component based - Would it be better to use XML/JSON/Text or a database to store game content?

I'm considering how to implement a component-based game, as that seems to be the hot thing and I like the idea of such a flexible design. One of the features of such a design is that adding new things to the game can be done through data, often presented as loading content through text files such as XML. This has the advantage of being human readable and easily editable in any text editor. On the downside, text can be slower to deal with, and you have to manage a large collection of data files. Similar text-based formats like JSON or config files would have similar benefits.

On the other side, there are small, portable databases like SQLite or Tokyo Cabinet. While not directly human readable, these files are easy to interface with, and I'd imagine some kind of editing tool would be preferable for game content design anyway. Using a DB allows for a consistent storage of config information and easy retrieval. You could serialize data into a DB for save games as well.

Performance wise, I think generally XML is faster for small files but a database scales better to large amounts of data. I'd imagine any real game is going to have a whole lot of game objects.

So the question: Which is the preferable approach? I'm leaning towards the DB, but I want to know if there are hidden pitfalls or real strong advantages to text files. Or if there are other alternatives besides these (serialize to binary format I guess?)

Answer

It might not come up so much for a small personal game, but one hard problem when it comes to game data is multi-user editing/versioning. We use a lot of small text files that get baked down to a small number of binary blobs by a build process. This makes life easier for designers since they have a lot of flexibility in their workflow. CCP, as a counter example, uses a central editing database that all designers connect to. This makes the build step unneeded (or at least a lot simpler) but it means you need to implement all the versioning and workflow features yourself, so they are bound to be simpler than other tools out there. You can deal with performance in either case, so the real question is what do you want for a designer workflow and how can you get there?