State Machine Implementation

[nilavya]

HI,

I have got a C++ application involving State machine.

Now I have got too many states and too many events which changes the
state of the Staet machine. So my code is somewhat unmanageable. It is
:
switch(nState)
{
case state1:
{
switch(nEvent)
{
case event1:
break;
case event2:
break;
case event3:
default:
break;
}
}
break;
case state2:
break;
default:
break;
}

Now the problem I have is that some of the handling in various
states-events combo is almost same. That makes my code redundant.Also
the function size is toooo large.
Is there any way or design to write this code. I heard about writing a
seperate function for each state. But not sure???

Can Anyone help me in this....


Thanks.
With Regards,
Bhagat Nirav K.

[anderberg]

Re: State Machine Implementation

On 2005-12-07, nilavya wrote:[color=blue]
> HI,
>
> I have got a C++ application involving State machine.
>
> Now I have got too many states and too many events which changes the
> state of the Staet machine. So my code is somewhat unmanageable. It is[/color]

[... switch/case ...]

Maybe a search for "object oriented state machine" may give some
ideas?

--
anderberg: People are just mutations. Some worse than others.

[EventHelix.com]

Re: State Machine Implementation

Use a Hierarchical State Machine. Checkout the following link:

http://www.eventhelix.com/RealtimeMa...ateMachine.htm

--
EventStudio System Designer 2.5 - http://www.EventHelix.com/EventStudio
Sequence Diagram Based System Design and Object Modeling Tool

[mlimber]

Re: State Machine Implementation

nilavya wrote:[color=blue]
> HI,
>
> I have got a C++ application involving State machine.
>
> Now I have got too many states and too many events which changes the
> state of the Staet machine. So my code is somewhat unmanageable. It is
> :
> switch(nState)
> {
> case state1:
> {
> switch(nEvent)
> {
> case event1:
> break;
> case event2:
> break;
> case event3:
> default:
> break;
> }
> }
> break;
> case state2:
> break;
> default:
> break;
> }
>
> Now the problem I have is that some of the handling in various
> states-events combo is almost same. That makes my code redundant.Also
> the function size is toooo large.
> Is there any way or design to write this code. I heard about writing a
> seperate function for each state. But not sure???
>
> Can Anyone help me in this....
>
>
> Thanks.
> With Regards,
> Bhagat Nirav K.[/color]

Consider the State pattern from _Design Patterns_ by Gamma et al.

Cheers! --M

[Tim Clacy]

Re: State Machine Implementation

nilavya wrote:[color=blue]
> HI,
>
> I have got a C++ application involving State machine.
>
> Now I have got too many states and too many events which changes the
> state of the Staet machine. So my code is somewhat unmanageable. It is[color=green]
>>[/color]
> switch(nState)
> {
> case state1:
> {
> switch(nEvent)
> {
> case event1:
> break;
> case event2:
> break;
> case event3:
> default:
> break;
> }
> }
> break;
> case state2:
> break;
> default:
> break;
> }
>
> Now the problem I have is that some of the handling in various
> states-events combo is almost same. That makes my code redundant.Also
> the function size is toooo large.
> Is there any way or design to write this code. I heard about writing a
> seperate function for each state. But not sure???
>
> Can Anyone help me in this....
>
>
> Thanks.
> With Regards,
> Bhagat Nirav K.[/color]

Try and make the machine look like a table of data, since that is what it
is. For a given state:

switch (event)
{
case ev1: action1; nextstate1; break;
case ev2: action2; nextstate2; break;
:
case evn: actionn; nextstaten; break;
}

You might be able to make the states hierarchical and use one of the many
Harrel /UML statechart patterns (boost, quantum framework):

This is quite a clean and simple implementation:

A Lightweight Implementation of UML Statecharts in C++

http://www.codeproject.com/samples/Statechart.asp

Alternatively, there are 'C' / C++ code generators for state machines that
use a GUI front end (e.g. IAR VisualSTATE).

[Neil Cerutti]

Re: State Machine Implementation

On 2005-12-07, nilavya <nilavya@gmail. com> wrote:[color=blue]
> HI,
>
> I have got a C++ application involving State machine.
>
> Now I have got too many states and too many events which
> changes the state of the Staet machine. So my code is somewhat
> unmanageable. It is:
>
> switch(nState)
> {
> case state1:
> {
> switch(nEvent)
> {
> case event1:
> break;
> case event2:
> break;
> case event3:
> default:
> break;
> }
> }
> break;
> case state2:
> break;
> default:
> break;
> }
>
> Now the problem I have is that some of the handling in various
> states-events combo is almost same. That makes my code
> redundant.[/color]

Any part of the code that's truly redundant can be moved to a
function that all the states can rely on.
[color=blue]
> Also the function size is toooo large. Is there any way or
> design to write this code. I heard about writing a seperate
> function for each state. But not sure???[/color]

You can define each state as a function, yes.

What you need is a function type that returns a pointer
to function of it's own type. This is not possible to
declare directly in C++.

Here's a finite state machine, which is implemented using
functions, that recognizes the word "cat". A struct is used to
simulate a recursive function type.

#include <iostream>

struct state_t;

typedef state_t state(char c);

struct state_t
{
state_t(state* p): process(p) { }
state *function;
};

state_t start(char);
state_t got_c(char);
state_t got_a(char);
state_t terminal(char);

state_t start(char c)
{
if (c == 'c') return state_t(got_c);
else return state_t(start);
}

state_t got_c(char c)
{
if (c == 'a') return state_t(got_a);
else if (c == 'c') return state_t(got_c);
else return state_t(start);
}

state_t got_a(char c)
{
if (c == 't') return state_t(termina l);
else if (c == 'c') return state_t(got_c);
else return state_t(start);
}

state_t terminal(char)
{
return state_t(0);
}

int main ()
{
char m[]="The fat cat sits on the hat.";
state_t machine = start;
int i;
int len = strlen(m);
for (i = 0; i < len && machine.functio n != terminal; ++i) {
machine = machine.functio n(m[i]);
}
if (i != len) {
std::cout << "found \"cat\" at " << i-2 << '\n';
} else {
std::cout << "\"cat\" not found.\n";
}
return 0;
}

It might be more efficient on your implementation to use generic
function pointers to simulate the recursive return type. I didn't
do that here since I find it more confusing.

--
Neil Cerutti

[Neil Cerutti]

Re: State Machine Implementation

On 2005-12-07, Neil Cerutti <leadvoice@emai l.com> wrote:[color=blue]
> On 2005-12-07, nilavya <nilavya@gmail. com> wrote:[color=green]
>> HI,
>>
>> I have got a C++ application involving State machine.
>>
>> Now I have got too many states and too many events which
>> changes the state of the Staet machine. So my code is somewhat
>> unmanageable. It is:
>>
>> switch(nState)
>> {
>> case state1:
>> {
>> switch(nEvent)
>> {
>> case event1:
>> break;
>> case event2:
>> break;
>> case event3:
>> default:
>> break;
>> }
>> }
>> break;
>> case state2:
>> break;
>> default:
>> break;
>> }
>>
>> Now the problem I have is that some of the handling in various
>> states-events combo is almost same. That makes my code
>> redundant.[/color]
>
> Any part of the code that's truly redundant can be moved to a
> function that all the states can rely on.
>[color=green]
>> Also the function size is toooo large. Is there any way or
>> design to write this code. I heard about writing a seperate
>> function for each state. But not sure???[/color]
>
> You can define each state as a function, yes.
>
> What you need is a function type that returns a pointer
> to function of it's own type. This is not possible to
> declare directly in C++.
>
> Here's a finite state machine, which is implemented using
> functions, that recognizes the word "cat". A struct is used to
> simulate a recursive function type.
>
> #include <iostream>
>
> struct state_t;
>
> typedef state_t state(char c);
>
> struct state_t
> {
> state_t(state* p): process(p) { }
> state *function;
> };[/color]

Sorry that got garbled before I pasted.

struct state_t
{
state_t(state* p): function(p) { }
state *function;
};

--
Neil Cerutti