You can do all that at the text level, if it makes the tooling easier; the tooling would just be operating on a canonically-chosen textual projective representa

You can do all that at the text level, if it makes the tooling easier; the tooling would just be operating on a canonically-chosen textual projective representation of graphical code (think: `go fmt`, but as a FUSE mount!), rather than against canonically textual code. Changes on the text layer would be propagated back down as changes to the persisted binary AST that the text represents. As such, any Unix tool used to manipulate text (and thereby textual source code), e.g. diff(1) + patch(1), could be used to manipulate the persisted binary AST through the text. `sed -i` would still work, even. But you could also have other tools, that manipulate the underlying binary AST directly; where the textual projection would be the thing to update in response.