By various sorts of happenstance, I’ve had quite a bit of exposure to data/file-type translation in projects I’ve been involved in. Given that, I don’t know why it still surprises me how difficult it can be. I think what gets me is that the broad strokes always seems pretty simple (“both formats are basically just a bag of polygons, right?”), and you can even get promising results based on that naive view. Fueled by that initial success, you start charging into the deeper details, and uh-oh, suddenly there’s some ‘little’ detail that threatens to swallow the whole project.

Another factor is that you don’t want the users of your translation facility to have to think about whether some given feature carries over perfectly into the translated realm, so you go nuts trying to support every little corner case, probably many that no user will ever encounter.