xcb: lock-free event processing

For details how this works refer to the documentation in the patch.

The follow-up patches will switch to calling processXcbEvents() on every
event loop iteration. With the existing code that would mean frequent
locking of shared data (event queue). Acquiring a lock is fast, but
lock contention isn't. To avoid potential problems, reimplement xcb event
processing to be lock-free. Besides theoretical performance benefits,
this definitally improves code readability in qxcbconnection.cpp. Thanks
to Mikhail Svetkin for questioning the design of the existing code.

Done-with: Mikhail Svetkin <mikhail.svetkin@qt.io>
Change-Id: I935f2b6ca802580f5c80205aef7b2f9afc172d26
Reviewed-by: Mikhail Svetkin <mikhail.svetkin@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

1 files changed, 2 insertions, 1 deletions

diff --git a/src/plugins/platforms/xcb/qxcbintegration.cpp b/src/plugins/platforms/xcb/qxcbintegration.cpp
index db8f816a86..b1a1ed93cd 100644
--- a/src/plugins/platforms/xcb/qxcbintegration.cpp
+++ b/src/plugins/platforms/xcb/qxcbintegration.cpp
@@ -46,6 +46,7 @@
 #include "qxcbbackingstore.h"
 #include "qxcbnativeinterface.h"
 #include "qxcbclipboard.h"
+#include "qxcbeventqueue.h"
 #if QT_CONFIG(draganddrop)
 #include "qxcbdrag.h"
 #endif
@@ -344,7 +345,7 @@ QAbstractEventDispatcher *QXcbIntegration::createEventDispatcher() const
 {
     QAbstractEventDispatcher *dispatcher = createUnixEventDispatcher();
     for (int i = 0; i < m_connections.size(); i++)
-        m_connections[i]->eventReader()->registerEventDispatcher(dispatcher);
+        m_connections[i]->eventQueue()->registerEventDispatcher(dispatcher);
     return dispatcher;
 }