Hey guys,

So I'm using a finite state machine for AI in my game. Up until recently it has been working perfectly - however, yesterday, for no discernable reason, this piece of code says that the currentAIstate's vfptr is inaccessible right when the instance of c_Zombie spawns:

void c_Zombie::logic( Uint32 deltaTicks )
{
    currentAIstate->AIlogic(collisionBox, c_Entity_manager::getPlayer()->getBox(), xVel, yVel, up, down,  left, right, alerted);

/*some other code..*/
    if(condition){
        setnextAIstate(someState);
    }

    changeAIState(currentAIstate);
}

the code for switching AI states goes like this:

void changeAIState(c_AIstate* currentAIstate)
{
	if ( nextAIState != AI_NULL )
	{
		delete currentAIstate;

		switch (nextAIState)
		{
		case CHILL:
			currentAIstate = new c_Chillmode(); break;
		case GO_TO: 
			currentAIstate = new c_Gotomode(); break;
		case ATTACK:
			currentAIstate = new c_Attackmode(); break;
		case DEAD:
			currentAIstate = new c_Deadmode(); break;
		}
		AIstateID = nextAIState;
		nextAIState = AI_NULL;
	}
}

void setnextAIState( int state )
{
	if ( state != DEAD )
	{
		nextAIState = state;
	}
	else
		nextAIState = DEAD;
}

I can't figure out why it just decided to stop working recently;

also, I read that allocating memory during runtime is a bad idea - what other options do I have for changing AI states, assuming that the method I'm using right now is a bad idea?

here's a link to the github if you need more info: https://github.com/johnmarinelli/8bit-sh/tree/master/cpp

or, just leave a comment with whatever info you need; I've already had to delete one feature of the game because of this, and I don't want to have to do it again.

Thanks for your time!

0xfeeefeee is the debug fill pattern for freed memory. If you get an access violation involving that address it means you're probably accessing an object after it's been deleted or something similar.

I know; I'm wondering why it's doing that since in my changeAIstate function I assign the pointer to a new AIstate right away.

After some toying around, I've found out that the problem is a bit more 'tangled' than I thought, so I don't expect any solutions - however, my other question still stands:

If using heap memory at runtime is generally a bad idea, how are state machines implemented in games? I have three of them in my game (one for the gamestates, one for AI, and one for a certain little puzzle), and am wary about using them because of the reason above.

For AI you don't necessarily need to create and delete a new state each time, as the others have said. I actually have written a few game state managers that create new states and swap them, as well as unload the old ones on change. Generally the way I handled that was by setting a flag and a temporary variable that holds a reference to the newly created gamestate, then have the gamestate physically swap when code isn't actually running through it in order to avoid access violations and such using the parent GameStateManager class.

Of course there are a lot of ways to design things like this.