Make the mutual recursion mechanism safer

By directly stopping the IO context there was a chance that a task would
get cancelled outright if all stars aligned in the wrong way. Stopping
the IO context could happen between posting the task to the context and
waiting for it. This approach is much safer as we cannot drop any work
this way.

src/plugin/bridges/vst3-impls/plug-view-proxy.h

@@ -235,6 +235,11 @@ class Vst3PlugViewProxyImpl : public Vst3PlugViewProxy {
             mutual_recursion_contexts.push_back(current_io_context);
         }
 
+        // Instead of directly stopping the IO context, we'll reset this work
+        // guard instead. This prevents us from accidentally cancelling any
+        // outstanding tasks.
+        auto work_guard = boost::asio::make_work_guard(*current_io_context);
+
         // We will call the function from another thread so we can handle calls
         // to  from this thread
         std::promise<TResponse> response_promise{};
@@ -242,9 +247,11 @@ class Vst3PlugViewProxyImpl : public Vst3PlugViewProxy {
             const TResponse response = bridge.send_message(object);
 
             // Stop accepting additional work to be run from the calling thread
-            // once we receive a response
+            // once we receive a response. By resetting the work guard we do not
+            // cancel any pending tasks, but `current_io_context->run()` will
+            // stop blocking eventually.
             std::lock_guard lock(mutual_recursion_contexts_mutex);
-            current_io_context->stop();
+            work_guard.reset();
             mutual_recursion_contexts.erase(
                 std::find(mutual_recursion_contexts.begin(),
                           mutual_recursion_contexts.end(), current_io_context));
@@ -254,7 +261,6 @@ class Vst3PlugViewProxyImpl : public Vst3PlugViewProxy {
 
         // Accept work from the other thread until we receive a response, at
         // which point the context will be stopped
-        auto work_guard = boost::asio::make_work_guard(*current_io_context);
         current_io_context->run();
 
         return response_promise.get_future().get();

src/wine-host/bridges/vst3.h

@@ -263,6 +263,11 @@ class Vst3Bridge : public HostBridge {
             mutual_recursion_contexts.push_back(current_io_context);
         }
 
+        // Instead of directly stopping the IO context, we'll reset this work
+        // guard instead. This prevents us from accidentally cancelling any
+        // outstanding tasks.
+        auto work_guard = boost::asio::make_work_guard(*current_io_context);
+
         // We will call the function from another thread so we can handle calls
         // to  from this thread
         std::promise<TResponse> response_promise{};
@@ -270,9 +275,11 @@ class Vst3Bridge : public HostBridge {
             const TResponse response = send_message(object);
 
             // Stop accepting additional work to be run from the calling thread
-            // once we receive a response
+            // once we receive a response. By resetting the work guard we do not
+            // cancel any pending tasks, but `current_io_context->run()` will
+            // stop blocking eventually.
             std::lock_guard lock(mutual_recursion_contexts_mutex);
-            current_io_context->stop();
+            work_guard.reset();
             mutual_recursion_contexts.erase(
                 std::find(mutual_recursion_contexts.begin(),
                           mutual_recursion_contexts.end(), current_io_context));
@@ -282,7 +289,6 @@ class Vst3Bridge : public HostBridge {
 
         // Accept work from the other thread until we receive a response, at
         // which point the context will be stopped
-        auto work_guard = boost::asio::make_work_guard(*current_io_context);
         current_io_context->run();
 
         return response_promise.get_future().get();