enough with umpteen "i18n.h" files. Consolidate on pbd/i18n.h
[ardour.git] / libs / pbd / event_loop.cc
index 67fa6cf49a5c132ecacbd9aca1ced2b60b8bdfc5..ea3f7a46afb5ae2b347967169c8ace80e69afc6a 100644 (file)
@@ -1,35 +1,93 @@
-#include <iostream>
+/*
+    Copyright (C) 2012 Paul Davis
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+*/
+
+#include <cstring>
+
+#include <pthread.h>
+
+#include "pbd/compose.h"
+#include "pbd/debug.h"
 #include "pbd/event_loop.h"
+#include "pbd/error.h"
 #include "pbd/stacktrace.h"
 
+#include "pbd/i18n.h"
+
 using namespace PBD;
 using namespace std;
 
-Glib::StaticPrivate<EventLoop> EventLoop::thread_event_loop;
-
 static void do_not_delete_the_loop_pointer (void*) { }
 
-EventLoop* 
-EventLoop::get_event_loop_for_thread() {
+Glib::Threads::Private<EventLoop> EventLoop::thread_event_loop (do_not_delete_the_loop_pointer);
+
+Glib::Threads::RWLock EventLoop::thread_buffer_requests_lock;
+EventLoop::ThreadRequestBufferList EventLoop::thread_buffer_requests;
+EventLoop::RequestBufferSuppliers EventLoop::request_buffer_suppliers;
+
+EventLoop::EventLoop (string const& name)
+       : _name (name)
+{
+}
+
+EventLoop*
+EventLoop::get_event_loop_for_thread()
+{
        return thread_event_loop.get ();
 }
 
-void 
-EventLoop::set_event_loop_for_thread (EventLoop* loop) 
+void
+EventLoop::set_event_loop_for_thread (EventLoop* loop)
 {
-       thread_event_loop.set (loop, do_not_delete_the_loop_pointer); 
+       thread_event_loop.set (loop);
 }
 
-/** Called when a sigc::trackable that was connected to using the invalidator() macro
- *  is destroyed.
- */
-void* 
+void*
 EventLoop::invalidate_request (void* data)
 {
         InvalidationRecord* ir = (InvalidationRecord*) data;
 
+       /* Some of the requests queued with an EventLoop may involve functors
+        * that make method calls to objects whose lifetime is shorter
+        * than the EventLoop's. We do not want to make those calls if the
+        * object involve has been destroyed. To prevent this, we
+        * provide a way to invalidate those requests when the object is
+        * destroyed.
+        *
+        * An object was passed to __invalidator() which added a callback to
+        * EventLoop::invalidate_request() to its "notify when destroyed"
+        * list. __invalidator() returned an InvalidationRecord that has been
+        * to passed to this function as data.
+        *
+        * The object is currently being destroyed and so we want to
+        * mark all requests involving this object that are queued with
+        * any EventLoop as invalid.
+        *
+        * As of April 2012, we are usign sigc::trackable as the base object
+        * used to queue calls to ::invalidate_request() to be made upon
+        * destruction, via its ::add_destroy_notify_callback() API. This is
+        * not necessarily ideal, but it is very close to precisely what we
+        * want, and many of the objects we want to do this with already
+        * inherit (indirectly) from sigc::trackable.
+        */
+
         if (ir->event_loop) {
-               Glib::Mutex::Lock lm (ir->event_loop->slot_invalidation_mutex());
+               Glib::Threads::Mutex::Lock lm (ir->event_loop->slot_invalidation_mutex());
                for (list<BaseRequestObject*>::iterator i = ir->requests.begin(); i != ir->requests.end(); ++i) {
                        (*i)->valid = false;
                        (*i)->invalidation = 0;
@@ -40,3 +98,136 @@ EventLoop::invalidate_request (void* data)
         return 0;
 }
 
+vector<EventLoop::ThreadBufferMapping>
+EventLoop::get_request_buffers_for_target_thread (const std::string& target_thread)
+{
+       vector<ThreadBufferMapping> ret;
+       Glib::Threads::RWLock::WriterLock lm (thread_buffer_requests_lock);
+
+       for (ThreadRequestBufferList::const_iterator x = thread_buffer_requests.begin();
+            x != thread_buffer_requests.end(); ++x) {
+
+               if (x->second.target_thread_name == target_thread) {
+                       ret.push_back (x->second);
+               }
+       }
+
+       DEBUG_TRACE (PBD::DEBUG::EventLoop, string_compose ("for thread \"%1\", found %2 request buffers\n", target_thread, ret.size()));
+
+       return ret;
+}
+
+void
+EventLoop::register_request_buffer_factory (const string& target_thread_name,
+                                            void* (*factory)(uint32_t))
+{
+
+       RequestBufferSupplier trs;
+       trs.name = target_thread_name;
+       trs.factory = factory;
+
+       {
+               Glib::Threads::RWLock::WriterLock lm (thread_buffer_requests_lock);
+               request_buffer_suppliers.push_back (trs);
+       }
+}
+
+void
+EventLoop::pre_register (const string& emitting_thread_name, uint32_t num_requests)
+{
+       /* Threads that need to emit signals "towards" other threads, but with
+          RT safe behavior may be created before the receiving threads
+          exist. This makes it impossible for them to use the
+          ThreadCreatedWithRequestSize signal to notify receiving threads of
+          their existence.
+
+          This function creates a request buffer for them to use with
+          the (not yet) created threads, and stores it where the receiving
+          thread can find it later.
+        */
+
+       ThreadBufferMapping mapping;
+       Glib::Threads::RWLock::WriterLock lm (thread_buffer_requests_lock);
+
+       for (RequestBufferSuppliers::iterator trs = request_buffer_suppliers.begin(); trs != request_buffer_suppliers.end(); ++trs) {
+
+               if (!trs->factory) {
+                       /* no factory - no request buffer required or expected */
+                       continue;
+               }
+
+               if (emitting_thread_name == trs->name) {
+                       /* no need to register an emitter with itself */
+                       continue;
+               }
+
+               mapping.emitting_thread = pthread_self();
+               mapping.target_thread_name = trs->name;
+
+               /* Allocate a suitably sized request buffer. This will set the
+                * thread-local variable that holds a pointer to this request
+                * buffer.
+                */
+               mapping.request_buffer = trs->factory (num_requests);
+
+               /* now store it where the receiving thread (trs->name) can find
+                  it if and when it is created. (Discovery happens in the
+                  AbstractUI constructor. Note that if
+               */
+
+               const string key = string_compose ("%1/%2", emitting_thread_name, mapping.target_thread_name);
+
+               /* management of the thread_request_buffers map works as
+                * follows:
+                *
+                * when the factory method was called above, the pointer to the
+                * created buffer is set as a thread-local-storage (TLS) value
+                * for this (the emitting) thread.
+                *
+                * The TLS value is set up with a destructor that marks the
+                * request buffer as "dead" when the emitting thread exits.
+                *
+                * An entry will remain in the map after the thread exits.
+                *
+                * The receiving thread may (if it receives requests from other
+                * threads) notice the dead buffer. If it does, it will delete
+                * the request buffer, and call
+                * ::remove_request_buffer_from_map() to get rid of it from the map.
+                *
+                * This does mean that the lifetime of the request buffer is
+                * indeterminate: if the receiving thread were to receive no
+                * further requests, the request buffer will live on
+                * forever. But this is OK, because if there are no requests
+                * arriving, the receiving thread is not attempting to use the
+                * request buffer(s) in any way.
+                *
+                * Note, however, that *if* an emitting thread is recreated
+                * with the same name (e.g. when a control surface is
+                * enabled/disabled/enabled), then the request buffer for the
+                * new thread will replace the map entry for the key, because
+                * of the matching thread names. This does mean that
+                * potentially the request buffer can leak in this case, but
+                * (a) these buffers are not really that large anyway (b) the
+                * scenario is not particularly common (c) the buffers would
+                * typically last across a session instance if not program
+                * lifetime anyway.
+                */
+
+               thread_buffer_requests[key] = mapping;
+               DEBUG_TRACE (PBD::DEBUG::EventLoop, string_compose ("pre-registered request buffer for \"%1\" to send to \"%2\", buffer @ %3 (key was %4)\n",
+                                                                   emitting_thread_name, trs->name, mapping.request_buffer, key));
+       }
+}
+
+void
+EventLoop::remove_request_buffer_from_map (void* ptr)
+{
+       Glib::Threads::RWLock::WriterLock lm (thread_buffer_requests_lock);
+
+       for (ThreadRequestBufferList::iterator x = thread_buffer_requests.begin(); x != thread_buffer_requests.end(); ++x) {
+               if (x->second.request_buffer == ptr) {
+                       thread_buffer_requests.erase (x);
+                       break;
+               }
+       }
+}