Caterpillar

I'm currently working code update implementation, with the goal of making code updates more reliable. I thought I had this pretty much figured out in an earlier prototype (with some adaptations required, because that prototype ran everything in a single process). Turns out, this isn't really the case. I must have had a bit of tunnel vision.

That old implementation is much more limited than I thought it was, and there's still lots to figure out to make it practical. So I'm back to the drawing board. I'm getting there. But working on this again, and having talked about the code database and a custom editor lately, doing this based on text files strikes me as deeply suboptimal.

To perform a successful code update, the compiler needs to figure out exactly what changed, so it can generate code to transition from the old version to the new. And the editor already had that information, but then threw it away. This would be much easier, if we didn't have to reconstruct those changes by comparing their end results.