ringbuffer: Remove locking mutex before notify_all, leading to pessimistic re-block of notified wait() thread.

Holding the same lock @ notify_all as the waiting thread, would lead to waking up the waiting thread,
which only would be blocked again as it attempts to re-acquire the lock.

Hence removing it.

A lock is also not required for the readPos or writePos field,
as it is atomic of memory_order_seq_cst.

1 files changed, 12 insertions, 6 deletions

diff --git a/include/jau/ringbuffer.hpp b/include/jau/ringbuffer.hpp
index 3302b45..893dc37 100644
--- a/include/jau/ringbuffer.hpp
+++ b/include/jau/ringbuffer.hpp
@@ -481,7 +481,8 @@ class ringbuffer {
                 dtor_one( localReadPos );
             }
             {
-                std::unique_lock<std::mutex> lockRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Also not required as readPos is atomic of memory_order_seq_cst
                 readPos = localReadPos; // SC-DRF release atomic readPos
                 cvRead.notify_all(); // notify waiting putter
             }
@@ -576,7 +577,8 @@ class ringbuffer {
                 localReadPos = ( localReadPos + togo_count - 1 ) % capacityPlusOne; // last read-pos
             }
             {
-                std::unique_lock<std::mutex> locRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Also not required as readPos is atomic of memory_order_seq_cst
                 readPos = localReadPos; // SC-DRF release atomic readPos
                 cvRead.notify_all(); // notify waiting putter
             }
@@ -658,7 +660,8 @@ class ringbuffer {
                 localReadPos = ( localReadPos + togo_count - 1 ) % capacityPlusOne; // last read-pos
             }
             {
-                std::unique_lock<std::mutex> lockRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockRead(syncRead); // SC-DRF w/ putImpl via same lock
+                // Also not required as readPos is atomic of memory_order_seq_cst
                 readPos = localReadPos; // SC-DRF release atomic readPos
                 cvRead.notify_all(); // notify waiting putter
             }
@@ -696,7 +699,8 @@ class ringbuffer {
                 new (const_cast<pointer_mutable>(array + localWritePos)) value_type( std::move(e) ); // placement new
             }
             {
-                std::unique_lock<std::mutex> lockWrite(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockWrite(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Also not required as writePos is atomic of memory_order_seq_cst
                 writePos = localWritePos; // SC-DRF release atomic writePos
                 cvWrite.notify_all(); // notify waiting getter
             }
@@ -738,7 +742,8 @@ class ringbuffer {
                 new (const_cast<pointer_mutable>(array + localWritePos)) value_type( e ); // placement new
             }
             {
-                std::unique_lock<std::mutex> lockWrite(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockWrite(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Also not required as writePos is atomic of memory_order_seq_cst
                 writePos = localWritePos; // SC-DRF release atomic writePos
                 cvWrite.notify_all(); // notify waiting getter
             }
@@ -824,7 +829,8 @@ class ringbuffer {
                 localWritePos = ( localWritePos + togo_count - 1 ) % capacityPlusOne; // last write-pos
             }
             {
-                std::unique_lock<std::mutex> lockRead(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Pessimization (would re-block notified wait() thread): std::unique_lock<std::mutex> lockWrite(syncWrite); // SC-DRF w/ getImpl via same lock
+                // Also not required as writePos is atomic of memory_order_seq_cst
                 writePos = localWritePos; // SC-DRF release atomic writePos
                 cvWrite.notify_all(); // notify waiting getter
             }