Difference between revisions of "Slicer3:EventBroker"

From NAMIC Wiki
Jump to: navigation, search
Line 51: Line 51:
 
* [http://java.sun.com/products/jms/javadoc-102a/index.html Java Message Service (JMS) API]
 
* [http://java.sun.com/products/jms/javadoc-102a/index.html Java Message Service (JMS) API]
 
* [http://en.wikipedia.org/wiki/Observer_pattern Wikipedia definition of Observer Pattern]
 
* [http://en.wikipedia.org/wiki/Observer_pattern Wikipedia definition of Observer Pattern]
* [http://xlobject.sourceforge A C++ implementation]
+
* [http://xlobject.sourceforge.net/ A C++ implementation]
 
* [http://sigslot.sourceforge.net/ Another C++ implementation]
 
* [http://sigslot.sourceforge.net/ Another C++ implementation]
 
* [http://doc.trolltech.com/4.3/signalsandslots.html The Qt implementation]
 
* [http://doc.trolltech.com/4.3/signalsandslots.html The Qt implementation]

Revision as of 16:56, 14 January 2008

Home < Slicer3:EventBroker

Currently

The basic idea of the EventBroker is that currently we have a lot of this kind of code in GUIs:

node->AddObserver(vtkCommand::ModifiedEvent, callbackCommand)

Problems

The problems with this:

  • node 'owns' the observer, but the callbackCommand is opaque so it doesn't know anything about what will happen when the event is invoked
  • the GUI needs to explicitly remove the observer before it is destroyed
  • node is not introspectable; you cannot get a list of observers on the node
  • there's no easy way to know what side effects will happen for any Set call (either a priori or experimentally).
  • there's no way to collapse events or disable them

Goals for Solution

So the EventBroker would be a singleton, perhaps owned by the ApplicationLogic that would look something like:

broker->RegisterObserver(node, vtkCommand::ModifiedEvent, this, callbackCommand);

The broker would do the following:

  • add DeleteEvent observers to both node and this so it can remove the observer automatically if either side is destroyed
  • keep an introspectable list of all observers it knows about
  • have an option to keep a log of all event invocations for debugging and performance analysis
  • have an option to turn off all event invocations
  • have an option to queue all event invocations and invoke them later
  • have methods to collapse redundant events in the queue
  • perhaps have method to pass event invocations from a processing thread to the main GUI thread?


Additional possible extensions:

  • rather than maintaining a distinct queue, the broker could queue events into the GUI event queue
  • the event queue could be protected by a mutex lock so that multiple threads can access the MRML scene in parallel but only the GUI thread talks to the display

References