What citation do you need? At the very least, a visual environment could always trivially display each syntax/AST node, which will be the same as seeing a plain text version of the code (but better -- closer to something like Apple's Dylan IDE or Smalltalk environments).

Strip out the text and see how clear it is.

Who said you need to "strip out the text"?

Text has been part of graphical representations for milennia.

The cool part is that you get to decide WHAT text you get to see -- you're not just left with seeing the lower level code.

It could be that, but it can also be boxes representing objects or functions, program modules, interacting programs, at any level of abstraction one likes.

Stripping the text out is a way to see how much information the graphical element conveys.

By allowing arbitrary text in so-called graphical languages, you have turned your original statement into a vacuous truth - an empty tautology - as all text can be diagrammed.

Strip out the "text" in your code and just leave the syntax - we'll see how readable that is.

Well, that would certainly demonstrate the value of the text aspect of the code...

Exactly. A graphical representation and a syntactic representation of code will both have text in them. They just have different syntax for equivalent structure.

You seem to be conflating visual programming languages/environments with pictography.

They are not about that.

I am suggesting that your concept of a visual language is just an enhanced presentation of one that is fundamentally textual.

And I'm suggesting the keyword is "enhanced".

Movies are fundamentally photography too.

The fact that you have to stretch for that analogy, rather than name a visual language that offers more than just an incremental improvement to the process of reading text, suggests that the alleged power of visual programming is mostly hypothetical at this point.

Confirmation bias much?

I see that you have given up on any attempt at a factual argument.

About as minimal a graphical representation as you can get would be to have a tree of application nodes with S and K combinators as leaf nodes.

You can compile any lambda calculus expression down to S and K and "run" them as a graph.

Doing this on real calculation probably wins a prize for non clarity though...