One of recent experiments in game development has been to orient my workflow away from code and towards data. When I use Flash, I’ve started using Ogmo Editor to design my levels, XML to handle settings, and Slang to handle game logic whenever possible. As a result, a lot of the code I’ve written recently can be easily reused across projects, and my classes are systemic and steer clear of situation-specific behavior. It’s been quite fulfilling and has done wonders for my iteration time; the focus on data instead of code means that I can take advantage of live-reloading for nearly every aspect of development. It’s not uncommon for me to work for an hour without ever closing my game, as I can simply reload all assets and data with the press of a button.

Today I’ve been working on Humphrey’s Tiny Adventure: Remastered, a post-compo version of my first 48 hour game. There are a number of places in the game where I need to set up events on a timeline, which I’ve done a few times already (see here and here), and each time I used some variation of traversal over a queue of function closures. While this approach worked well enough, it was tedious to set up and a pain to debug, not to mention the fact that it was anything but systemic. The most criticized aspect of the Ludum Dare version of Humphrey was that the intro cutscene was too long, and I agree. I think I knew that even before I released the game, but there was no way I was diving into that code to change it.

For Humphrey: Remastered, I was determined to achieve the same type of event scheduling in a data-driven way. Thanks to Slang, I was able to do just that. Here are the scripts for a scene I’ve been testing that involve two actors; Humphrey and Abe. This won’t be in the finished game, but it’s a good demonstration of the system’s capabilities.

action {
        print "start"
        delay 3
        done
}

action {
        print "after delay"
        give-cue "humphrey done"
        done
}

action {
	print "start abe"
	done
}

action {
	await-cue "humphrey done" {
		print "got cue"
		done
	}
}

Actions are called sequentially by the actor that the script is attached to. Each action block defines a set of statements that will be run as the events are executed. Calling done moves to the next action, and delay causes the actor to stop executing actions for the given number of seconds.

The give-cue and await-cue functions allow actors to pass messages between each other. In the above example, Humphrey give a cue to tell that he’s finished a set of actions, and Abe, who has been waiting for that cue, executes his final action in response.

There are only a few control functions involved, but so far the system has proven to be very powerful, and more than capable enough for the needs of this project. With the addition of message passing and response, specialized actor classes will be able to define custom behaviors while allowing the system to remain pure.

I’m quite happy with the way this has turned out. It’s fun to experiment with what can be done with Slang even in its current, quite minimal state.