomic<Atomic>
+    {
+        static_assert(sizeof(std::atomic<Atomic>) < CacheLineSize,
+                      "atomic<Atomic> is too large to pad to one cache line");
+        using std::atomic<Atomic>::atomic;
+
+      private:
+        char _pad[CacheLineSize - sizeof(std::atomic<Atomic>)]{};
+    };
+
     /// @brief Underlying storage.
     std::array<Slot, Capacity> m_slots;
 
     /// @brief The front of the queue; claimed by consumers via fetch_add in pop.
     /// @details Aligned so that m_head and m_tail do not share a cache line.
-    alignas(CacheLineSize) std::atomic<std::size_t> m_head{0};
+    CacheLinePaddedAtomic<std::size_t> m_head{0};
 
     /// @brief The back of the queue; claimed by producers via fetch_add in push_impl.
     /// @details Aligned so that m_head and m_tail do not share a cache line.
-    alignas(CacheLineSize) std::atomic<std::size_t> m_tail{0};
+    CacheLinePaddedAtomic<std::size_t> m_tail{0};
 
     /// @brief Set to true by stop(); causes push() to return false and pop() to
     ///        return std::nullopt instead of blocking.
     /// @details Aligned on its own cache line so that the single stop() write
     ///          does not cause false sharing with m_tail updates in push_impl().
-    alignas(CacheLineSize) std::atomic<bool> m_stopped{false};
+    CacheLinePaddedAtomic<bool> m_stopped{false};
 
     /// @brief Counts items currently in the queue; consumers block on this when
     ///        the queue is empty.