Postmortem: Zombie's SpecOps: Rangers Lead the Way

Someone once said, "Experts are just people who have already made all the mistakes in their field." If this old saying is remotely true, I must be well on my way. After two years battling on the front lines at Zombie, I look back on the creation of SpecOps as a crusader looks back on victory in the Holy Land. Through the haze of past battles won and lost, I will now try to remember what it was that we did right and wrong, and how this product finally hit the shelves.

To begin, let’s define the responsibilities of our game engine. Our game engine manages many aspects of the game: sound, graphics, physics, game core, and others. Taken alone, these components don't do anything — they’re just tools. The game-specific AI uses these tools to perform logical game actions. In this manner, a complete game is composed of the game engine plus the game-specific AI.

As such, the SpecOps game is composed of the Viper game engine and the SpecOps AI and resources. In this article, I will primarily focus on the design aspects of the Viper engine, which was the first and most important step in creating SpecOps. In Table 1, you’ll find a description of the project and its development environment. With these design requirements in mind, let’s take a look at how each of the major components of the game engine was implemented.

Game Core

The game core provides an interface for all of the game engine components. Among other things, it defines and manages "objects" in the engine and allows them to be passed between components. For example, the game core might first send an object to the physics engine to be moved and then to the graphics engine to be drawn.

In the Viper engine, we chose to implement the objects hierarchically. Figure 1 shows a small portion of this hierarchy. The hierarchy allows memory optimizations because objects only require allocation of the structures that they use. It also provides some object-oriented structure to the program. In C, there was one .C file for each object type, with the same .H relationship as in Figure 1.

Figure 1. Hierarchical object relationship from SpecOps.

The only objects that the game core referenced and modified were those at the top of the hierarchy: Generic Object, Static Object, and Dynamic Object. The rest of the hierarchy is composed of game-specific data structures (such as the Character class). The support modules for these were separated from the rest of the code, so that it would be easy to remove them from the game engine code. This helped ensure the separation of game-specific logic from the game engine.

This design was expandable and easy to work with. While building the engine, we could separate out the SpecOps-specific code to create demos and even other games. Doing so at several points during the development cycle forced us to continually clean up any breaches of the design and maintain the reusability of the engine.

Game Editor

0ur game editor let us introduce new resources into the game engine, modify game play, debug game logic, and print out diagnostics. An editor environment needs to be easy to learn and use, so that the game designers and artists can make game modifications without a programmers being involved. It also needs to be continually maintained, providing access to new features of the engine as they are created and having bugs fixed or the interface modified to save time.

I wanted the Viper editor to be totally integrated into the game engine. This would help ensure that new game engine code didn’t break the editor code, and would allow us to add new game engine features to the editor environment. I created an EDITOR compiler switch that created an encapsulating .DLL into the executable game file upon compilation. This .DLL ran the game as a child thread, allowing us to suspend the execution of the game and modify the contents of the game’s memory. With this switch, we could activate functions in the game engine that we didn’t want in the release version.

When active, the Viper editor displays a simple console . Command strings can be entered into this console to perform a variety of actions in the game engine. For instance, we would frequently use this functionality to load terrain geometry, create a binary space partition (BSP) tree of that geometry, and save the BSP tree to disk in the native file format.

The Viper editor console.

Overall, I was pretty happy with this editor. While it’s more cryptic than the user interfaces of the Quake and Unreal editors, it has several benefits that those systems don’t offer. First, all of our programmers were able to add functionality to it. Once our team learned how to use the editor’s string commands (which took about five minutes), any programmer could add functions to support the code they wrote. Had I written a MFC-based GUI editor, I would have been supporting the editor on full-time basis because few team members were familiar with Windows code. Another point in favor of our editor was the fact that our geometry was being created separately by CAD tools, so we didn’t need geometry creation facilities such as those found in the Quake and Unreal editors. Simply put, our needs were different from those games, and the command string interface fit our needs fairly well.

Yet, the editor did have drawbacks. Some tasks were difficult to complete using the command string interface. For example, to place dynamic objects (such as enemies and pickups) in the environment effectively, the game designers really needed a CAD-like, windowed interface that showed the entire level at various angles and allowed designers to place objects with mouse clicks. This could have been added fairly easily, but we lacked the time to implement it. As we advance the Viper engine, I’d like to see the editor seamlessly built into an existing CAD package, such as 3D Studio Max. That way, we could take advantage of an interface that the artists are already familiar with and draw upon pre-existing features.