On 23.6.2011. 22:04, Christophe Henry wrote:
In our project, we need a generic interface to refer to different state machine (back-end). For example, in the following codes, we want
to use SMBase* to refer to any concrete state machine. SMBase should have similar interface as state machine.
struct User { A(SMBase* s) : sm(s) (); SMBase* sm; ... // User can use sm->start(), sm->process_event() etc }
struct SMBase { virtual void start() = 0; ... // how to design the rest ??? }
The problem of designing SMBase is the function "process_event". We want it to to be virtual function so SMWrapper can implement it. >But it also needs to be a template according to state machine interface. But we cannot have a virtual template function.
How can we achieve/design this generic state machine interface?
Hi Jerry,
Yes, virtual template methods are not possible in C++ :(
(Note to C++ Standard Comittee: I'd consider donating a kidney for this one :) )
This leaves you with few possibilies. I see 2 at a first glance (suggestions welcome):
- virtual void process_event_event1(); virtual void process_event_event2(); ...etc (not fun but quite in the OO spirit)
- virtual void process_event(boost::any) followed by any_casts until you find the correct one (not fun either)
All considered, solution 1 looks better. If you can define a common interface for state machines, then they are bound to expect common events.
I have a combination of the above two which is quite fun :) Use boost.variant on all event types:
typedef boost::variant<ev1, ev2, ev3> event_variant
virtual void process_event( event_variant e ) = 0
and to use its apply_visitor function in derived class to force MSM to instantiate all needed process_event templates (untested):
template<typename SM> struct ProcessEvent { ProcessEvent( SM & sm ) : sm_( sm ) {} template<typename event> void operator( event & e ) { sm_.process_event( e ); }
SM & sm_; };
template <typename SM> class SMWrapper : public SMBase { virtual void process_event( event_variant ev ) { boost::variant::apply_visitor( ev, ProcessEvent<SM>( sm_ ) ); } };
Ah yes I didn't think about this solution, it's clearly more elegant. I need to give it a try. Good one :) Thanks, Christophe