Advanced Gtk+ Sequencer - News
Refactored key c source files
Item posted by Joël Krähemann <jkraehemann> on Thu 19 Apr 2018 06:49:50 PM UTC.
Wow, it is really a lot of work to rework the entire code base. In case you missed it, here is what I do for short:
- make all objects thread-safe "the user" interacts with
- replace existing locks looked-up on the mutex manager by per-class locks
Most objects get additional fields like:
pthread_mutex_t *obj_mutex;
pthread_mutexattr_t *obj_mutexattr;
You should be able to access these fields as having the per-class lock. E.g. the newly introduced AgsPluginPort has following function:
pthread_mutex_t* ags_plugin_port_get_class_mutex();
In theory you only access the obj_mutex field as having the class mutex lock. For the plugin port it would look like this:
pthread_mutex_t *plugin_port_mutex;
plugin_port = AGS_PLUGIN_PORT(gobject);
/* get base plugin mutex */
pthread_mutex_lock(ags_plugin_port_get_class_mutex());
plugin_port_mutex = plugin_port->obj_mutex;
pthread_mutex_unlock(ags_plugin_port_get_class_mutex());
The code above is from ags_plugin_port_get_property() function. Note, strings should basically been duplicated as it is just a pointer to a memory location.
Thread-safe key objects
So far, I reworked the key source files:
- ags_audio.c
- ags_channel.c
- ags_recycling.c
- ags_audio_signal.c
- ags_recall.c
- ags_recall_factory.c
- ...
Since the objects inheriting of AgsRecall need to lock the object, too. There is still much to do. Further AgsRecall got better semantics of processing stages. The process first/last flags was just removed because it is not needed, anymore.
There are some considerations that the engine won't perform as desired. However, the advantage of more intuitive locking mechanism is evident.
The situation in 1.4.x is basically magic. This shall change.
by Joël
Powered by Savane 3.13-ec04.
Corresponding source code