Crosscut

If input code and rendered code don't need to be the same, then you can optimize each for its purpose. For example, one might decide that significant whitespace is best for readability, but still might not want it in a language, because it could be harder to write. With a code database and a custom editor, we wouldn't have to make that trade-off.

Rendered code could use whitespace to show how code is structured. Or it could use other means to highlight that structure, like drawing boxes. We could go totally crazy and render code as a control flow graph. At that point, the border between traditional (text-based) and visual programming is pretty fuzzy.

And likewise, we could optimize the input code for its respective purpose. I'm not claiming that the example I presented yesterday is the right way to do it. But generally speaking, the characters that we actually type, would not need to unambiguously define the code. Only the whole input stream, consisting of characters, other keys, touch gestures, or whatever else, would need to.