Tim Bunce <[EMAIL PROTECTED]> wrote: >On Thu, Jul 25, 2002 at 01:01:59PM -0500, James G Smith wrote: >> My questions for now are two: >> (1) should I place the state machine outside the general application >> namespace and, if so, what are some suggestions for where to put >> it? I am trying to place general modules in general locations. > >Tricky. We don't have a namespace for state machines and the 'control flow >utilities' section of the module list is a mish-mash of modules: > http://search-beta.cpan.org/modlist/Control_Flow_Utilities > >Searching on state machine is interesting: > http://search-beta.cpan.org/search?mode=all&query=state+machine >but offers little practical help. It seems that state machine modules >have always been closely tied to other modules (POE::Session, >Bio::Tools::StateMachine::AbstractStateMachine, Win32::CtrlGUI::State) > >Is there any chance you could reuse/extend/subclass POE::Session?
Arthur Bergman raised a similar question, which I will try to answer here also. I did look at POE and (today, looked again) at the docs for POE::Session and POE::NFA. POE seems to assume that I know what state I want next, or at least what event to call next. But I don't. All I have are the previous state of the machine and a bunch of data (in this case, from a web form). I need to use the data to find the next state and then transition to that state, calling any code that might be associated with that transition. Basically, once the object is set up, you feed it any existing context (e.g., existing state) and a hash of data. Based on the hash and the existing state, it will attempt to transition to a new state, producing a new context and a new hash of (possibly munged) data. The state machine in this case is a passive beast (though images of ribosomes come to mind, which aren't at all passive). I also need to be able to sub-class and inherit any transitions, states, and associated code in the process. I'm not finding this likely with POE, at least not without writing a lot of code, which I already have (though not involving POE) -- then POE becomes a bit heavy just to track state. >Otherwise perhaps StateMachine::Gestinanna would be a reasonable >name (showing the association with the Gestinanna framework) and >distribute it as a separate module - if it really is separate. It doesn't access any other modules in the Gestinanna:: namespace. It doesn't assume it is running in a web environment (I think it might be possible to use it in a Perl/Tk environment, but haven't looked yet -- of course, I'd love to use the xml views to create Tk dialogs :) I think it would be reasonable to make it a separate distribution. Another possible name would be StateMachine::Wizard, since it is designed to drive wizard-like applications. That might interfere with the Wizard:: namespace though if Wiedmann ever wanted to move part of his Wizard::State code to the StateMachine:: namespace. -- James Smith <[EMAIL PROTECTED]>, 979-862-3725 Texas A&M CIS Operating Systems Group, Unix
