I think you might want to take a look at this:

http://ffnnkk.org/

Fact is that graphical programming languages are only good for certain kinds of programs which is why they have only gained traction in the data acquisition world ( labview, lego mindstorm..etc ) where the metaphor of "data-flow" fits in very well with the domain. Sometimes they are just really slow to create general purpose programs compared to traditional text-based programming. I still remember the labs I had to do in Labview where to insert an element into an array, I had to drag like 3 different blocks unto the screen, convert the data into right types the and then wire them up, just inefficient.

While I haven't used it, Processing (http://processing.org/) sounds like it may be worth a look.

It's the output of Processing that's visual. In graphical PLs, it's the input that's graphically manipulated; take a look at Scratch, you drag a "for loop" from a toolbox and drop it into a visual "control flow", then watch the FOR snap into place and presto.

http://scratch.mit.edu/

processing is a framework for creating graphics and visualization. i am sure this is a good way to write the graphical language container, but i was looking more for what graphical metaphors should exist if we wanted to solve problems using a graphical layout.

All right, for starters: Most algorithms can be decomposed into graph problems. (Trees are a subset of graphs.) Set operations, as you noted, can usually be visualized with venn diagrams. Multiple inheritance may make more sense as a tree growing together. A garbage collector's arena could show its data fading away when it hasn't been accessed for a while. I like the idea of a hash of a variable affecting its color, making structurally equal items easier to spot.

Have you ever seen lavaps (http://www.isi.edu/~johnh/SOFTWARE/LAVAPS/index.html), by the way?

To turn the question on its head, I wonder what concepts remain undiscovered because people tend to brainstorm visually (on whiteboards, etc.), but some ideas are intrinsically hard to depict visually. (Speaking as somebody who is more textually-minded, but has become good at spatial thinking because of 3D programming.)

Brainstorming in terms of sound (harmonic scales or rhythms) could bring out different solutions, for example. I find anthropomorphizing components of a problem (like the actor model) helpful, too.

i like the idea of any vr, or augmented reality components being brought into the programing language environment.

the way i see it, text based languages lend them selves well to linear thinking and linear processing. i think we are moving to a place where we need to move passed linear thinking and move more towards networked processes.

it seems to me that on large sites, you might want to do A/B testing on a number of features to see which ones work better, this means staged deployment, testing and programing are done in a really complex way. we should accept this as part of the pattern and design environments and tools to make this easier to debug, without a whole bunch of "if" statements cluttering the code to determine the configuration state.

i also see writing functions now as part of the design process, and i always like some of the metaphors in UML, but UML also always lacked into translating directly into code.