Caterpillar

If code is represented as text files, and we have a code update, there's not much we can do except compare two big blobs of code (either as text, or in its pre-compiled form), to figure out what the specific changes are between them. This is not only unreliable, but also wasteful. Because the editor already had that information and threw it away.

If we had a code database and a custom editor that updated the database directly, we'd know about every single change as it happens. The compiler would compile just the new code, and generate code to transition from old version to new. There would be no perceptible delay between finishing to type something, and that making it into the live program.

Alas, the time has not come yet. Building the infrastructure required for this would be a huge distraction right now, and set any hope of productive use back by months, at best. So I'm left with having to find a technique that works well enough for now, based on source code in text files.