Try optimizing by acquiring RwLock fewer time

src/lib.rs

@@ -736,16 +736,27 @@ impl SystemHandle {
         &self,
         recipient: Recipient<M>,
     ) -> Result<(), SendError> {
+        let mut event_subscribers = self.event_subscribers.write();
+
         // Send the last cached value (if there is one)
-        if let Some(last_cached_value) =
-            self.event_subscribers.read().last_value_cache.get(&TypeId::of::<E>())
+        if let Some(last_cached_value) = event_subscribers.last_value_cache.get(&TypeId::of::<E>())
         {
             if let Some(msg) = last_cached_value.downcast_ref::<E>() {
                 recipient.send(msg.clone().into())?;
             }
         }
 
-        self.subscribe_recipient::<M, E>(recipient);
+        // Duplicate the contents of Self::subscribe_recipient() to reuse the same WriteGuard
+        // for better performance.
+        let subs = event_subscribers.events.entry(TypeId::of::<E>()).or_default();
+        subs.push(Box::new(move |e| {
+            if let Some(event) = e.downcast_ref::<E>() {
+                let msg = event.clone();
+                recipient.send(msg.into())?;
+            }
+
+            Ok(())
+        }));
 
         Ok(())
     }
@@ -1325,9 +1336,9 @@ mod tests {
 
         struct Subscriber;
         impl Actor for Subscriber {
-            type Message = ();
             type Context = Context<Self::Message>;
             type Error = ();
+            type Message = ();
 
             fn started(&mut self, context: &mut Self::Context) {
                 context