Execute cross-context events in correct increasing time order.

Propagate value changes / message sends breadth first instead of deptch first.


git-svn-id: http://svn.drobilla.net/lad/trunk/ingen@2282 a436a847-0d15-0410-975c-d299462d15a1

1 files changed, 36 insertions, 14 deletions

diff --git a/src/engine/MessageContext.cpp b/src/engine/MessageContext.cpp
index 5a180a10..6e08eb8d 100644
--- a/src/engine/MessageContext.cpp
+++ b/src/engine/MessageContext.cpp
@@ -32,16 +32,17 @@ namespace Ingen {
 
 
 void
-MessageContext::run(NodeImpl* node)
+MessageContext::run(NodeImpl* node, FrameTime time)
 {
 	if (ThreadManager::current_thread_id() == THREAD_PRE_PROCESS) {
 		assert(node);
 		Glib::Mutex::Lock lock(_mutex);
-		_request = node;
+		_non_rt_request = Request(time, node);
 		_sem.post();
 		_cond.wait(_mutex);
 	} else if (ThreadManager::current_thread_id() == THREAD_PROCESS) {
-		_requests.write(sizeof(NodeImpl*), &node);
+		Request r(time, node);
+		_requests.write(sizeof(Request), &r);
 		// signal() will be called at the end of this process cycle
 	} else if (ThreadManager::current_thread_id() == THREAD_MESSAGE) {
 		cout << "Message context recursion at " << node->path() << endl;
@@ -54,33 +55,55 @@ MessageContext::run(NodeImpl* node)
 void
 MessageContext::_run()
 {
-	NodeImpl* node = NULL;
+	Request req;
 
 	while (true) {
 		_sem.wait();
 
-		// Run a node requested by the pre-process thread
+		// Enqueue a request from the pre-process thread
 		{
 			Glib::Mutex::Lock lock(_mutex);
-			node = _request.get();
-			if (node) {
+			const Request req = _non_rt_request;
+			if (req.node) {
+				_queue.insert(req);
+				_end_time = std::max(_end_time, req.time);
 				_cond.broadcast(); // Notify caller we got the message
-				run_node(node);
 			}
 		}
 
-		// Run nodes requested by the audio thread
+		// Enqueue (and thereby sort) requests from audio thread
 		while (has_requests()) {
-			_requests.full_read(sizeof(NodeImpl*), &node);
-			run_node(node);
+			_requests.full_read(sizeof(Request), &req);
+			if (req.node) {
+				_queue.insert(req);
+			} else {
+				_end_time = req.time;
+				break;
+			}
+		}
+
+		// Run events in time increasing order
+		// Note that executing nodes may insert further events into the queue
+		while (!_queue.empty()) {
+			const Request req = *_queue.begin();
+
+			// Break if all events during this cycle have been consumed
+			// (the queue may contain generated events with later times)
+			if (req.time > _end_time) {
+				break;
+			}
+
+			_queue.erase(_queue.begin());
+			execute(req);
 		}
 	}
 }
 
 
 void
-MessageContext::run_node(NodeImpl* node)
+MessageContext::execute(const Request& req)
 {
+	NodeImpl* node = req.node;
 	node->message_run(*this);
 
 	void*      valid_ports = node->valid_ports();
@@ -94,8 +117,7 @@ MessageContext::run_node(NodeImpl* node)
 			for (PatchImpl::Connections::iterator c = wires.begin(); c != wires.end(); ++c) {
 				ConnectionImpl* ci = dynamic_cast<ConnectionImpl*>(c->get());
 				if (ci->src_port() == p) {
-					ci->dst_port()->pre_process(*_engine.message_context());
-					run_node(ci->dst_port()->parent_node());
+					_queue.insert(Request(req.time, ci->dst_port()->parent_node()));
 				}
 			}
 		}