Writing generators for graphs with decompositions

Thanks for the suggestions.

Adding events for transitions would be different if I do it in the Watch example way compared to what jpt suggested. In the Watch example,  the Event role is connected to an object(button). However, in my case I have many different kinds of events which lead to a transition. Events from UI components and async events(10-15 kinds of). The events from UI components(only buttons currently) just happens to be present in another graph-"Screen Editor". Do I have to add a separate object for each class of events and associate then in the "Transition" relationship? Once I have defined the buttons in the Screen Editor graph, I want them to be available while I use the State editor graph to define the events for transitions.

I could not find the error log file in the working directory. Is it in the c:\program Files directory?

The semantics of a button event is the following: the value specifies the state to which the transition occurs. For example,

 <button label="SKIP" width="200" height="200" x="412" y="500" textsize="40" icon="none" adjtext="30">
                <event name="clicked">
                    <action preconditions="no" name="transition">
                        <parameter>
                            <type>state</type>
                            <name>n</name>
                            <value>4192</value>
                        </parameter>
                    </action>
                </event>
        </button>
        
There are timeout events, for example:
<event name="TimeOut">
            <action preconditions="no" name="transition">
                <parameter>
                    <type>state</type>
                    <name>n</name>
                    <value>4172</value>
                </parameter>
            </action>
        </event>
There are async events which also cause transition, for example :
      <event name="PulseCoreStatus" arg1="F1">
        <action preconditions="no" name="transition">
          <parameter>
            <type>state</type>
            <name>n</name>
            <value>4181</value>
          </parameter>
        </action>
      </event>       
In the current design there are also background actions(don't know what these are called in Statechart terminology but probably similar to the concept of onEnter) which are parametrized, for example:
    <backgroundactions>
      <rocosmessage name="FDSessionStart" type="EVENT_MESSAGE" senderid="RCP" receiverid="NA">
        <parameter type="unsigned int" name="id" vartype="text">1</parameter>
        </rocosmessage>
    </backgroundactions>

A complete state example is here : http://pastebin.com/DnMK6rmR

I'm not clear on how to write the generator for the visual syntax the modeler is going to describe. My current schema seems quite complex for what I'm trying to do after all your suggestions.    I will be very interested if you have some refactoring suggestions for my xml schema.

This is my current simple state machine description which doesn't include clustering states,orthogonality of states or refinement of states. I'm trying to scale up my design ( and the xml schema) to avoid exponential blow-up of states using Harel statechart visual notation. For meeting my requirements, i also need to extend the formal semantics for statechart as described in Harel's paper (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.131.7628&rep=rep1&type=pdf). Essentially w3c scxml doesn't yet have a domain specific requirement that I have and is good for describing a generic statechart.
Reading through the original paper, I found that your watch example is similar to the Citizen watch example Harel described in his work.
The major difference to Harel's statechart solution in my requirement is that events carry parameters and I need a dispatching mechanism for events and handling of actions.


Exploring on this front, I found that Adobe's mxml(help.adobe.com/en_US/flex/using/WS2db454920e96a9e51e63e3d11c0bf69084-7fae.html) is very close to reusing widgets in states to avoid state explosion. So, the semantics is something like if you define an EXIT button, you can includeIn s1,s2,s3. But in my state design, EXIT button and the associated event has to be mentioned in every state definition where its required. However, mxml doesn't allow me to specify the other async events and actions which are not associated with UI elements in their schema. That is done in actionscript3 code.


Edited by iceman - 22.Jul.2011 at 13:53

stevek wrote: Let's step back for a second and look at the wider picture. You have states, transitions and events. You have screens with buttons and texts. Some events are caused by buttons, some are asynchronous. You want a graphical language for the state machine part, and another for the screen part. And you are making an XML schema.

Thanks indeed for summarizing. Thats exactly what I want, but the meta-representation is internal and not necessarily xml

Thanks Steven,
There is a lot of information in this thread of discussion which can be used to drastically improve my design and handle higher complexity as I have pointed out, and possible solutions you have suggested. I'm planning to solve the problems step-by-step.
The first is to start with simple FSM example and generate the exact same xml (http://pastebin.com/DnMK6rmR). The domain model (and writing the generator) for that might be more complicated and not pertaining to the best practices, but it will be useful for a reference step for improvement of the model. Wondering, how to write the generator for that. I'm working on this now.

The next one is to define models according to your suggestions for multiple events triggering same transition(and to take care of state explosion) as in the last post. For that next step, what representation other than xml can be used?